Research of Expressway Vehicle Detection Based on Faster R-CNN and Domain Adaptation

摘要

For the traditional object detection model affected by weather, illumination and occlusion factors, the application scene is single, high missing rate of small vehicle and poor detection results problems in the expressway scene, this paper put forward an improved domain adaptive Faster R-CNN algorithm, by adding image-level, instance-level domain classifier, consistency loss component, and improvement RPN network. Adopting multi-scale training during the training and using mining the hard samples fine tuning the train model. The improved model can get the gain by 4.8%. The experimental results show that the adaptive domain is effective to deal with the domain transfer between different domains, and the improved model can significantly raise the detection performance of small-scale object detection. It is proved that the improved method effectively enhance the accuracy and robustness of the model, and also has certain generalization ability.