Abstract

Pedestrian detection is widely used in daily life, but it is difficult to study in highway environment, such as occlusion overlap. In order to reduce error detection rate of highway pedestrian detection, an algorithm based on YOLOv5s was proposed. Since vehicle occlusion leads to the reduction of effective features of targets, CBAM attention and SE-NET mechanism module is introduced in the network of YOLOv5s to maximize the extraction of effective features. In order to prevent the spatial characteristic information in the trunk network from being damaged, CBAM module is added at the beginning and end of the structure, and SE-Net attention module is added in the neck network, that is, after the detection layer C3 module, the weight information obtained is connected with the subsequent Conv module, so that the model pays more attention to the pedestrian area. Due to low detection accuracy caused by pedestrian overlap. YOLOv5s was designed by combining DIOU_NMS candidate box screening mechanism. The results show that the mean average precision of YOLOv5s (IOU=0.5) increases by 0.48, and the value of Recall of the improved algorithm increases by 0.51 respectively. The improved pedestrian detection algorithm improves the accuracy of target box regression. Thus, the confidence of pedestrian detection is improved. Based on the improvement strategies mentioned above, the detection speed is 32fps, which meets the requirements of real-time detection.

Full Text
Published version (Free)

Talk to us

Join us for a 30 min session where you can share your feedback and ask us any queries you have

Schedule a call