Computer Vision Based Human Detection

Md Ashikur. Rahman

To cite this version:

Md Ashikur. Rahman. Computer Vision Based Human Detection. International Journal of Engineer-
ing and Information Systems (IJEAIS), 2017, 1 (5), pp.62 - 85. �hal-01571292�

HAL Id: hal-01571292

https://hal.archives-ouvertes.fr/hal-01571292
Submitted on 2 Aug 2017

HAL is a multi-disciplinary open access L’archive ouverte pluridisciplinaire HAL, est

archive for the deposit and dissemination of sci- destinée au dépôt et à la diffusion de documents
entific research documents, whether they are pub- scientifiques de niveau recherche, publiés ou non,
lished or not. The documents may come from émanant des établissements d’enseignement et de
teaching and research institutions in France or recherche français ou étrangers, des laboratoires
abroad, or from public or private research centers. publics ou privés.
International Journal of Engineering and Information Systems (IJEAIS)
ISSN: 2000-000X
Vol. 1 Issue 5, July– 2017, Pages: 62-85

Computer Vision Based Human Detection

Md. Ashikur Rahman
Dept. of Computer Science and Engineering
Shaikh Burhanuddin Post Graduate College
Under
National University, Dhaka, Bangladesh
[email protected]

Abstract: From still images human detection is challenging and important task for computer vision-based researchers. By
detecting Human intelligence vehicles can control itself or can inform the driver using some alarming techniques. Human
detection is one of the most important parts in image processing. A computer system is trained by various images and after making
comparison with the input image and the database previously stored a machine can identify the human to be tested. This paper
describes an approach to detect different shape of human using image processing. This thesis mainly based on shape based
detection. Shape of the input image is extracted using a operator namely cany operator. Different images are used to train up the
system. Then after training the system with the input image when a test image is provided to detect, test image is then compared
with the database. If a certain threshold value is found then the test image is considered as the specific human. The average
accuracy and precision rate achieved by the system is above 93%.

Keyword: Computer Vision, Human Detection, Edge detection

1. INTRODUCTION
Analysis of visual scenes involving humans is one of the very popular, yet demanding applications of Computer Vision. Some of
the tasks that fall under this domain are face recognition, gesture recognition and tracking the whole body. The motivation stems
from the desire to improve human computer interaction which has been one of the general goals of artificial intelligence. Detecting
humans in still images is a relatively new field. This domain is rich and challenging because of the need to segment rapidly
changing scenes in natural environments. Additional momentum has been provided by the technological advances in the real time
capture, transfer and processing of images. Today, the basic capability of a smart surveillance system would be to detect if
humans are indeed present in the captured frames/images. This paper deals with detecting humans with fairly upright position in
images. Human detection in images is a challenging task owing to variable appearance and wide range of poses that they can adopt.
Hence, a robust feature set is needed that allows the human form to be discriminated clearly even in cluttered backgrounds and
difficult illumination. Below we present some applications which will justify the need of a robust human detector using Canny
Operator. In this thesis, an approach has been used to detect different human. To detect human, firstly the system is trained with
various database. Shape based detection approach is used in this thesis.
1.2 What is Computer vision?
Computer vision techniques are used in such systems and human detection. Human detection and identification is an essential
task for many applications such as Human-Robot-Interaction (HRI), video surveillance, human motion tracking, gesture
recognition and human behavior analysis. Among many applications, we are interested in the field of HRI. As because intelligent
robots should coexist with humans in a human-friendly environment, they must be aware of humans in their proximity, and identify
them. Often, a single static camera is used for human detection due to its low cost and easy handling. However, a single camera
is not practical for human detection by a mobile robot because the robot (camera) and the human are moving relatively each other,
and the illumination conditions and backgrounds are changeable. Hence, the depth cues from a stereo camera can help to detect and
identify humans effectively in mobile robot applications. Thus, stereo-based vision is used to detect and identify the humans in this
paper.
1.3 Background of the Research
Human detection in real world scenes is a challenging problem. Intelligent vehicles refer to cars, trucks, buses etc. on which
sensors and control systems have been proposed and assisted driving task. In recent years a variety of approaches have been
proposed and impressive results have been reported on a variety of data bases.
1.4 Objective of the project

www.ijeais.org
62
International Journal of Engineering and Information Systems (IJEAIS)
ISSN: 2000-000X
Vol. 1 Issue 5, July– 2017, Pages: 62-85
An intelligent machine must have the capability to detect various types of human. This thesis will be able to detect different types
of object after training with various databases. The main human of this thesis is to detect specific object from still images. The major
parts are:
1. Gray scale conversion.
2. Shape detection
3. Image comparisons
4. Human detection
1.5 Scope of the project
This paper presents a novel system for the real time detection and tracking from a moving vehicle. The scope of the work is
versatile. Detecting human beings in images has gained prominence in the field of Computer Vision, with applications in the fields.
1.6 Proposed System’s overview
This thesis presents a computer vision based human detection system using boundary based approach. By using camera this
project firstly capture the real time images. After detecting moving object by segmentation method and eliminating noises this
project detects the object. By comparing the boundary with predefined templates it can detect human. This approach used in this
system has advantages over other human detection systems in its speed, simplicity, learning, capability and robustness to small
changes in the images. This system can not only detect humans but also other object we want. This thesis is very efficient to train
with different moving objects.
1.7 Project Organization
This thesis is organized into five chapters. This second chapter presents related works done previously with their success and
limitation. Chapter three explains the details of the detection methodology used by us to tackle the problem of human detection,
including experimental setup, feature vector generation algorithm, datasets used, evaluation methodology used and configuration
settings for the feature descriptors tested. Chapter four explains the steps adopted towards choosing the best feature descriptor for
compressed images. Finally in the last chapter explains the results obtained and concludes the paper by discussing its key
contributions, limitations as well as future works.
1.8 Conclusion
In this paper, we use a combination of feature extraction and learning frame work to classify whether an image contains/does not
contain human(s). We also propose a feature descriptor that is resistant to compression.

2. RELATED WORKS
2.1 Introduction

Human detection is closely related to general object recognition techniques. It involves two steps
- feature extraction and training a classifier as shown in Figure 2.

