Pedestrian Detection Algorithm Based on Improved YOLOv3_tiny

Abstract

In view of the low detection accuracy of YOLOv3_tiny algorithm on small pedestrian target, a pedestrian detection method to improve YOLOv3_tiny is proposed. Firstly, the head and shoulder of the pedestrian are taken as the detection object. The K-means++ algorithm is used to improve the k-means algorithm and cluster the data set to get the anchor frame with higher accuracy. Secondly, add a target prediction layer with a resolution of 52 × 52 to the multi-scale prediction section, which improves the detection accuracy of the small target. In addition, the backbone structure of YOLOv3_tiny is light-weighted, which further compresses the model memory and reduces the deployment burden of the model while maintaining accuracy. A comparative experiment is conducted on the self-made pedestrian dataset which tags the head and shoulder. The results show that the recall of the improved algorithm is 0.8743, which is 11.78% higher than that of the original algorithm; The average accuracy is 77.69%, which is better than the classical algorithms, such as SSD, fast RCNN and Yolo v3; the model size is 26M, and the test time of single image is 0.0789s, which can achieve efficient and real-time detection.

Research on lightweight pedestrian detection based on improved YOLOv5

“…Xu et al solved the efficiency problem of pedestrian detection by model reconstruction and pruning of the YOLOv3 network but did not consider this special case of pedestrians in gauge frequency due to the high number or in a dense state, and there will be a high leakage rate in detection [7]. Liu et al addressed the effectiveness of pedestrian detection in hazy weather with a weighted combination layer that combines multi-scale feature maps with squeezing and excitation blocks, but the additional computation does not apply to embedded devices and the detection frame rate is low [8].…”