Optimization method of road target detection based on infrared and visible light image fusion

In order to improve the accuracy of road target detection in the field of road traffic safety, the idea of multi-scale feature image fusion in image fusion technology is used to achieve fusion and the idea of Ghost bottleneck module of GPNet is used to reduce the complexity of the algorithm. An innovative infrared and visible light fusion and target detection network is established. The network consists of three parts: selective image fusion module, lightweight target detection module and fusion quality and detection accuracy discrimination network. Three sets of tests were conducted in daytime, nighttime and special weather (rain, fog, etc.) under 30-40\ km/hurban conditions with average vehicle speed as data sets. The experimental results showed that the average gradient was increased by 5.64881, the cross entropy was 0.93668, the edge strength was 56.9457, the information entropy was 0.925208781, the mutual information was 1.000548571, the peak signal-to-noise ratio was 3.053893252, Qab was 0.342882208, Qcb was 0.20898381, and the mean square error was reduced by 0.08. The AP, mAP and Recall output by the lightweight object detection network were all at the optimal level, verifying the advantages of the innovative application of infrared and visible light technology in road obstacle detection.