Figure 2.1 Components of Human Detection System

The image feature set that needs to be extracted should be the most relevant ones for object detection or classification, while
providing invariance to changes in illumination, changes in viewpoint and shifts in object contours. Such features can be based on
points [1] and [2], blobs (Laplacian of Gaussian [3] or Difference of Gaussian [4]), intensities [5], gradients [6] and [7], colour,
texture, or combinations of several or all of these [8]. The final descriptors need to characterize the image sufficiently well for the
detection and classification task at hand. We will divide the various approaches to descriptor selection into two broad categories:
Sparse representations are based on local descriptors of relevant local image regions. The regions can be selected using either

www.ijeais.org
63
International Journal of Engineering and Information Systems (IJEAIS)
ISSN: 2000-000X
Vol. 1 Issue 5, July– 2017, Pages: 62-85
key point detectors, image fragments or parts detectors. On the other hand, dense representations are based on image
intensities, gradients or higher order differential operators. Image features are often extracted densely (often pixel-wise) over
an entire image or detection window and collected into a high-dimensional descriptor vector that can be used for discriminative
image classification or labeling the window as object or non-object.
2.2 Local Shape-Based Human Detection
Mori et al . [ Mori, 2002] model human body configurations where body part templates are represented by local Shape Context.
In the later work, they apply normalized cuts segmentation and use shape, shading and focus cues for retrieving the body parts. M.
Oren et al. [Oren, 1997] use Haar wavelet coefficients to build o global human model. Edgar Seeman, Bastian Leibe et al
[Leibe,2003] studied different shape based human detection algorithms. There are mainly two kinds of shape based detection
techniques: Global approach and Local approach.
2.2.1 Global Approach
Global approach is known as Global chamber matching technique [Gavrila,2000] and Local approach is known as Implicit
Shape Model (ISM) [Harris, 1998].Object shape silhouettes to image structure are match on the Global chamfer matching
approach. For that purpose, a silhouette is shifted over the image and a distance Dchamfer (T,L) between a silhouette T and the
Edge image at each image location L is calculated.

2.2.2 Local Approach- Implicit Shape Model (ISM)

The local approach is subdivided into three subsections. Such as: Model training, Hypothesis generation, segmentation and
verification. The ISM is trained by extracting local features from training images [Leibe, 2000].Then modeling their spatial
occurrence distributed on the object. An Interest Point Detector is applied for each training images. After training with the images
Hypothesis is generated and finally segmentation and verification is performed to detect the human.
By using the Interest Poit Detector local features are calculated. There are Interest Point Detector techniques such as Harris
detector [Harris ,1988], Deference of Gaussian detector [Lowe, 2004] Harris-Laplace detector [ Mikolajczyk, 2001], Hessian-
Laplace [ Mikolajczyk, 2004]
2.3 Dense Descriptors of Image Regions
One of the primary works using simple image intensities is the Eigen faces approach of [9]. Approaches using image gradient
descriptors are [7], where histograms of gradients have been used. The Census algorithm [10] transforms the intensity space to an
order space, where a bit pattern is formed by looking at the orders of a given pixel with its neighbors. [11] uses an improved version
of this algorithm where they somewhat alleviate the problem of counting salt- and-pepper noise in a pixel multiple times. [12]
proposed a method in which the penalty for an order flip is proportional to the intensity difference between the two flipped pixels,
thereby improving the noise immunity. Finally, [13] presents a statistical approach whose match measure can be tuned to the
underlying error process. All of these methods assume that the pixel locations do not vary across the two patches and are thus
inappropriate for a feature matching problem where the pixel locations might undergo some shift. [14] makes feature descriptors
with point pairs that are invariant to Gaussian noise. A penalty is awarded if there is an order change for a point pair between the two
patches and such penalties for different pairs are summed in order to determine the difference between the two features. The Local
Binary Patterns (LBP) descriptor [15], which is a variant of the Census approach [10] has also shown promise in texture
description [16]. As the LBP operator produces a rather high dimensional histogram and is therefore difficult to use in the context of
a region descriptor, a Center-Symmetric LBP (CS- LBP) which only compares center-symmetric pairs of pixels was considered for
feature description in [17]. More recently, [18] came up with the idea of Center Symmetric-Local Ternary Patterns (CS-LTP) to
make CS-LBP resistant to noise together with proposing a global order based descriptor (Histogram of Relative Intensities (HRI))
which handles saturation and illumination changes better.
2.4 Work in human Detection
[19] describes a pedestrian detector based on a polynomial SVM using rectified Haar wavelets as input descriptors, with a parts
(subwindow) -based variant in [20]. However we find that linear SVMs (weighted sums of rectified wavelet outputs) give similar
results and are much faster to calculate. [19] shows results for pedestrian, face, and car. [21] takes a more direct approach, extracting
edge images and matching them to a set of learned exemplars using chamfer distance. This has been used in a practical real-time
pedestrian detection system [22]. [23] built an efficient moving person detector, using AdaBoost [24] to train a chain of
progressively more complex region rejection rules based on Haar-like wavelets and space-time differences. [6] built an articulated

