Research On Depth Estimation Algorithm Based On Light Field Multi-view Image Sequence

The depth estimation of scene is one of the key problems in computer vision,but the traditional single aperture imaging system can only record the intensity information of the scene in the corresponding position of the detector plane,and lose the direction information of the light.Therefore,it can not perceive the depth information of the scene.For this reason,we use the camera array to get the complete four-dimensional light field information,and get the depth information of the scene according to the disparity relation among different viewpoints.In view of the large error of the stereo matching algorithm commonly used in multi view image processing,this paper introduces the processing method of epipolar plane image(EPI)to obtain the disparity.Specifically,according to the formation principle and structural characteristics of EPI,the relationship between disparity and depth can be obtained.In most cases,people first perceive color,which is an important information that distinguishes objects,and can be used as the main distinguishing feature.Therefore,this paper first studies the depth estimation of the color image sequence,and make full use of the difference of the color information to get the edge information of the image,and then use the consistency of the color information to get the complete depth image.By experimentation on the more authoritative data set in the industry,the results are compared with the results of the more perfect methods in the industry,which shows that the algorithm has a good effect on the edge preserving and the time complexity is significantly reduced.However,in some industrial background,it is impossible to get color image information of the scene,so this paper also studies the depth estimation of the gray image sequence.Based on the relatively mature edge detection algorithm of gray image,an adaptive threshold Canny edge detection algorithm withthe largest inter class variance is proposed,which combines the least square method to get the slope to get the depth information of the edge.Experiments on the grayscale dataset collected in our laboratory show that the proposed algorithm can estimate the depth information of the scene more accurately.