WCSE 2022 Spring
ISBN: 978-981-18-5852-9 DOI: 10.18178/wcse.2022.04.065

Object Detection of Roadside Pedestrians and Vehicles Based on Improved YOLO v5

Qirui Li, Wanyu Deng, Huijiao Xu, Xiaoting Feng

Abstract— Target detection algorithms based on deep learning are more and more widely used in the field of autonomous driving. At present, the target detection of roadside pedestrians and vehicles in the field of autonomous driving mainly faces the problems of too many types of vehicles, complex detection target backgrounds, overlapping of people and vehicles, and too many small targets. In view of the existing problems, this paper improves the YOLO v5 algorithm, optimizes the feature extraction network of YOLO v5, adds a small target detection head, improves the ability of the backbone network to extract target features, and strengthens the detection of small targets; adds the Selayer attention mechanism to improve the model sensitivity to channel features and enhances the network's ability to detect occluded targets. The experimental results show that the improved YOLO v5 model can achieve an average P of 91.3% for the detection of 10 kinds of roadside pedestrians and vehicles, and the mAP@0.5 of 80.1%. Compared with the original YOLO v5, the P is increased by 8.4%, and the mAP@0.5 is increased by 5.2%.

Index Terms— YOLO v5, target detection, attention mechanism, small target detection.

Qirui Li
College of Automation, Shenyang Institute of engineering
Wanyu Deng
College of Automation, Shenyang Institute of engineering
Huijiao Xu
College of Automation, Shenyang Institute of engineering
Xiaoting Feng
College of Automation, Shenyang Institute of engineering

[Download]


Cite: Qirui Li, Wanyu Deng, Huijiao Xu, Xiaoting Feng, " Object Detection of Roadside Pedestrians and Vehicles Based on Improved YOLO v5, " WCSE 2022 Spring Event: 2022 9th International Conference on Industrial Engineering and Applications, pp. 554-560, Sanya, China, April 15-18, 2022.