www.ijeais.org
64
International Journal of Engineering and Information Systems (IJEAIS)
ISSN: 2000-000X
Vol. 1 Issue 5, July– 2017, Pages: 62-85
body detector by incorporating SVM based limb classifiers over 1st and 2nd order Gaussian filters in a dynamic programming
framework similar to those of [25] and [26]. [27] uses combinations of orientation position histograms with binary-threshold
gradient magnitudes to build a parts based method containing detectors for faces, heads, and front and side profiles of upper and
lower body parts. [28] uses a combination of [7] and [15] to build a more efficient descriptor and also try to overcome occlusion.
2.5 Different Types Edge Detector
2.5.1 Introduction
Edge detection refers to the process of identifying and locating sharp discontinuities in an image. The discontinuities are abrupt
changes in pixel intensity which characterize boundaries of objects in a scene. Classical methods of edge detection involve
convolving the image with an operator (a 2-D filter), which is constructed to be sensitive to large gradients in the image while
returning values of zero in uniform regions. There are an extremely large number of edge detection operators available, each
designed to be sensitive to certain types of edges. Variables involved in the selection of an edge detection operator include Edge
orientation, Noise environment and Edge structure. The geometry of the operator determines a characteristic direction in which it is
most sensitive to edges. Operators can be optimized to look for horizontal, vertical, or diagonal edges. Edge detection is difficult in
noisy images, since both the noise and the edges contain high frequency content. Attempts to reduce the noise result in blurred
and distorted edges. Operators used on noisy images are typically larger in scope, so they can average enough data to discount
localized noisy pixels. This results in less accurate localization of the detected edges. Not all edges involve a step change in intensity.
Effects such as refraction or poor focus can result in objects with boundaries defined by a gradual change in intensity [1].
The operator needs to be chosen to be responsive to such a gradual change in those cases. So, there are problems of false edge
detection, missing true edges, edge localization, high computational time and problems due to noise etc. Therefore, the objective is
to do the comparison of various edge detection techniques and analyze the performance of the various techniques in different
conditions. There are many ways to perform edge detection. However, the majority of different methods may be grouped into two
categories:
Gradient based Edge Detection:
The gradient method detects the edges by looking for the maximum and minimum in the first derivative of the image.
Laplacian based Edge Detection:
The Laplacian method searches for zero crossings in the second derivative of the image to find edges. An edge has the one-
dimensional shape of a ramp and calculating the derivative of the image can highlight its location.
Suppose we have the following signal, with an edge shown by the jump in intensity below: Suppose we have the following
signal, with an edge shown by the jump in intensity below:

If we take the gradient of this signal (which, in one dimension, is just the first derivative with respect to t) we get the following:

www.ijeais.org
65
International Journal of Engineering and Information Systems (IJEAIS)
ISSN: 2000-000X
Vol. 1 Issue 5, July– 2017, Pages: 62-85

Clearly, the derivative shows a maximum located at the center of the edge in the original signal. This method of locating an edge
is characteristic of the “gradient filter” family of edge detection filters and includes the Sobel method. A pixel location is declared an
edge location if the value of the gradient exceeds some threshold. As mentioned before, edges will have higher pixel intensity values
than those surrounding it. So once a threshold is set, you can compare the gradient value to the threshold value and detect an edge
whenever the threshold is exceeded. Furthermore, when the first derivative is at a maximum, the second derivative is zero. As a
result, another alternative to finding the location of an edge is to locate the zeros in the second derivative. This method is known as
the Laplacian and the second derivative of the signal is shown below:

2.5.2 Edge Detection Techniques

2.5.2.1 Sobel Operator
The operator consists of a pair of 3×3 convolution kernels as shown in Figure 1. One kernel is simply the other rotated by 90°.

Figure 2.2: Masks used by Sobel Operator

These kernels are designed to respond maximally to edges running vertically and horizontally relative to the pixel grid, one
kernel for each of the two perpendicular orientations. The kernels can be applied separately to the input image, to produce separate
measurements of the gradient component in each orientation (call these Gx and Gy). These can then be combined together to find the
absolute magnitude of the gradient at each point and the orientation of that gradient. The gradient magnitude is given by:
Typically, an approximate magnitude is computed using:

which is much faster to compute. The angle of orientation of the edge (relative to the pixel grid) giving rise to the spatial gradient
is given by:

2.5.2.2 Robert’s cross operator:

The Roberts Cross operator performs a simple, quick to compute, 2-D spatial gradient measurement on an image. Pixel values at
each point in the output represent the estimated absolute magnitude of the spatial gradient of the input image at that point. The

www.ijeais.org
66
International Journal of Engineering and Information Systems (IJEAIS)
ISSN: 2000-000X
Vol. 1 Issue 5, July– 2017, Pages: 62-85
operator consists of a pair of 2×2 convolution kernels as shown in Figure 2. One kernel is simply the other rotated by 90°[4]. This is
very similar to the Sobel operator.

Figure 2.3: Masks used for Robert operator

These kernels are designed to respond maximally to edges running at 45° to the pixel grid, one kernel for each of the two
perpendicular orientations. The kernels can be applied separately to the input image, to produce separate measurements of the
gradient component in each orientation (call these Gx and Gy). These can then be combined together to find the absolute magnitude
of the gradient at each point and the orientation of that gradient. The gradient magnitude is given by:

although typically, an approximate magnitude is computed using:

which is much faster to compute.

The angle of orientation of the edge giving rise to the spatial gradient (relative to the pixel grid orientation) is given by:

2.5.2.3 Prewitt’s operator:

Prewitt operator is similar to the Sobel operator and is used for detecting vertical and horizontal edges in images.

Figure 2.4: Masks for the Prewitt gradient edge detector

2.5.2.4 Laplacian of Gaussian:

The Laplacian is a 2-D isotropic measure of the 2nd spatial derivative of an image. The Laplacian of an image highlights
regions of rapid intensity change and is therefore often used for edge detection. The Laplacian is often applied to an image that has
first been smoothed with something approximating a Gaussian Smoothing filter in order to reduce its sensitivity to noise. The
operator normally takes a single gray level image as input and produces another gray level image as output.

www.ijeais.org
67
International Journal of Engineering and Information Systems (IJEAIS)
ISSN: 2000-000X
Vol. 1 Issue 5, July– 2017, Pages: 62-85

The Laplacian L(x,y) of an image with pixel intensity values I(x,y) is given by:

Since the input image is represented as a set of discrete pixels, we have to find a discrete convolution kernelthat can approximate
the second derivatives in the definition of the Laplacian. Three commonly used small kernels are shown in Figure 4.

Figure2.5: Three commonly used discrete approximations to the Laplacian filter.

