Hardware Acceleration of YOLO-based Convolutional Neural Network Detector by Deep Learning Processor Unit for Intelligent Autonomous Mover
Kuan-Hung Chen, Chung-Bin Wu, Yin‐Tsung Hwang, Hua-Luen Chen, Jie-Min Lin, Chih‐Peng Fan · 2022 IET International Conference on Engineering Technologies and Applications (IET-ICETA) · 2022
For intelligent autonomous vehicles, they play an important role in user-friendly usage in crowded environments, so the recent focus has been on providing object recognition and collision avoidance functions. In this study, a Xilinx Deep Learning Processor Unit (DPU)-based Light Convolutional Neural Network (CNN) model was implemented for object detection and classification, and FPGA-based hardware acceleration was used to efficiently detect pedestrians and other objects. With the DPU-based implementation, the accuracy of the proposed object classification and face orientation detection model reaches up to 90% on image datasets collected in supermarkets. Compared to previous GPU-based software implementations, the proposed approach provides better frames-per-second (FPS) performance for real-time applications.