Research on highway vehicle detection based on faster R-CNN and domain adaptation

In order to solve the problems of the high missing detection rate of small target vehicles, the low detection ability and the single application scene when the traditional target detection model in the actual highway scene that due to factors such as bad weather, light changes, occlusion. This paper proposes an improved domain adaptive Faster R-CNN algorithm. By adding image-level and instance-level domain classifiers and consistency loss components to solve the problem of domain offset caused by the inconsistent distribution between training samples and actual samples. And the RPN network is improved by using multi-scale training and mining difficult samples for secondary training during the training process to improve the performance of the model. The improved model can increase the gain by 4.8%. The experimental results show that the domain adaptive component is effective for the migration between different sample domains, and the performance of small-scale target detection is significantly improved. The improved method can effectively improve the accuracy and robustness of the model, and has certain generalization ability.