Because these kernels are approximating a second derivative measurement on the image, they are very sensitive to noise. To
counter this, the image is often Gaussian Smoothed before applying the Laplacian filter. This pre-processing step reduces the high
frequency noise components prior to the differentiation step.
In fact, since the convolution operation is associative, we can convolve the Gaussian smoothing filter with the Laplacian filter
first of all, and then convolve this hybrid filter with the image to achieve the required result. Doing things this way has two
advantages: Since both the Gaussian and the Laplacian kernels are usually much smaller than the image, this method usually requires
far fewer arithmetic operations.
The LoG (`Laplacian of Gaussian')[6] kernel can be pre-calculated in advance so only one convolution needs to be performed at
run-time on the image.
The 2-D LoG function centered on zero and with Gaussian standard deviation has the form:

Figure 2.6: The 2-D Laplacian of Gaussian (LoG) function. The x and y axes are marked in standard deviations

www.ijeais.org
68
International Journal of Engineering and Information Systems (IJEAIS)
ISSN: 2000-000X
Vol. 1 Issue 5, July– 2017, Pages: 62-85

Figure2.7: Discrete approximation to LoG function with Gaussian S = 1.4

Note that as the Gaussian is made increasingly narrow, the LoG kernel becomes the same as the simple Laplacian kernels shown
in figure 4. This is because smoothing with a very narrow Gaussian (< 0.5 pixels) on a discrete grid has no effect. Hence on a
discrete grid, the simple Laplacian can be seen as a limiting case of the LoG for narrow Gaussians.

2.5.2.5 Canny Edge Detection Algorithm

The Canny edge detection algorithm is known to many as the optimal edge detector. Canny's intentions were to enhance the
many edge detectors already out at the time he started his work. He was very successful in achieving his goal and his ideas and
methods can be found in his paper, "A Computational Approach to Edge Detection". In his paper, he followed a list of criteria to
improve current methods of edge detection. The first and most obvious is low error rate. It is important that edges occurring in
images should not be missed and that there be no responses to non-edges. The second criterion is that the edge points be well
localized. In other words, the distance between the edge pixels as found by the detector and the actual edge is to be at a minimum. A
third criterion is to have only one response to a single edge. This was implemented because the first two were not substantial enough
to completely eliminate the possibility of multiple responses to an edge. Based on these criteria, the canny edge detector first
smoothes the image to eliminate and noise. It then finds the image gradient to highlight regions with high spatial derivatives. The
algorithm then tracks along these regions and suppresses any pixel that is not at the maximum (nonmaximum suppression). The
gradient array is now further reduced by hysteresis. Hysteresis is used to track along the remaining pixels that have not been
suppressed. Hysteresis uses two thresholds and if the magnitude is below the first threshold, it is set to zero (made a non-edge). If the
magnitude is above the high threshold, it is made an edge. And if the magnitude is between the 2 thresholds, then it is set to zero
unless there is a path from this pixel to a pixel with a gradient above T2.
Step 1:-
In order to implement the canny edge detector algorithm, a series of steps must be followed. The first step is to filter out any
noise in the original image before trying to locate and detect any edges. And because the Gaussian filter can be computed using
a simple mask, it is used exclusively in the Canny algorithm. Once a suitable mask has been calculated, the Gaussian smoothing
can be performed using standard convolution methods. A convolution mask is usually much smaller than the actual image. As a
result, the mask is slid over the image, manipulating a square of pixels at a time. The larger the width of the Gaussian mask, the
lower is the detector's sensitivity to noise. The localization error in the detected edges also increases slightly as the Gaussian width is
increased.
Step 2:-
After smoothing the image and eliminating the noise, the next step is to find the edge strength by taking the gradient of the
image. The Sobel operator performs a 2-D spatial gradient measurement on an image. Then, the approximate absolute gradient
magnitude (edge strength) at each point can be found. The Sobel operator uses a pair of 3x3 convolution masks, one
estimating the gradient in the x-direction (columns) and the other estimating the gradient in the y-direction (rows). They are shown
below:

www.ijeais.org
69
International Journal of Engineering and Information Systems (IJEAIS)
ISSN: 2000-000X
Vol. 1 Issue 5, July– 2017, Pages: 62-85

The magnitude, or edge strength, of the gradient is then approximated using the formula:
|G| = |Gx| + |Gy|

Step 3:-
The direction of the edge is computed using the gradient in the x and y directions. However, an error will be generated when
sumX is equal to zero. So in the code there has to be a restriction set whenever this takes place. Whenever the gradient in the x
direction is equal to zero, the edge direction has to be equal to 90 degrees or 0 degrees, depending on what the value of the gradient
in the y-direction is equal to. If GY has a value of zero, the edge direction will equal 0 degrees. Otherwise the edge direction will
equal 90 degrees. The formula for finding the edge direction is just:

Theta = invtan (Gy / Gx)

Step 4:-
Once the edge direction is known, the next step is to relate the edge direction to a direction that can be traced in an image. So if
the pixels of a 5x5 image are aligned as follows:

Then, it can be seen by looking at pixel "a", there are only four possible directions when describing the surrounding pixels - 0
degrees (in the horizontal direction), 45 degrees (along the positive diagonal), 90 degrees (in the vertical direction), or 135 degrees
(along the negative diagonal). So now the edge orientation has to be resolved into one of these four directions depending on which
direction it is closest to (e.g. if the orientation angle is found to be 3 degrees, make it zero degrees). Think of this as taking a
semicircle and dividing it into 5 regions.
Therefore, any edge direction falling within the yellow range (0 to 22.5 & 157.5 to 180 degrees) is set to 0 degrees. Any edge
direction falling in the green range (22.5 to 67.5 degrees) is set to 45 degrees. Any edge direction falling in the blue range (67.5 to
112.5 degrees) is set to 90 degrees. And finally, any edge direction falling within the red range (112.5 to 157.5 degrees) is set to 135
degrees.
Step 5:-
After the edge directions are known, non-maximum suppression now has to be applied. Non- maximum suppression is used to
trace along the edge in the edge direction and suppress any pixel value (sets it equal to 0) that is not considered to be an edge. This
will give a thin line in the output image.

