Pedestrian Partial Occlusion Detection Method Based on Improved YOLO Network Structure
Guangxi Chen, Zhan Baoying, Min Tang, Xie Chen · 2019
The problem of pedestrian detection accuracy is low due to partial occlusion of pedestrians in practical application scenarios, a pedestrian occlusion detection method based on improved YOLOv3 network structure is proposed. By improving the activation function and clustering method to strengthen the feature output layer of the shallow layer of the network, which effectively make the accuracy of detection better. The experimental results show that the mAP of proposed method is improved by 2.63 percent in the INRIA data set, and the improved network has good robustness and generalization ability, and guarantees the real-time of detection.