Font Size: a A A

Research On Vehicle Target Extraction Algorithm

Posted on:2016-03-13Degree:MasterType:Thesis
Country:ChinaCandidate:J LiuFull Text:PDF
GTID:2308330503977441Subject:Pattern Recognition and Intelligent Systems
Abstract/Summary:PDF Full Text Request
As the main research contents of aerial image processing and translation, Target extraction has important implications for the late senior visual tasks, and is very worthy of study in the computer vision field. In this paper, based on single image and video sequences perspectively, the vehicle target extraction algorithm is studied respectively. Meanwhile, for the multi-pose problem exists, this paper later discusses four analysis of obtaining the target-spindle, and ultimately extracts the targets.In terms of processing single image, this paper studys a target extraction method based on a study of the salient features and superpixel. Firstly, it builds a response graph according to the salient features of the target vehicle, while carrying on the super pixel segmentation at the rectangular frame of the target detected. Then by analyzing each super pixel block in response to the strength of the response graph, it selects target sub-block to connect into the final target. Experimental results show that a better segmentation results can be obtained under the shadow of the object in the case of little effect.In terms of processing video sequence, this paper studys a target extraction method based on significant movement of the target. Firstly, by the use of affine transformation model to estimate the background motion parameters, it achieves a good background motion compensation. Then, after the differential operation to compensation image, it makes a brief outline of vehicle target, and extracts the target mass by a subsequent filtering process. Experimental results show that this algorithm achieves better segmentation results, with strong robustness.To realize the object extraction based on pose estimation, this paper converts it to analyze the target spindle extraction problem and then shows four methods based on comparative analysis of the PC A, RANSAC, Hough and direction of movement. The experiment finds that the fourth method is far better than the other three algorithms in terms of accuracy and stability. After the spindle is obtained, the pose angle of the target can be calculated, enabling the target pose -based extraction.
Keywords/Search Tags:Target extraction, salient features, super pixel segmentation, background motion compensation, differential operation, spindle extraction
PDF Full Text Request
Related items