Step 6:-

www.ijeais.org
70
International Journal of Engineering and Information Systems (IJEAIS)
ISSN: 2000-000X
Vol. 1 Issue 5, July– 2017, Pages: 62-85
Finally, hysteresis [12] is used as a means of eliminating streaking. Streaking is the breaking up of an edge contour caused by the
operator output fluctuating above and below the threshold. If a single threshold, T1 is applied to an image, and an edge has an
average strength equal to T1, then due to noise, there will be instances where the edge dips below the threshold. Equally it will also
extend above the threshold making an edge look like a dashed line. To avoid this, hysteresis uses 2 thresholds, a high and a low. Any
pixel in the image that has a value greater than T1 is presumed to be an edge pixel, and is marked as such immediately. Then, any
pixels that are connected to this edge pixel and that have a value greater than T2 are also selected as edge pixels. If you think of
following an edge, you need a gradient of T2 to start but you don't stop till you hit a gradient below T1.
2.6 Detection and Tracking Using Combination of Thermal and
Visible Imaging
The previous method was useful in indoors but in outdoor environment it require a high resolution camera. In this case a
fusion of Infrared Camera and Visible imaging is used. This significantly reduces the processing time and power required for
detecting human. The operational diagram for such kind of processing system is shown below.
The main process of the technique are Segmentation and Classification. These are used in many detection techniques.
2.6.1 Segmentation
The most important part of any real time human detection technique based on images is firstly detecting the still objects from the
background and ROI(Region of Interests).This process is known as segmentation. Most methods use intensity, texture and contrast
properties in the image over a period of time to construct a back ground model. The back ground model is updated by averaging
frames over a period of time to a account for slow changes in illumination. The back ground model is the subtracted on a pixel basis
from the current image. There proposed system segment the object using the information of the temperature. Humans are segment by
examining the hot object . Again not all hot object are humans for this reason this stage were also used for eliminating small clusters
of hot objects that are unlikely to be people.
2.6.2 Classification
After segmentation there is a group of hot objects after that it is require classifying the hot objects which are human and which
are not Amitage et al. used straightforward shape analysis. They used vertical histogram projection to get the shape of the object.
They find that human have a shape, similar to a normal Gaussian curve which is differ from other hot objects likes cars, busses etc.
by examining the shapes they classify the humans.
2.7 Summary
In this chapter various methods of human detection is described briefly. The main purpose of these researches is to detect human
more still images by computer. The works has some limitations for this reason researches are going on to improve the techniques.
The next chapter presents sour proposed human detection methodology.
3. PROPOSED HUMAN DETECTION METHODOLOGY
3.1 Introduction
In the previous chapter we have discussed various human detection systems, their successes and limitations. In this chapter we
will discuss on proposed human detection system. The discussion is broken into several modules: Image acquisition, edge detection,
and finally human detection by analyzing the shapes. Section 3.2 describes the proposed system architecture. Section 3.3 describes
the detail of human detection technique.

3.2 Proposed System Architecture

Figure 3.1 shows a basic architecture of proposed human detection. In this propose system, images are captured using a digital
camera. These images are passed through the human detection module. In the human detection module, input RGB images to
convert into Gray-scale images; then normalized boundary is compared with predefined templates and if sufficient match is found
then human is bounded by a rectangular box. After detecting human from the real-time image , the system can take several actions.
Such as it can aware about the presence of the human by making alarm or displaying some light signal instructions.

www.ijeais.org
71
International Journal of Engineering and Information Systems (IJEAIS)
ISSN: 2000-000X
Vol. 1 Issue 5, July– 2017, Pages: 62-85

Figure 3.1 Proposed System Architecture

3.3 Details of Human Detection
Although motion is a very important cue for recognizing actions, when we look at such images, we can more or less understand
human actions in the picture. This is mostly true in news or sports photographs, where the people are in stylized poses that reflect an
action.. We used a digital camera to capture the image. From the segmented images boundary is detected and detected is compared
with predefined templates for matching.
Human detection is closely related to general object recognition techniques. It involves two steps-training or learning phases and
classifier for human detection phases Figure 3.2 presents the flow chart of the proposed human detection method. Each part of the
work will be described in the following subsections. Subsection 3.3.1 focuses on the images Acquisition. Subsection 3.3.2 focuses
on the Gray-scale conversion from RGB images. Subsection 3.3.3 describes on shape/ boundary detection method. Subsection 3.3.4
describes Normalization technique and finally subsection 3.3.5 focuses on pattern matching approaches.

www.ijeais.org
72
International Journal of Engineering and Information Systems (IJEAIS)
ISSN: 2000-000X
Vol. 1 Issue 5, July– 2017, Pages: 62-85

Figure 3.2 Flowchart of The Proposed System

3.3.1 Images Acquisition

We can get images various way such as digital camera, scanner and store image on computer hard disk.
3.3.2 Gray Scale Conversion
There are several methods for image conversion. In this paper, we use the gray scale conversion for further process to detected
human area by traditional approach.

Newi = ( Ri + Gi + Bi ) / 3………………………………..(3.1)

www.ijeais.org
73
International Journal of Engineering and Information Systems (IJEAIS)
ISSN: 2000-000X
Vol. 1 Issue 5, July– 2017, Pages: 62-85

Figure 3.3 Gray Scale Conversion

3.3.3 Edge Detection
The Canny edge detection algorithm is known to many as the optimal edge detector. Canny's intentions were to enhance the
many edge detectors already out at the time he started his work. He was very successful in achieving his goal and his ideas and
methods can be found in his paper, "A Computational Approach to Edge Detection". In his paper, he followed a list of criteria to
improve current methods of edge detection. The first and most obvious is low error rate. It is important that edges occurring in
images should not be missed and that there be no responses to non-edges. The second criterion is that the edge points be well
localized. In other words, the distance between the edge pixels as found by the detector and the actual edge is to be at a minimum. A
third criterion is to have only one response to a single edge. This was implemented because the first two were not substantial enough
to completely eliminate the possibility of multiple responses to an edge. Based on these criteria, the canny edge detector first
smoothes the image to eliminate and noise. It then finds the image gradient to highlight regions with high spatial derivatives. The
algorithm then tracks along these regions and suppresses any pixel that is not at the maximum (nonmaximum suppression). The
gradient array is now further reduced by hysteresis. Hysteresis is used to track along the remaining pixels that have not been
suppressed. Hysteresis uses two thresholds and if the magnitude is below the first threshold, it is set to zero (made a non-edge). If the
magnitude is above the high threshold, it is made an edge. And if the magnitude is between the 2 thresholds, then it is set to zero
unless there is a path from this pixel to a pixel with a gradient above T2.
Step 1:-
In order to implement the canny edge detector algorithm, a series of steps must be followed. The first step is to filter out any
noise in the original image before trying to locate and detect any edges. And because the Gaussian filter can be computed using
a simple mask, it is used exclusively in the Canny algorithm. Once a suitable mask has been calculated, the Gaussian smoothing
can be performed using standard convolution methods. A convolution mask is usually much smaller than the actual image. As a
result, the mask is slid over the image, manipulating a square of pixels at a time. The larger the width of the Gaussian mask, the
lower is the detector's sensitivity to noise. The localization error in the detected edges also increases slightly as the Gaussian width is
increased.
Step 2:-
After smoothing the image and eliminating the noise, the next step is to find the edge strength by taking the gradient of the
image. The Sobel operator performs a 2-D spatial gradient measurement on an image. Then, the approximate absolute gradient
magnitude (edge strength) at each point can be found. The Sobel operator uses a pair of 3x3 convolution masks, one
estimating the gradient in the x-direction (columns) and the other estimating the gradient in the y-direction (rows). They are shown
below:

