Research On Traffic Surveillance Video Enhancement Methods Under Complex Environment

In recent years,with the rapid development of Intelligent Traffic(IT)technology,the processing methods of traffic monitoring video have become more and more mature,bringing great convenience to human beings.However,how to make the traditional traffic monitoring video processing method effective in the complex natural environment is still a hot topic in the field of traffic video.The paper focuses on the intensive research on the enhanced algorithm of complex environment in traffic surveillance video.Firstly,for the traditional traffic monitoring video processing method,the accuracy degradation,vehicle false detection and algorithm failure will occur in the case of strong shadow interference.This paper proposes a shadow removal method without manual interaction and adaptive ability.First,after the video frame is converted from the RGB color space to the HSV color space,the non-downsampled shearlet transform is performed.Secondly,the transform coefficient is assumed to be Gaussian,and the weighted mask of each scale is calculated by using the mean and standard deviation of the transform coefficients.Then,according to the zero-tree distribution characteristic of the multi-scale transform coefficients,the coarse-scale weighted mask is used to correct the fine-scale weighted mask,and the weighting masks of the respective scales and the respective color channels are linearly combined to obtain a common mask;Again,the adaptive segmentation threshold is calculated by fitting the maximum entropy method based on the least squares method,and the common mask is binarized;Finally,the moving vehicle region is determined by voting.The experimental results show that the algorithm can effectively remove the static/motion shadows in the traffic surveillance video and suppress the shadow interference.The average Euclidean distance between the output vehicle trajectory and the real trajectory of the traditional Meanshift algorithm is reduced by 95%,and no target loss occurs.Phenomenon enhances the robustness of intelligent analysis algorithms.Second,in the traffic monitoring video,in addition to strong shadow interference will make the video degraded,the rainy weather has also had a great impact on the post-processing of traffic monitoring video.Based on this,this paper proposes a traffic monitoring video raining method with fast processing speed,ability to discriminate rainfall,and spatial domain-frequency domain combination.First,the video frame is converted from the RGB color space to the YUV color space.Secondly,the non-downsampled shearlet transform is performed on the video frame and the low frequency sub-band is cleared,thereby obtaining an image rich in edge and contour information,and then the maximum is utilized.The inter-class variance method obtains all the edge information maps;then,the saliency mapping method is used to calculate the depth map of the video frame,and the bilateral filtering and non-downsampling shearlet transform are performed,and the high-frequency transform coefficients are retained to obtain the main edge information map;The edge information map,the main edge information map and the frame difference of two consecutive frames make the raindrop/rain line area and the rainfall decision.Finally,the precipitation is counted.If it is medium or heavy rain,it is directly repaired by the curvature-driven diffusion method.The pixels in the raindrop/rainline area,otherwise a rainwater test is performed again,and the traffic monitoring video after the rain is finally obtained.The experimental results show that the proposed algorithm can not only remove the raindrops/rain lines in the video,but also better preserve the shape and texture details of the objects,which can avoid the loss of tracking targets in existing algorithms.Finally,the work of this paper is summarized and forecasted.