Font Size: a A A

The Research Of Target Detection Method Based On Moving Camera

Posted on:2015-05-09Degree:MasterType:Thesis
Country:ChinaCandidate:J Y YangFull Text:PDF
GTID:2298330467455090Subject:Pattern Recognition and Intelligent Systems
Abstract/Summary:PDF Full Text Request
The algorithm of moving target detection in dynamic scene is a hot direction inimage processing and image understanding fields. It is also the future developmentdirection of video processing, and it has attached great attention of the experts in thecurrent image field and the artificial intelligence field. In this paper, the robustestimation on corner points to extract those corner points only belongs to moving targetin the base of the corner points detection and tracking. And combined with thecoordinate information of goals to extract the target region from the initial segmentationimage. The realization to detect moving target under dynamic scene finally. The mainwork contents are as follows:This paper studied of the random sample consensus algorithm combined with rankconstraint theory, and introduced the least median square method based on rankconstraint to realize the estimation and classification to the image corner points.Experimental results show that these two methods have good stability, high precisionand high breakdown rate, and they can extract the moving target feature points detected.At the same time, it is introduced a classification algorithm based on minimumcurvature radius in this paper. Through the observation to trajectory curves of image’scorner points, it can be seen obviously that distorted degree of background track curvesis different from the prospect track curves’, when the relative motion between thecamera and the moving target is a large quantity. Accordingly, the image’s all cornerpoints are divided into background points and foreground points by setting anappropriate minimum radius curvature threshold to distinguish background track curvesand the prospect trajectory curves, and it is in order to complete subsequent the targetextraction. Experimental results show that this method can extract the target featurepoints effectively, while the relative movement is a large amount between thebackground motion and object motion. And it is simple, low complexity, high stabilitycompared with the above two kinds of method. In this paper, it is introduced a method that combines with coordinate informationof object points to detect the moving target by extracting the regions contain the targetonly. It segmented image by Isoperimetric algorithm firstly, and then used the targetpoints coordinates information to extract object region from the segmentation image.And connect boundary, expand and corrosion on the object region extracted and otheroperations to enrich the target area in order to obtain the complete target. Experimentalresults show that this method can be used to detect moving object. And it is proposed animproved approach basis on the above method to improve the speed of Isoperimetricsegmentation, which cut out a sub image contains the complete target from the originalimage, and then do the Isoperimetric segmentation to the sub image. Experimentalresults show that the improved method reduces the dimension of the coefficient matrixin linear equation, speed up the solution of linear equation, improve the Isoperimetricsegmentation efficiency.
Keywords/Search Tags:Target detection, Isoperimetric segmentation, The dynamic background, Rank constraints, Radius of curvature
PDF Full Text Request
Related items