www.ijeais.org
74
International Journal of Engineering and Information Systems (IJEAIS)
ISSN: 2000-000X
Vol. 1 Issue 5, July– 2017, Pages: 62-85

The magnitude, or edge strength, of the gradient is then approximated using the formula:
|G| = |Gx| + |Gy|

Step 3:-
The direction of the edge is computed using the gradient in the x and y directions. However, an error will be generated when sum
X is equal to zero. So in the code there has to be a restriction set whenever this takes place. Whenever the gradient in the x direction
is equal to zero, the edge direction has to be equal to 90 degrees or 0 degrees, depending on what the value of the gradient in the y-
direction is equal to. If GY has a value of zero, the edge direction will equal 0 degrees. Otherwise the edge direction will equal 90
degrees. The formula for finding the edge direction is just:

Theta = invtan (Gy / Gx)

Step 4:-
Once the edge direction is known, the next step is to relate the edge direction to a direction that can be traced in an image. So if
the pixels of a 5x5 image are aligned as follows:

Then, it can be seen by looking at pixel "a", there are only four possible directions when describing the surrounding pixels - 0
degrees (in the horizontal direction), 45 degrees (along the positive diagonal), 90 degrees (in the vertical direction), or 135 degrees
(along the negative diagonal). So now the edge orientation has to be resolved into one of these four directions depending on which
direction it is closest to (e.g. if the orientation angle is found to be 3 degrees, make it zero degrees). Think of this as taking a
semicircle and dividing it into 5 regions.

www.ijeais.org
75
International Journal of Engineering and Information Systems (IJEAIS)
ISSN: 2000-000X
Vol. 1 Issue 5, July– 2017, Pages: 62-85
Therefore, any edge direction falling within the yellow range (0 to 22.5 & 157.5 to 180 degrees) is set to 0 degrees. Any edge
direction falling in the green range (22.5 to 67.5 degrees) is set to 45 degrees. Any edge direction falling in the blue range (67.5 to
112.5 degrees) is set to 90 degrees. And finally, any edge direction falling within the red range (112.5 to 157.5 degrees) is set to 135
degrees.
Step 5:-
After the edge directions are known, non-maximum suppression now has to be applied. Non- maximum suppression is used to
trace along the edge in the edge direction and suppress any pixel value (sets it equal to 0) that is not considered to be an edge. This
will give a thin line in the output image.
Step 6:-
Finally, hysteresis [12] is used as a means of eliminating streaking. Streaking is the breaking up of an edge contour caused by the
operator output fluctuating above and below the threshold. If a single threshold, T1 is applied to an image, and an edge has an
average strength equal to T1, then due to noise, there will be instances where the edge dips below the threshold. Equally it will also
extend above the threshold making an edge look like a dashed line. To avoid this, hysteresis uses 2 thresholds, a high and a low. Any
pixel in the image that has a value greater than T1 is presumed to be an edge pixel, and is marked as such immediately. Then, any
pixels that are connected to this edge pixel and that have a value greater than T2 are also selected as edge pixels. If you think of
following an edge, you need a gradient of T2 to start but you don't stop till you hit a gradient below T1

www.ijeais.org
76
International Journal of Engineering and Information Systems (IJEAIS)
ISSN: 2000-000X
Vol. 1 Issue 5, July– 2017, Pages: 62-85

Figure 3.4 Original Image and Edge Detected Image of Canny Operator

3.5 Normalization or Resize

Normalization phase scale up all edge images into equal size. Each edge image is scaled to a rectangular image of M`xN`
resolution. In our project, normalized images are of 180x200 resolutions. The edge image B[(0,0), (xm,ym)] is scale to image
N[(0,0), (64x64)] using the equation 3.3.

N ( xi , yi ) = B ( xi × Sx, yi × Sy )……………………….(3.3) Where, Sx =

64/xm and Sy = 64/ym

Figure 3.5 Example Scenario of Normalizing Method (A ) Source Image (B ) Before Normalizing (C) After Normalizing

3.3.6 Pattern Matching

After normalizing, the contour image is compared with the predefined template images. In section 33 we will describe template
image formation technique. The detail of our matching algorithm is as follows:

Step 1:
Read the contour image, and save the co-ordinates that hold 1
Step 2:
Compare with known templates and measure hit and miss score.
a) Hit Score:

www.ijeais.org
77
International Journal of Engineering and Information Systems (IJEAIS)
ISSN: 2000-000X
Vol. 1 Issue 5, July– 2017, Pages: 62-85
If Tst ( x, y ) = Tmp ( x, y ) = 1 then increment Hit Score.

b) Miss Score:
If Tst ( x, y ) ≠ Tmp ( x, y ) then increment Miss Score.

Step 3:
Calculate the Hit ratio

Hit Ratio:
H
Hr =
H+M where, H = Total Hit Score and
M = Total Miss Score

Step 4:
If Hr is greater than predefined threshold then human is localized and bounded using a rectangle color box.

The threshold is selected through experiment. We considered the threshold value is as 0.75. This box is drawn by calculating the
minimum and maximum values of X and Y co-ordinates.

