Passive Ranging With Monocular System Based On Image Registration Algorithms

Ranging based on monocular images is a technical difficulty in imaging detection, which has both military and civilian value. Passive ranging based images is the very process in which the depth variance of an imaged target was transformed into the distance from a camera to the target, and image registration is the key technology in depth clue extraction. Centering on the relative technical requirements of the National Natural Science Foundation of China (No.60872136), this thesis has done a study on image registration and its application in distance estimation based on monocular imaging system.An analysis on the point-based image registration and the current were performed. These work include modified coefficient of p in auto-adaptive threshold Harris corn detection in subpixel level, adjusted for layer numbers of Gaussian pyramid and SIFT descriptor in SIFT key points extractionh, and improved the descriptor of SURF. Our experiments show that the new algorithms have obvious improvement in running speed and rate of success matching.Aiming at application of these matching algorithms in monocular passive range system, the author has discussed the ranging equations under conditions of there exists displacement between two sampling instance or not. Among these equations, the quadratic equation is simpler and faster, is suitable only for exist a displacement for the observer between two sampling instance, while, the quartic equation has not this limit. It is given the experimental data.Our study shows that compare with Harris and SURF, the SIFT algorithm has the best stability and more suitable for passive ranging with monocular system. At last, a summary was given on transplantation the algorithm into a DSP chip.