The Research On Feature-based Method For Camera Self-calibration

The paper is focused on the research on feature-based camera self-calibration, including interest point detection, feature correspondence among sequential images and camera self-calibration. With the in-depth development of photoelectric technologies and digital sequential image processing and analysis, more and more attention has being paid to the research area of structure from motion. Moreover, many navigation and control tasks requiring the motion parameters of the camera and Euclidean information of the scene, such as object tracking, obstacle avoidance and map building and so on, also require the fact that the intrinsic parameters are known. Therefore, all these applications have made the calibration of the camera a research hotspot in computer vision. Based on the analysis of existed methods, an integral camera self-calibration system is designed, and some improvements on better understanding about related algorithm are expected to make. The main work in this thesis includes:On the research of interest point detection, different kinds of existed methods for interest point detection are first categorized and summarized, followed by a detailed evaluation of several typical methods. To solve the problem on setting the threshold on the interest point, a novel approach to interest point detection based on phase congruency is proposed. Experimental results show that the accuracy of detection and localization of different kinds of interest points can be achieved.On the research of feature correspondence among sequential images, we summarize different kinds of stereo matching algorithms and discuss the main difficulties associated with the problem. Some theories on fundamental matrix, together with several estimating method, are introduced. With uniqueness and epipolar constraints, a robust method is utilized to solve the correspondence problem and estimate the fundamental matrix simultaneously. Experiments with real image sequences validate the effectiveness of the approach.On the research of camera self-calibration, we introduce different camera self-calibration methods under various circumstances as constant intrinsic parameters with general motions and restricted motions and variable intrinsic parameters under general motions. To evaluate the performance of different algorithms, a prototype system of camera self-calibration is proposed and implemented. For each key algorithm of the system, detailed descriptions and analysis of the experimental results are made extensively, together with their advantages and disadvantages pointed out. Finally, several discussions and future research areas and applications of camera self-calibration are also presented.