Research On Key Technologies Of Vehicle Detection On Highway At Night Based On Intelligent Image Processing

Using camera to monitor vehicles is the basis of road traffic parameter acquisition and abnormal event detection.Compared with the daytime,the road image at night is foggy and blurry,and the noise is larger.So that the existing algorithms are still difficult to meet the requirements of vehicle detection at night.Therefore,based on the existing video surveillance equipment,it is of great theoretical significance and application value for highway traffic control to study vehicle target detection in the scene of night highway and improve the accuracy of vehicle target detection at night.Aiming at the nighttime highway scene,this paper analyzes the scene characteristics and vehicle detection difficulties of nighttime highway.And it focuses on the study of night image enhancement algorithm and improved Faster R-CNN target detection method.Finally,a set of vehicle target detection method suitable for night highway is formed.The main work and contributions of this paper are as follows:(1)Aiming at the problems of haze degradation,high noise level and inapplicability of existing image enhancement algorithms in night data set,an image enhancement method suitable for nighttime highway was proposed.First,the input image is decomposed into structure layer and texture layer using TV-L1 model.Then,in view of the problem that the atmospheric light value is globally consistent and the light source area fails when the dark channel prior defogging algorithm is used in the structure layer,the atmospheric light map is estimated by the Gaussian center surround function,and the color distortion caused by the failure of the dark channel prior theory in the light source area is compensated by combining the bright channel prior.Finally,in view of the problem that texture layer denoising is easy to lose details,a denoising method based on guided filtering is proposed.The experimental results demonstrate the effectiveness of the image enhancement method in the nighttime highway scene.(2)Aiming at the problems of vehicle ambiguity and large scale variation in nighttime vehicle target detection,a nighttime vehicle detection algorithm based on improved Fast-R-CNN was proposed.First of all,in view of the problem that the Faster R-CNN backbone network is not sensitive to the location of night vehicle features,the attention mechanism is used to improve the Res Net-50.By calculating the correlation coefficient between the internal features of the feature map,the ability to represent important features can be enhanced.Secondly,in view of the large scale changes of vehicles,the feature pyramid network is improved.A balanced feature enhancement structure is designed,which fully integrates the semantic features of different pyramid levels to avoid weakening the semantic information of non-adjacent levels.Finally,aiming at the disadvantage of the fixed threshold of NMS algorithm,soft-NMS is used to optimize the post-processing operation and reduce the number of missed vehicles.(3)Aiming at the lack of data set applicable to vehicle target detection on night expressway at present,this paper collects the surveillance video of expressway at night time and constructs the night vehicle data set on mountain expressway.On this basis,a set of vehicle detection algorithm suitable for night expressway is established by synthesizing all the improvement strategies in this paper,and tests are carried out respectively for each improvement strategy and the whole algorithm.The experimental results show that the proposed method can effectively improve the quality of nighttime expressway images,and improve the accuracy of vehicle target detection while ensuring the detection efficiency.