Research And Implementation Of Camera Track’s Self-adaptive Algorithm Based On Image Matching

With the continuous development of computer application technology, research on computer vision technology and theoretical is constantly being improved, especially in the visual inspection, visual navigation and automated assembly in the field that is being widely used.. Typically, vision sensor (camera) as acquisition, processing the image data of the tool is due to its advantages of low power consumption, large observation area and access to information-rich. But in the face of some harsh environment which is not suitable for people to operation, the scope of the camera will be limited, such as forests, deserts, battlefield environment. In order to expand its scope and realize their autonomy and automation, the image resources obtained by camera can be used to make the camera return is the problem to be solved. The first condition of self-realization returned is to restore the movement route of the camera.When the camera is moving in space, using only the image information acquired by the camera restore the camera’s movement routes without other case of the navigation apparatus of the navigation information is the main task of this paper. This paper is focused on using monocular camera of moving to obtain a sequence of images, with the environmental characteristics are identified and matched by vision systems, according to the global motion which is caused by camera’s motion, the relative motion information of camera should be accessed to recover the camera motion route.This article mainly studies the relationship between the camera’s motion and the point’s motion, taking advantage of each pixel in the image movement to calculate the global motion parameters, and then recover the camera motion trajectory. First, in order to achieve the balance of efficiency and accuracy, chose six affine transformation model to describe the global motion parameters. Second, to select SIFT algorithm which the number of feature points’s num is large and stable as the image matching method, and then to remove false matching points based on neighborhood weighted average algorithm, after that use the RANSAC(Random Sample Consensus) algorithm to get rid of foreign points, then estimate the motion parameters by use of the point pairs. Third, in order to solve the problem of low accuracy and large complexity in the process of image matching, the adaptive sampling and adaptive distance threshold image matching strategy is proposed, using the adaptive threshold value method to improve the matching accuracy and using the adaptive sampling method to reduce the number of feature points, and then reduce the time consuming. The test results show that the improved matching algorithm can better describe the movement, it can improve the parameter precision meanwhile reduce the operation time, and it is more suitable for real-time system.