Pedestrian detection has a wide range of application scenarios in the fields of intelligent monitoring and security, but there is a problem that it cannot achieve sufficient accuracy at sufficient speed. In this context, this paper proposes an improved YOLOv3 pedestrian detection method using combined attention mechanism. Firstly, K-means clustering algorithm is used to extract the prior box of INRIA dataset, so as to obtain the prior boxes more suitable for INRIA dataset. Secondly, SE block is added to the three effective feature layers derived from the backbone Darknet53, which brings a small number of parameters, but can add weight to the features of each channel. Thus removing the side effects of ineffective or weak feature layers. The addition of SE block can effectively improve the accuracy with a small number of parameters. Finally, on the INRIA dataset, the improved YOLOv3 algorithm improves mAP by 3.15% compared with YOLOv3, and is also higher than the similar two-stage algorithm.
Improved YOLO Pedestrian Detection Algorithm Based on Attention Mechanism
2022-10-12
1359370 byte
Conference paper
Electronic Resource
English
Pedestrian Detection Based on YOLO Network Model
British Library Conference Proceedings | 2018
|Ship detection based on improved YOLO algorithm
IEEE | 2023
|Using YOLO-based pedestrian detection for monitoring UAV
SPIE | 2019
|Infrared Image Pedestrian Detection via YOLO-V3
IEEE | 2021
|