3.4 Template Image Formation Technique

To train the human detection system we collect several template images to detect the shape/contour of human. This system
prepares the template by detecting the shape area, eliminating noises, detecting edge, and finally scales to 180×200. This boundaries
of the human are used for matching with every captured image. We have taken 400 images of different human in the different poses
and in different scenarios. All these template images are kept in size
180×200 resolutions. Figure shows example of several template images and figure shows the extracted boundaries of those
images.

www.ijeais.org
78
International Journal of Engineering and Information Systems (IJEAIS)
ISSN: 2000-000X
Vol. 1 Issue 5, July– 2017, Pages: 62-85

Figure 3.6 Example Template images and shape templates

3.5 Summary
In this chapter we have discussed different modules of proposed human detection system. This technique is mainly focused on
detecting objects, noise elimination, shape detection and linking, and finally matching module. Next chapter presents experiments,
results with related discussions.
4 EXPERIMENT, RESULT AND DISCUSSIONS
4.1 Introduction
This chapter presents experimental results together with valuable discussion. The experimental result focuses on following areas:
results of conversion method, results of filtering method, results of boundary detection method and finally results of human detection
method. Section 4.2 describes the experiment setup. Section 4.3 present the results of human detection.
4.2 Experiment Setup
In this thesis we uses core2 duo 2.8GHz PC with 1 GB RAM. We also use application software for subtract the background of
an image. This image will be input to the system for further processing.
The proposed method is implemented using Microsoft Visual C++(6.0) , OpenCV library function, OpenGL programming
language and Adobe Photoshop CS2.

www.ijeais.org
79
International Journal of Engineering and Information Systems (IJEAIS)
ISSN: 2000-000X
Vol. 1 Issue 5, July– 2017, Pages: 62-85

Figure 4.1: Computer System

4.3. Experimental Results of Proposed System
This proposed system can detect human from white background still image. This section will present the result of the system.
Subsection 4.3.1 show the result of color conversion method, Subsection 4.3.2 present the result of boundary detection method,
Subsection 4.3.3 present the result of resizing method and finally Subsection 4.3.4 show the result of human detection method.
4.3.1 Result of Color Conversion
Color can be converted by use various method but we proposed the traditional approach that can convert the image from RGB to
GRAY.

www.ijeais.org
80
International Journal of Engineering and Information Systems (IJEAIS)
ISSN: 2000-000X
Vol. 1 Issue 5, July– 2017, Pages: 62-85
Figure: 4.2 A) Source Image B) Gray Image

4.3.2 Results of Boundary or Edge Detection

Boundary can be detected by scanning the image using masking but we proposed an algorithm, which is called ( canny edge
algorithm ) that scans the image for detecting edges. Figure 4.5 shows the detected boundaries by using canny edge algorithm.

Figure 4.3: Result of edge detect

Performance of human detection varies, based on boundary detection techniques. Table 4.2 shows the comparison of human
detection using canny edge operator boundary detection methods.

Table 4.1: Experiment Result of human Detection where Boundary is Detected using canny edge Scanning.
Edge Detection Method

Experiment Accuracy (%)

(Scan Edge)
1 92.4

2 95.5

3 93.33

4 96.97

5 90.87

6 94.2

Average =93.87

www.ijeais.org
81
International Journal of Engineering and Information Systems (IJEAIS)
ISSN: 2000-000X
Vol. 1 Issue 5, July– 2017, Pages: 62-85

4.3.3 Result of Resizing Method

For resize the we used a formula, which given below:

N ( xi , yi ) = B ( xi × Sx, yi × Sy )………………………….(3.3) Where,

Sx = 64/xm and Sy = 64/ym

Figure 4.4: Result of Resizing Image

4.3.4 Results of human Detection

This thesis finally gives a visual output by surrounding a rectangular box of the detected human. Figure 4.7 shows a sample
visual output. Left side of the window is the original image that is inputted by subtracting the background and right side of the
window shows the detected human by making a rectangular box around the human.

Figure 4.5: Sample Visual Output of Human Detection

www.ijeais.org
82
International Journal of Engineering and Information Systems (IJEAIS)
ISSN: 2000-000X
Vol. 1 Issue 5, July– 2017, Pages: 62-85

Figure: 4.6 shows example results of human detection. Only true pedestrian are detected and bounded by a rectangular green
box.

Performance of human detection method

Table 4.3 represents the performance evaluation of out proposed human detection method. Equation 4.1 and 4.2 defines the
accuracy and precision of human detection method
Pp = C / C + F * 100
Here, Pp = Precision (%) of human detection
P = Number of total human
C = Number of Correctly detected human
F = Number of False detection

Table 4.2 : Performance of Human Detection Method

Experiment Human Detected False Detection Precision
Correctly (F) (%)
© (Pp)

#1 95 5 95
#2 38 0 100
#3 280 9 96.88
#4 378 25 93.79
#5 488 10 97.99
#6 370 12 96.85

#7 520 19 96.47

www.ijeais.org
83
International Journal of Engineering and Information Systems (IJEAIS)
ISSN: 2000-000X
Vol. 1 Issue 5, July– 2017, Pages: 62-85

Performance of human detection method in term of accuracy is presented in figure 4.9 . In experiment number #7, we can see
from the figure that the accuracy is highest. The duration of this experiment is 60 seconds and total correctly detected human is 120
among 150 human. In this case lighting condition was satisfactory. In some cases slight degradation of accuracy is observed. This is
due to variation of noises and cluttered background. Figure 4.10 presents the precision of human detection method for various
experiments. In some cases the precision is very low. This is also due to variation background and noise. Overall performance of
the detection method is satisfactory. The average precision is 92.72% and the average accuracy is
90.05%.
From the above experiment we can say that this very efficient and successful in detecting human in different poses. The outcome
can be used in the area where human or pedestrians are related. Mainly the idea of this thesis was come from the idea of autonomous
driving. When in future there will be no driver on the vehicle and the vehicle will able to drive itself then the idea of this thesis can
be used.

