This project aims at efficiently handling vehicle parking services across parking lots. Vehicles are uniquely identified using their number plate information and tracked across camera locations. The location of a parked car is monitored for easy reference and access. Parking slot vacancy is also computed for managing the parking capacity of a parking lot. Focus has been placed in making the system robust for handling low-light surroundings and unconstrained camera angles.
- Vehicle Detection Models are based on the YOLOv4 architecture executed using the darknet framework
- ALPR is handled using a custom Keras Model for Licence Plate Detection and a YOLO based Detector for performing Numberplate character OCR
- TensorFlow - 1.15.4 (CPU) / TensorFlow 1.13.1 (GPU - Colab)
- Keras - 2.2.4
- FastAPI
- OpenCV
- imutils
- nest-asyncio
- pyngrok
- asgiref
- Python 3.6+
- To install required packages on local system:
pip install -r requirements.txt- For CPU Only execution, do the corresponding changes in
./darknet/Makefile
GPU = 0
CUDNN = 0
CUDNN_HALF = 0
- Building YOLO Darknet Binaries
cd darknet && makepython main.pyTo test out the repository on Google Colab, check out the notebooks folder
- Vehicle Detection: https://arxiv.org/abs/2011.04244
- Licence Plate Detection in Unconstrained Scenes: http://sergiomsilva.com/pubs/alpr-unconstrained/