Indian ANPR (Automatic Number Plate Recognition)
Documentation
1. Introduction
Automatic Number Plate Recognition (ANPR) is a computer vision-based technology used
to detect and recognize vehicle license plates from images or videos. This program is
specifically designed for Indian vehicle number plates. It uses a combination of YOLOv8
(for object detection) and EasyOCR (for text recognition) to extract vehicle registration
numbers and identify the type of vehicle (Car, Truck, Bus, Motorcycle). The program
processes a video file, detects vehicles and plates frame by frame, applies OCR to extract
the text, and saves the recognized results into Excel, CSV, and log files in Google Drive.
2. Methodology (Procedure)
1. Input Setup: Video file is provided as input and Google Drive is mounted for
reading/writing results. 2. Preprocessing: Video is read frame by frame, skipping frames
for efficiency. 3. Detection: YOLOv8 detects vehicles and possible license plate regions. 4.
OCR Extraction: Cropped plate regions are passed to EasyOCR to recognize text. 5.
Validation: Plates are cleaned, normalized, and checked against Indian formats. 6. Vehicle
Classification: YOLO assigns type (Car, Truck, Bus, Motorcycle). 7. Saving Results:
Recognized numbers and vehicle types are stored in CSV, Excel, Debug Images, and
Logs.
3. Algorithm
The program uses two main algorithms: A. YOLOv8 (Object Detection) - Detects vehicles
and number plate regions in each frame. - Uses deep learning to predict bounding boxes
and classes. - Real-time, high-speed detection. B. EasyOCR (Optical Character
Recognition) - Extracts alphanumeric text from cropped number plate images. - Uses
CNNs and RNNs to detect and recognize text sequences. - Post-processing ensures
normalization and validation of results. C. Post-Processing - Normalizes extracted text
(removes spaces, dashes, replaces O→0). - Validates against expected Indian plate
formats. - Removes duplicates and saves final results.
4. Accuracy, Input and Output
Accuracy: - YOLOv8 Accuracy: ~70–90% depending on video clarity and lighting. -
EasyOCR Accuracy: ~60–85% for Indian plates (due to variations in fonts, blur, etc.). -
Overall System Accuracy: ~65–80% under normal conditions. Input: - A video file (.mp4)
containing vehicles on the road. Output: - CSV File: Contains vehicle numbers and types. -
Excel File: Same as CSV but with formatting for readability. - Log File: Frame-wise
detection logs with confidence levels. - Debug Images: Cropped license plate images for
verification. Example Output: | Vehicle Number | Vehicle Type | |----------------|--------------| |
TN01AB1234 | Car | | AP22XY5678 | Truck | | KA09CD4321 | Motorcycle |
5. Conclusion
This program demonstrates a robust Indian ANPR system using YOLOv8 and EasyOCR.
It identifies number plates, validates them, and classifies vehicles into categories. The
results are stored in Excel, CSV, and debug images, making it practical for traffic
monitoring, toll collection, and security applications. Future improvements can include
training YOLOv8 with custom Indian datasets, using OCR models fine-tuned for license
plates, and applying super-resolution for clearer text recognition. In conclusion, the system
is a practical, automated, and scalable ANPR pipeline for Indian traffic environments.
