基于YOLO的轻量红外图像行人检测方法

红外图像中行人的快速检测一直是计算机视觉领域的热点和难点。针对红外图像行人目标检测算法检测速度和检测精度难以平衡,算法模型体积较大,在中低性能设备中难以部署和实时运行的问题,提出了一种基于YOLO算法的轻量红外图像行人检测方法。在分析了MobileNet v3等轻量网络在YOLO v3算法上的性能和特点之后,该方法提出了引入注意力机制的轻量特征提取网络(CSPmini a)、特征融合模块和解耦检测端分类回归结构三种改进措施,在满足网络模型轻量的情况下保证了一定的检测精度。实验表明,该方法有效的实现了红外图像行人目标检测的准确性和快速性。

An infrared image pedestrian detection method based on YOLO algorithm

Fast detection of pedestrian in infrared image has always been a hot and difficult problem in computer vision field.Aiming at the problems of difficulty in balancing detectionspeed and detection accuracy of infrared image pedestrian target detection algorithms,the algorithm model is large,and the difficulty in deploying and running in real time in low and medium performance devices,a lightweight pedestrian detection method in infrared image based on YOLO algorithm is proposed.After analyzing the performance and characteristics of MobileNet v3 and other lightweight networks based on YOLO v3 algorithm,the method proposes three improvements,a lightweight feature extraction network(CSPmini a)that introduces an attention mechanism,a feature fusion module and a classification and regression structure of decoupling detection end.Under the condition that the network model is lightweight,certain detection accuracy is guaranteed.The experiments show that the method is effective in achieving the accuracy and speed of pedestrian targets in infrared images.