sontranbk/mot_licenseplate

This product features an Intelligent Transport System with Multiple object tracking and License Plate detection. Main functions: Multiple object tracking with Re-id support Object counting and area-contour counting License Plate (Vietnam Standard) detection, with text recognition IoT system oriented with Jetson Nano deployment as edged devices

Getting started

Pip

pip install -r requirements.txt
cd YOLOX
pip install -v -e .

Nvidia Driver

Make sure to use CUDA Toolkit version 11.2 as it is the proper version for the Torch version used in this repository: https://developer.nvidia.com/cuda-11.2.0-download-archive

torch2trt

Clone this repository and install: https://github.com/NVIDIA-AI-IOT/torch2trt

Download a pretrained model

Download pretrained yolox_s.pth file: https://github.com/Megvii-BaseDetection/YOLOX/releases/download/0.1.1rc0/yolox_s.pth

Copy and paste yolox_s.pth from your downloads folder into the 'YOLOX' folder of this repository.

Convert model to TensorRT

python tools/trt.py -n yolox-s -c yolox_s.pth

Runing with YOLOX-s

python detector.py

References

• YOLOX
• ByteTrack

mot_licenseplate's Issues

Tối ưu phần cứng

Jetson Nano đã có thể nhận được phần GPU. Tuy nhiên khi chạy project, Jetson Nano đã bị đơ và treo máy.
Các bước nâng cấp: chuyển file weights từ format .pt sang .onnx để tăng tốc độ, còn lại chưa thay đổi quá sâu.