Research On Wide Baseline Stereo Matching Algorithm Based On Topological Invariance

Wide-baseline stereo matching technology possesses higher precision for visual measurement, but also small limitations and constraints and broad prospect. Currently, it is a hot research field in computer stereo vision technology. Many fields depend on the matching precise of the real-time information to ensure the effectiveness of the system, such as aerial mapping, resource analysis, optical and radar tracking, vehicle cruise system, missiles and terrain contour matching, as well as medical diagnosis, weather forecasting, virtual reality etc. Due to time-consuming and laborious, the traditional manual matching methods can not achieve the high accuracy and no longer meet the needs of modern technology.This paper focuses on wide-baseline matching algorithms. For different graphic characteristics of images, the paper proposes two novel algorithms of wide-baseline matching: based on consistency of topological relations point matching algorithm and based on surrounding information of line features and combines topological constraints line matching algorithm. These two algorithms are provided with stability to illumination changes, geometric distortion, as well as other factors. Specific studies are as follows:(1)Based on topology, proposes a wide-baseline stereo vision matching algorithm. It not only sets up a more advanced model on matching, but discards mismatches by iteratively computing the fitness of the feature matches. The model is the combination of double topological configuration includes the consistency of first topological relationship (CFTR) and the consistency of second topological relationship (CSTR).(2)This paper presents a method for wide-baseline stereo vision matching of line segments. Color histogram based matching is performed in order to get initial candidate matches. The choice of the HSV quantization space is crucial to simplify the complexity of the calculation. In addition, presents a new scheme for estimating the epipolar geometry from segment matches.(3)The experiments are done for matching algorithms including the detection of feature points and line segments, features matching and elimination of mismatches. In the end, carry on testing performance of the algorithms and error analysis in order to confirm that the two algorithms proposed in this paper have high anti-interference, low false matching rate, and strong real-time. They can reach the matching standard.