The Study Of Computer Vision Techniques Applied Into Low Penetration And Precision Strike

Based on the requirements of practical applications, the dissertation anatomizes StereoVision and Image Registration techniques in Computer Vision field and put the emphasis on theessential and most difficult part involved in both of them, i.e. the correspondence problem. Thedissertation proposes two novel approaches to solve the problem so as to technically find howto apply computer vision techniques into low penetration and precision strike. During the Persian Gulf War in 1990s and the US-Iraq War in 2003, television viewersaround the world have witnessed the new effectiveness of modern high-tech weapons. It showsthat modern national defense is in urgent need of new science and high technology. At present,low penetration and precision strike are the two practical tactics developed in modern warfarethat can effectively fulfill an air attack mission. With the growing of air defense power, it ismore necessary to exert all the power of new technologies on air strike power so as to improvetheir effectiveness as possible, meanwhile assuring the safety of attackers. Computer Vision is a new technique developed rapidly in recent years, which has beenapplied in a variety of domains and is gradually accepted by more and more researchers.Generally speaking, it is the technique to process visual information taking advantage ofcomputers and can thus be regarded as a remote sensing technique, with which engineeringsystems can improve their automation, intelligence, and other performance. In fact, it is not anew idea to apply computer vision techniques into low penetration and precision strike, butthere exist many technical problems to solve. The dissertation intends to study two of them, oneis Stereo Vision, and the other is Image Registration. In both of them, there is the commoncrucial part, i.e. the correspondence problem, which is also the most difficult task to solve. Inthe dissertation, it is cast as an engineering optimization problem, but has to be solved in twoways according to different application requirements. For example, a detailed digital terrainmap is needed in low penetration while the position of the target point needs to be locatedautomatically in precision strike. Thus, the difficulties involved and/or the methods employedmay be various in different applications. In the case of low penetration, Stereo Vision techniqueis employed and an area-based matching method is developed in the dissertation, with which adense disparity map can be obtained directly without the need of the feature extraction and thepost-interpolation processes. In the case of precision strike, after analyzing several candidateschemes Point Mapping Based Image Registration technique is employed and a feature-basedmatching method is developed, with which images are aligned and their spatial information iscombined correctly, quickly, and automatically. Genetic Algorithms are well known as the optimization technique that can be search forglobal optimal solutions. As an optimization task, the correspondence problem can be solvedmaking use of them. In the dissertation, two novel matching methods are proposed for the twoapplications, respectively. One is used to estimate a dense disparity map and the other is used tofind the matches quickly and correctly. In their implementations, for the first case, a feasibledisparity map is viewed as an individual and the disparity value at each of its pixel sites isviewed as a chromosome, which is represented by a binary string. Therefore, one individualcontains a lot of chromosomes, which is very different from the classical GAs. Accordingly,two types of guided crossover and one type of guided mutation are designed for the encoding,and four matching constraints are formulated into the evaluation function. Hence, the proposedalgorithm can obtain a dense disparity map with efficiency. For the second case, first a newchromosomal encoding scheme is proposed, so called the partial permutation, then for thoseencoding chromosomes five matching constraints are considered, five seclection methods andthree replacem...