Image Fusion Traffic Detection Method Based on Improved YOLOv5
Aiming at the problem of low recognition accuracy of YOLOv5s model and large number of network model pa-rameters in complex traffic environment ,an algorithm for pedestrian and vehicle target detection based on feature fusion of infrared and visible images is proposed. Based on the YOLOv5s algorithm ,firstly ,a progressive image fusion network is used to generate visible and infrared image datasets;secondly ,GSConv convolution is used to replace the original convolu-tion in the feature fusion part,which reduces the number of model parameters and computational volume,and the CA position attention mechanism is introduced so that the network pays more attention to the positional information;and the EIOU-Loss loss function is used to replace the original loss function to accelerate the convergence speed and improve the regression accuracy. Finally ,the target detection and recognition experiments for pedestrians and vehicles are carried out on the M3FD dataset. The experimental results show that the improved YOLOv5s improves the mAP by 11.6％,the model size by 4.20％, the number of parameters by 8.05％,and the detection speed by 9.09％ compared with the original network for traffic de-tection under infrared conditions with complex background.
infrared and visible light imagestraffic target detectionimage fusionattention mechanism
万方数据
维普
