ANPR utilizing YOLOX object detection from a custom trained dataset to detect license plates, and deepsort object tracking. The detection frames then pipelined to Tesseract OCR to detect characters (TODO)
Don't forget to install yolox first, guide can be found on the official YOLOX repository:
Megvii-BaseDetection YOLOX
or you can just simply
python3 ./setup.py
Detailed PyTesseract installation guide can be read here: www.projectpro.io, Install PyTesseract
to run detection on a video:
python3 ./demo.py --path ./path/to/video
Original dataset (VOC2017 format) that is used can be found at andrewmd - Car License Plate Detection However, training is done on COCO JSON Format dataset that could be found here: COCO ANPR - Google Drive
- Optimize model accuracy
- Optimize OCR detection
- Restructure project
- Optimize engine efficiency