Front Rear Car Recognition for Edge Devices

Automatic License Plate Recognition (ALPR) is one of the most interest research topics in recent years because of its vast applications ranging from law enforcement to intelligent transportation systems. Although the ALPR pipeline has grown and matured to have a standard format, different extensions were introduced to it like the addition of driver distraction detection, and the front-rear detection model. We propose an extension to the pipeline that adds the ability to detect car orientations (front, rear) with an accuracy of 96%, which is comparable to the state-of-the-art models while maintaining low memory, file size on disk, and decent performance on edge devices like raspberry pi 4 around 15 frames per second (FPS) and more than 150 Frame per second in CPU.