Object Detection System Based on Faster R-CNN
Jiangdong Lu, Dongfang Li, Meng Wang, Boyan Mi, Penglong Wang, Zhuochen Dai, Fen Zheng · 2022
Aiming at the low efficiency of image target detection in cloud computing mode, a target detection system suitable for edge devices is designed. First, the system selects Faster R-CNN in the deep learning algorithm as the target detection and recognition model, and trims the network feature extraction layer through the residual module. Second, a proposal region extraction sub-network with adjustable anchor boxes is used to obtain proposal regions more quickly by setting a convolutional sliding window of reasonable size. Finally, a complete target detection system is built using hardware such as Raspberry Pi development board and Intel neural computing stick. The experimental results on the KITTI dataset show that the system achieves good detection results, and achieves a faster recognition speed without reducing the target detection accuracy, which can meet the real-time requirements of offline work.