Title: Video stream safety helmet recognition method based on improved faster R-CNN

Authors: Xian Yang; Longjin Chen; Zongyi Wang; Shuyu Lin; Danqing Luo; Ziwen Cai; Yuxin Lu

Addresses: Hainan Power Grid Co., Ltd., Haikou, 570311, Hainan, China ' Hainan Power Grid Co., Ltd., Haikou, 570311, Hainan, China ' China Southern Power Grid Research Institute Co., Ltd., Guangzhou, 510663, Guangdong, China ' Hainan Power Grid Co., Ltd., Haikou, 570311, Hainan, China ' Hainan Power Grid Co., Ltd., Haikou, 570311, Hainan, China ' China Southern Power Grid Research Institute Co., Ltd., Guangzhou, 510663, Guangdong, China ' China Southern Power Grid Research Institute Co., Ltd., Guangzhou, 510663, Guangdong, China

Abstract: In order to improve the recognition accuracy and speed of safety helmets, a video stream safety helmet recognition method based on improved Faster R-CNN is proposed. By using the weighted average method to greyscale video stream images, the processing flow is simplified, and median filtering is used to remove noise and improve image quality. The improved Faster R-CNN significantly improves the accuracy and robustness of object recognition by combining the PISA module and deformable convolutional layer (DCN). The PISA module optimises sample weight allocation through importance-based sample weights (ISR) and CARL loss functions. DCN enhances the accuracy of feature extraction by learning the offset of receptive field sampling points. The experimental results show that the safety helmet recognition accuracy of our method reaches 97%, and the recognition speed reaches 30FPS. The improved Faster R-CNN can effectively address the challenge of improving target recognition accuracy in complex scenarios.

Keywords: improved faster R-CNN; video stream; recognition of safety helmets; sample weight.

DOI: 10.1504/IJISTA.2025.145637

International Journal of Intelligent Systems Technologies and Applications, 2025 Vol.23 No.1/2, pp.202 - 216

Received: 24 Sep 2024
Accepted: 10 Jan 2025
Published online: 09 Apr 2025 *

Full-text access for editors Full-text access for subscribers Purchase this article Comment on this article