Pedestrian Detection Based on Improved YOLOv3 Network
Yongbo Yu, Shaochen Yan, Xinpeng Hao · 2023
Pedestrian detection has important applications in driverless and road detection. For the common problems of pedestrian detection such as multi-scale target and occlusion, this paper improves the YOLOv3 network structure by introducing CBAM attention mechanism and SPP structure to improve the expression ability of the output feature layer of the backbone network, and greatly improves the perceptual field to enhance the network's perception of features. By conducting experiments on the INRIA Person dataset, The experimental results show that the improved YOLOv3 network improves the detection accuracy from 89.95% of the original YOLOv3 network to 92.06%, an improvement of 2.3 percentage points, which is a significant improvement in accuracy. Compared with YOLOv3, the improved network in this paper has more outstanding effect on real-time detection.