Study On Dense Matching Method For Close-Range Stereo Image

In recent years,3D reconstruction based on image has become a research focus of photogrammetry, whose core aspect is the image matching. Matching based on close-range images has many shortcomings, including a high rate of repeatability texture and discontinuous parallax, the front overlaying the background and other negative characteristics. All above characteristics make the dense matching based on close-range images become a difficult task. So far, main mature methods are sparse matching and quasi-dense matching which are both base on feature points matching. However, these mature methods can’t acquire dense point clouds data. Therefore, this paper sets dense matching method for close-range stereo image as its key research content and makes some improvements based on the analysis and studies on the methods of image matching. And finally an experimental system of dense matching for close-range images is established.The core of digital close range photogrammetry is image matching. In order to obtain better spatial data, the paper analyzes the current space data acquisition and stereo image matching techniques and put forward the stereo image dense matching method. To obtain dense matching data, this article improves the matching efficiency and effectiveness in the aspects of matching similarity measure model, epipolar constraint, parallax constraint and image color analysis. At last, the digital close-range image dense matching system is constructed and the effectiveness of the method is verified in example. The main conclusions and results are as follows:(1) Based on analyzing and summarizing the existing image feature extraction algorithms, and combining the close-range image features, just like the shorter photographic baseline, complex texture and occlusions, the paper selects the SURF algorithm to extract the feature points and search the matching points, which have rotate and scale invariant features. Then use these robust feature points to calculate epipolar line parameters, which can reduce the error brings by manually selection in traditional methods.(2) In stereo matching, it’s important to judge whether the two points separately located in target image and reference image are homonymy points. Therefore, after comparing and analysing current match measure models, this paper make most use of the color information and pixels’ spatial distance information, introducing the thought of weight to build a comprehensive match measure models on the basis of the color and the pixels’ distance weight, which lay a foundation for match measure models.(3) In order to improve the efficiency of the image matching, the search strategy during the image matching is very important. This paper introduces the commonly used epipolar line constraint strategy, which can make the two dimensional search down to a one dimensional search. At the same time, to prevent the error propagation, the paper introduces triangle constraint nets based on the matched feature points’ parallax, which can improve the efficiency of image matching.(4) In order to get more match points and overcome drawbacks of parallax discontinuous and matching holes, the paper optimizes the initial matching results using the image color analysis results. It bases on the assumption of that the pixels in the same segment have the same parallax.(5) Design the prototype system of the digital close-range image dense matching, and realize the algorithm of dense matching method, and use the stereo images collecting from the vehicle-borne3D data acquiring system to verify the effectiveness of the proposed method.