Research on Pedestrian Detection Based on Improved YOLOv3
Conference: ICETIS 2022 - 7th International Conference on Electronic Technology and Information Science
01/21/2022 - 01/23/2022 at Harbin, China
Proceedings: ICETIS 2022
Pages: 4Language: englishTyp: PDF
Authors:
Wang, Hansong; Liang, Quan; Xiong, Neng; Hu, Jinjing (School of Computer Science and Mathematics, Fujian University of Technology, Fujian Provincial Key Laboratory of Big Data Ming and Application, Fuzhou, China)
Abstract:
The YOLOv3 algorithm is an anchor-based algorithm, which is difficult to cover all shapes of targets. It is often necessary to perform cluster analysis to determine the optimal anchors before training. These anchors have weak generalization performance and increase the complexity of detection. Cause the detection speed to slow down. Aiming at the different poses of pedestrians in pedestrian detection, the paper proposes an improved YOLOv3 algorithm for pedestrian detection. The anchor free method is used to cover all shapes of pedestrian poses, and the HSV data enhancement method is used to process the training images to improve the accuracy and speed of pedestrian detection. The experimental results show that compared with the original algorithm, in the INRIA pedestrian dataset and the Caltech pedestrian dataset, the improved algorithm in the article has increased accuracy by 9.23% and 5.01%. The speed has increased by 0.021 s and 0.025 s, respectively.