4.4 Summary
In this chapter we have presented experimental results of the proposed human detection system.
In the next chapter we conclude our work by mentioning the major contributions, limitations of the system and our future works.
5. CONCLUSION AND FUTURE WORKS

This thesis work describes the human detection from still images using the contour/boundary based matching. This system is
tested in different position for detecting human. From the experimental result we conclude that performance of the system is
satisfactory.

5.1 Contribution

This thesis has great contributions in the field of image analysis and object recognition. In this thesis we have presented an
algorithm for detecting human. The proposed system runs with satisfactory success rates. The contribution of our work can be
summarized as follows.
- A novel method for detecting human from still images.
- A novel method for shape/contour detection from still images.
- A satisfactory performance in detecting human using contour/shape matching.
- Average accuracy and precision of human detection method in the system is 93.05%
And 95.72%.
5.2 Limitation and Future Works
The major limitation of the thesis is that the system is not dynamic. It only can detect human from still images.
Another limitation of the thesis is that the system cannot detect if the multiple human is present
There are situation such as when a human is just backside of another human, this thesis cannot detect the two human separately.
Another limitation of this thesis is that the system cannot subtract the background automatically. It is to do by manually for the
system input images to detect the human.
In future we will make this thesis more robust against the various limitations and detect multiple human.
5.3 Concluding Remarks
The ultimate goal of this thesis was to established an efficient, robust and user friendly system to detect human. Our achievement
from this research is satisfactory. We aspire to do more research in the same field.

www.ijeais.org
84
International Journal of Engineering and Information Systems (IJEAIS)
ISSN: 2000-000X
Vol. 1 Issue 5, July– 2017, Pages: 62-85
REFERENCES
[1] Harris, C., Stephens., M.: A combined corner and edge detector. Alvey Vision Conference (1998) 147–151
[2] Mikolajczyk, K., Schmid, C.: An affine invariant interest point detector. 7th European Conference on Computer Vision 1
(2002) 128–142
[3] Lindeberg, T.: Feature detection with automatic scale selection. International Journal of Computer Vision 30 (1998) 79–116
[4] Lowe, D.G.: Local feature view clustering for 3D object recognition. Conference on Computer Vision and Pattern Recognition
(2001) 682–688
[5] Kadir, T., Brady, M.: Scale, saliency and image description. International Journal of Computer Vision 45 (2001) 83–105
[6] Cordelia, R.R., Schmid, C., Triggs, B.: Learning to parse pictures of people. European Conference on Computer Vision (2002)
700–714
[7] Dalal, N., Triggs, B.: Histograms of oriented gradients for human detection. IEEE Computer Society Conference on Computer
Vision and Pattern Recognition 1 (2005) 886–893
[8] Martin, D.R., Fowlkes, C.C., Malik, J.: Learning to detect natural image boundaries using local brightness, color, and texture
cues. IEEE Transactions on Pattern Analysis and Machine Intelligence 26 (2004) 530 –549
[9] Sirovitch, L., Kirby, M.: Low-dimensional procedure for the characterization of human faces. Journal of the Optical Society of
America 2 (1987) 586–591
[10] Zabih, R., Woodfill, J.: Non-parametric local transforms for computing visual correspondence. ECCV ’94:
Proceedings of the Third European Conference-Volume II on Computer Vision (1994) 151–158 on Pattern Analysis and
Machine Intelligence 20 (1998) 415–423
[11] Mittal, A., Visvanathan, R.: An intensity-augmented ordinal measure for visual correspondence. CVPR ’06: Proceedings of
the 2006 IEEE Computer Society Conferenceon Computer Vision and Pattern Recognition (2006) 849–856
[12] Singh, M., Parameswaran, V., Ramesh, V.: Order consistent change detection via fast statistical significance testing. IEEE
Conference on Computer Vision and Pattern Recognition, 2008 (2008)
[13] Gupta, R., Mittal, A.: SMD: A locally stable monotonic change invariant feature descriptor. 10th European Conference on
Computer Vision (2008) 265–277
[14] Ojala, T., Pietikainen, M., Harwood, D.: A comparative study of texture measures with classification based on featured
distributions. Pattern Recognition 29 (1996) 51–59
[15] Ojala, T., Pietik¨ainen, M., M¨aenp¨a¨a, T.: Multiresolution gray-scale and rotation invariant texture classification with local
binary patterns. IEEE Transactions on Pattern Analysis and Machine Intelligence 24 (2002) 971–987
[16] Heikkil¨a, M., Pietik¨ainen, M., Schmid, C.: Description of interest regions with local binary patterns. Pattern Recognition 42
(2009) 425–436
[17] Gupta, R., Patil, H., Mittal, A.: Robust order-based methods for feature description. In: IEEE Conference on Computer Vision
and Pattern Recognition. (2010)
[18] Papageorgiou, C., Poggio, T.: A trainable system for object detection. International Journal of Computer Vision 38 (2000) 15–
33
[19] 20. Mohan, A., Papageorgiou, C., Poggio, T.: Example-based object detection in images by components. IEEE Transactions
on Pattern Analysis and Machine Intelligence 23 (2001) 349–361
[20] 21. Gavrila, D., Philomin, V.: Real-time object detection for smart vehicles. Proceedings of the 7th International Conference
on Computer Vision 87{93 (1999)
[21] 22. Gavrila, D.M., Giebel, J., Munder, S.: Vision-based pedestrian detection: The
[22] PROTECTOR system. IEEE Intelligent Vehicles Symposium (2004) 13–18
[23] Jones, M., Viola, P., Viola, P., Jones, M.J., Snow, D., Snow, D.: Detecting pedestrians using patterns of motion and
appearance. International Conference on Computer Vision (2003) 734–741
[24] Schapire, R.E.: The boosting approach to machine learning, an overview. MSRI Workshop on Nonlinear Estimation and
Classification (2002) 25International Journal of Computer Vision 43 (2001) 45–68
[25] Krystian Mikolajczyk, C.S., Zisserman, A.: Human detection based on a probabilistic assembly of robust part detectors.
European Conference on Computer Vision (2004) 69–82
[26] Wang, X., Han, T.X., Yan, S.: An hog-lbp human detector with partial occlusion handling. International Conference on
Computer Vision (2009)

www.ijeais.org
85