Research Of Multi-view Image Matching Technology For Point Clouds Acquisition

Image matching is one of the basic tasks of image processing. It is used to match the images which were obtained from different conditions of the same objects or scene. Since image matching is the precondition of applications of image mosaic, image reconstruction and object recognition, many researchers have studied on this problem and proposed a lot of matching algorithms.An improved bidirectional SIFT feature matching algorithm is proposed for the high mismatching rate and low operation efficiency of SIFT algorithm. Based on Image-Matching-Technology with SIFT algorithm as the research object, we analyzed the principle, implementation process, and the main advantages and disadvantages in detail, and also improved the defect. Finally, the main results are as follows:(1)To remove the mismatching, there are two steps to follow. First, SIFT bidirectional matching algorithm is taken to eliminate part of the mismatch. Second, disparity gradient constraint and RANSAC algorithm are used to purify the matching points. On the basis of traditional SIFT algorithm in this paper, we processed two matching in positive and negative for SIFT feature vector. It will improve the accuracy of the matching, and is important to realize automatic calibration of space posture and position using matching points. In order to increase the number of matching points fundamentally and ensure the accuracy of matching, this paper also puts forward the method of using intersection and sets for mismatching removing, and selects sets as the matching points. Based on the intersection with high accuracy the fundamental matrix was estimated and then used for removing the mismatching points in the sets.(2)To improve the speed of process, it is not only K nearest method was taken at the beginning of matching, but disparity gradient constraint is also improved, in order to reduce the iterations to lower the consuming time. Traditional disparity gradient constraint algorithm can eliminate about 20% of the match probably, and each of the iterations can only remove a match point by mistake, what is more, when the number of iterations are too much, the running speed of the program can be seriously influenced. In this paper, the improved disparity gradient constraint algorithm was adopted, in which the third step of "remove the biggest disparity gradient and the corresponding point on" was changed into "get rid of all of the disparity gradient and 3 times greater than the minimum disparity gradient and the corresponding point on". After that each of the iterations can remove more than one false match point, and the number of iterations is reduced, the time of the algorithm is reduced too.Experimental results show that the algorithm presented in this paper not only keeps the basic characteristics of SIFT algorithm, but also has many advantages such as plenty matching points, high matching accuracy, no duplicate points, matching efficiency higher merit and so on. In a certain aspect of the matching points’ number, the improved algorithm make false matching points decrease 4%-7% through a series of purification to automatically calibrate the position and orientation of camera in the space; in a certain aspect of the algorithm run time, the improved algorithm make run time decrease 5%-10%, the algorithm of image matching results and real-time are improved, the new algorithm also improves image matching results effect and real-time ability.