基于无人驾驶领域的飞速发展,为提高道路行人目标检测的速度和精度,提出一种基于YOLOv5网络改进的YW-YOLO的道路行人目标检测方法,在YOLOv5模型的neck结构中改入RepGFPN,充分交换高级语义信息和低级空间信息,添加自适应融合机制,引入SimAM注意力模块机制,提高算法的特征提取能力,在损失函数方面,使用Optimal Transport Assignment优化损失函数.实验结果表明,所提算法与原算法相比,在道路行人类别数据集上识别精确率由38.1%提升到52.6%,检测速度由29.4 fps提高到30.8 fps,具有更好的检测效果.
Pedestrian recognition research on unmanned roads with improved YOLOv5 algorithm
Based on the rapid development of the unmanned field,in order to improve the speed and accuracy of road pedestrian target detection,a road pedestrian target detection method based on YOLOv5 network improved by YW-YOLO is proposed,which is changed into RepGFPN in the YOLOv5 model's neck structure,which fully exchanges the high-level semantic information and the low-level spatial information,adds the adaptive fusion mechanism,introduces the SimAM attention module mechanism,which improves the feature extraction ability of the algorithm,and in terms of loss function,optimal transport assignment is used to optimize the loss function.The experimental results show that the proposed algorithm in this paper,compared with the original algorithm,the recognition accuracy rate on the road pedestrian category dataset is improved from 38.1%to 52.6%,detection speed increased from 29.4 fps to 30.8 fps,which has a better detection effect.
NETL
NSTL
万方数据
