Research On Multi-sensor Image Matching Based On Line Features

In the application of multi-sensor image fusion, medical image analysis and navigation system based on computer vision, multi-sensor image matching is always needed. Because of the different imaging mechanism of different image sensors, the image features are different, and the gray level of the image is very different. The traditional image matching algorithm is dependent on the gray distribution of the image. It is difficult to apply to the multi-sensor image matching.This paper is based on the background of the heterologous scene matching with complex terrain conditions, and developing the real-time and reliable multi-sensor image matching algorithm to meet the requirements of the actual application of the heterologous image matching algorithm is research content. Local feature of image which is studied deeply and used widely in computer vision field in recent years is used to deal with the difficulty of multi-sensor image matching algorithm.The main work of this paper is as follows:In the first part, the definition of image matching and the transformation models used in the work of image matching are introduced, then the image matching algorithm is classified according to the difference of image information, and some representative algorithms of these image matching algorithms are briefly described and analyzed. The main advantage of the feature based image matching algorithm is that the time complexity is low and the accuracy is high, and it is a hot research direction in the field of image matching technology. The advantage of the region based image matching algorithm is simple, but the fatal weakness of this method is the computational time is too large. The biggest advantage of image matching algorithm based on transform domain is easy to implement, but it needs to be further improved in the matching precision.The second part mainly analyzes and compares three kinds of commonly used straight line extraction method. Firstly, the method of using Hough transform to extract line is introduced, the time complexity of this algorithm is high, the line features missing rate is relatively high. Then the linear extraction algorithm based on edge detection and phase encoding is analyzed, the algorithm is based on the detection efficiency and accuracy of the image edge detection algorithm and the phase grouping algorithm is too high, and the weak feature extraction is too much. Then the LSD line extraction algorithm is analyzed and explained, the proposed algorithm can extract the linear feature of the sub pixel level in linear time. And the missing rate is relatively low; this is the most outstanding performance of several algorithms.In the third part, the analysis and experiment verification of the image matching algorithm based on line feature are described. Firstly, the extraction and selection of line segment feature and the calculation method of the construction rules and relevant data are introduced; Then introduced the similarity measure between the line segments, the search strategy of the same name, line segments to the matching rule, the line segment characteristics based on the relationship between the line segment feature matching method and the method of eliminating the false match. Finally, the method of the proposed method is verified by experiments, and the optimal threshold is selected by a large number of experiments. The real-time performance and matching performance of the proposed algorithm is verified by comparison with other methods.The fourth part mainly summarizes the innovation points of the paper, and analyzes the shortcomings of the paper.