Implementation of a Real-time Uneven Pavement Detection System on FPGA Platforms

Abstract

Potholes or uneven road surfaces can lead to flat tires, suspension damage, or even accidents. A real-time uneven pavement detection system can provide drivers with information beforehand to reduce car damage and safety risk. It also can be used to inform road repair and maintenance departments to save the efforts of manual inspection. The proposed real-time detection system employs the YOLO-v4 algorithm as the detection model followed by quantization using the Vitis-AI framework for model compression so that the developed system can be performed on the Xilinx FPGA platform without compromising on accuracy and speed. Experimental results show that the proposed system can obtain 28 FPS with four-thread running at 300 MHz.