Research On Obstacle Detection In Rain Scene Based On Stereo Vision

With the rapid development of artificial intelligence,computer and other high-tech,the automobile industry is undergoing a major transformation,among which the development of intelligent vehicles has been an important part of the development of the automobile industry.Environmental awareness is the basis of intelligent vehicles.Stereo vision sensors have been paid more and more attention in the industry because of their low cost,strong adaptability and rich information collection.Intelligent vehicles are seriously interfered by raindrops when driving in rainy days,so this paper focuses on the road scene in rainy days,using the Attentive GAN to carry out in-depth rain removal operation on the raindrop pollution image collected by the camera,and then carry out obstacle detection.First of all,after comparing the two methods with better rain removal effect,this paper uses the Attentive GAN to remove the rain from the images collected by the camera in rainy days.Different from the general Generative Adversarial Networks,the network adds the attention map to the generation network and the discrimination network respectively.By injecting attention information,the generating network and the discriminating network will pay more attention to the raindrop polluted area and its surrounding structure,which can effectively improve the training speed and the quality of the generated image.Secondly,the vehicle stereo vision system is constructed.Based on the stereo vision imaging model,the internal and external parameters of the camera are calculated as the basis of 3D reconstruction.Through stereo matching,the disparity map corresponding to the stereo image is obtained,and then the 3D data describing the road scene is obtained,which is the basis of the subsequent obstacle detection.Then,the disparity map is processed,and the U-disparity map and V-disparity map are obtained respectively.The obstacle detection algorithm based on U / V disparity is studied,and the road information in the disparity map is eliminated by Hough transform,so as to realize the obstacle detection.Finally,carry out the vehicle road experiment,train the rain model,use the stereo vision system to collect the rain scene image,carry out the obstacle detection after the rain removal operation,compare the detection results before and after the rain removal,and then verify the effectiveness of the method proposed in this paper.