一种改进的YOLOv3红外图像行人检测方法

An Improved YOLO v3 Infrared Image Pedestrian Detection Method

针对夜间交叉路口光照差和基于视觉行人检测存在不连续的问题,提出了一种改进的YOLO v3红外图像行人检测方法。建立了城市道路交叉路口行人红外图像数据集,使用k-means++算法重新聚类行人数据集,改进边框回归损失函数和网络结构,在NVIDIA GTX1060平台上进行了模型测试。实验结果表明,改进后的YOLO v3算法模型平均准确率较改进前提高了5.75%,检测速度达到24 f/s。

In view of the poor illumination at night intersections and the sustainability problem of vision-based pedestrian detection,an improved YOLO v3 infrared image pedestrian detection method is proposed.An infrared image data set of pedestrians at urban road intersections is established,the k-means++algorithm is used to obtain the width and height of the target candidate frame,the frame regression loss function is improved,and the network structure is modified.Experimental verification is carried out on the NVIDIA GTX1060 platform.The results show that the average accuracy of the improved model is 5.75%higher than that of the original YOLO v3,and the detection speed reaches 24 f/s.