Aiming at the problems of low clarity and resolution of infrared images and insufficient illumination of visible images,an improved multi-modal pedestrian detection algorithm IMV5(improved multimodal YOLOv5)based on YOLOv5 was designed by fusing the characteristics of visible and infrared images.The traditional cascade fusion method was improved,and combined with the attention mechanism,a multi-modal feature fusion module PMWM(pedestrian modal adaptive weight fusion module)was designed to fuse visible and infrared images to improve the detection effect after feature fusion.The optimized spatial pyra-mid pooling structure was added to improve the detection effect while keeping the receptive field unchanged.The target detection was performed on the feature layer to predict the probability and location of pedestrians and realize the pedestrian detection func-tion.Experimental results show that the detection effect of IMV5 algorithm on KAIST pedestrian detection dataset is significantly improved.
维普
万方数据