Visual Enhancement And Perception Based On Generative Adversarial Networks In Complex Environment

Vision based object detection is an important part of intelligent vehicle environment perception system.However,in the process of driving,the road image collected by the sensors will be disturbed by different environments(such as different weather and light conditions),and also affected by the occlusion of the target to be measured.These problems affect the accuracy and robustness of the algorithm.Therefore,this paper focuses on these issues,combined with the generation of confrontation network method and visual perception algorithm,to study the visual enhancement and visual perception in complex environment.The main research contents are as follows:(1)Considering the requirements of the robustness of vision based target recognition,the driving situation in different weather and light conditions is analyzed.The assumption that there is a shared hidden space between the mixed image domains is proposed.The mapping relationship between different image domains is established.The multi coder weight sharing method is used to improve the generation of confrontation network model.The results show that it is more advanced than the current one Compared with other methods,the model has a good effect in the tasks of fog removal,rain removal and light enhancement.(2)In order to solve the problem that occlusion between targets leads to the decrease of accuracy of target detection algorithm,from the point of view of the proposed area,a repulsion force model is introduced to optimize the fast-rcnn.At the same time,the improved non maximum suppression algorithm is used for the post-processing of the generated box.(3)In view of the situation that the difficult samples are not enough,which leads to missed detection and false detection,this paper starts from the feature level perspective,and establishes a model of feature mask generation combining with the generation countermeasure network.The model can increase the difficult samples from the features,simulate the target occlusion in the real environment,and increase the perception ability of the algorithm to the features with strong robustness,so as to strengthen the target Detection algorithm detection performance.The results show that the method combined with feature mask has a good effect on occluded target recognition.(4)According to the image enhancement model and object detection model,the visual perception system in complex environment is established,which is verified on different scene databases,and then the detection process of visual perception system is optimized.The results show that the system has strong anti-interference ability.