Year: 2024
Type: conference-paper
Abstract: Aiming at the problem that the pedestrian detection effect of illumination change, occlusion and different postures is not good in intelligent traffic scene, an optimized YOLOv8 is proposed to enhance its pedestrian detection ability. Based on YOLOv8, Coordinate Attention (CoordAtt) is introduced into the backbone network to enhance the sensitivity of the model to pedestrian location information and improve the detection accuracy. On the public test benchmark of Caltech Pedestrian Dataset, compared with the baseline algorithm, the proposed algorithm improves mAP0.5 by 1.3%, mAP0.5: 0.95 by 0.3%, and LAMR by 1.8%. Compared with the popular algorithm, mAP0.5 and mAP0.5: 0.95 are as high as 0.875 and 0.543 respectively, and LAMR is as low as 0.286, which is better than the popular algorithm. The algorithm has been tested to show that the accuracy of excellent performance detection has been significantly improved, and the phenomenon of missed detection has been greatly reduced, which can serve the intelligent transportation more effectively and reduce the potential safety hazards caused by missed detection.
Authors: Yurong Yue, Lihui Pan, Wei Shan
Institution: Huaibei Normal University
Language: English
Subfield: Computer Vision and Pattern Recognition
Field: Computer Science
Domain: Physical Sciences
Open Access status: gold
OpenAlex ID: W4405284416