Study On The Strategy And Improvement Method Of Multi-view Matching Technology

Using multi-view matching technology to precisely, automatically and rapidly processstereo images, and then to obtain highly dense three-dimensional(3D) geospatial information, iscrucial to the effective application of high resolution remote sensing imagery. Focusing on suchaspects as model performance, reliability, precision, speed and so on, this paper discusses andstudies the strategies and improvement methods of multi-view matching technology, and themain contents and innovations of the paper are listed as follows.1. Based on a deep analysis of the difficulties and developing trends in image matching,improvement aspects of multi-view matching technology is discussed, and correspondingmulti-view matching models, methods, and constraints are then summarized and compared.2. Since the matching information collected by multi-view matching line models isdifferent, a method to integrate various matching lines is proposed. Through such methods asmulti-view matching and candidate checking, various matching information can be integratedby the method to improve the successful rate of image matching. And thus, the informationintegration quality of multi-view matching models is improved.3. The necessity and feasibility of stereo selecting process is analyzed. And based on theanalysis, a feature-point-guided stereo selecting method is proposed. According to method,matching measure significance of extracted feature points is analyzed, and the matching qualityof searching images is obtained and sorted. And then, through a step calculation method ofmulti-view matching measures, searching images with the best quality and least amount isselected for multi-view matching to decrease the negative influence caused by low qualitysearching images. Applying stereo selecting method, the image filtering and identifying qualityof multi-view matching models is improved.4. For ground surfaces along stereo epipolar planes, a local slope calculation method isproposed to rapidly identify discontinuity features. The method provides technical support forthe self-choosing of multi-view matching models and the self-extension of matching window,and therefore, greatly improves the matching reliability of discontinuity features.5. A multi-view matching strategy plan is designed and constructed to orderly organizeand effectively combine auxiliary data, matching models and constraints used in multi-viewmatching process. And correspondingly, a reliability evaluation model is proposed to evaluate multi-view matching results. The model uses eight reliability evaluation factors and stepanalysis method to verify the reliability of matching results, and then determine whether theapplication process of matching strategies should be continued or stopped. So, it can greatlydecrease redundant calculation brought by excessive application of matching strategies, andsignificantly improve the reliability of matching results.6. The geometrical constraint of epipolar lines is introduced into multi-view least squarematching process, and through weight transformation and quality controlling methods, theprecision and convergence speed of multi-view least square matching is greatly improved, andthe precision of matching results is refined to sub-pixel level.7. Potential dense computing tasks existing in multi-view matching process are analyzedand extracted, and based on the analysis, a coarse granularity CPU multi-core parallelcomputing method and a coarse granularity CPU parallel computing method is proposed. Basedon two given parallel computing platforms, the feasibility and practicality of the two methods isverified by a few experiments, and some accelerating efficiency conclusions are obtained. Theexperimental results demonstrate that the speed of dense computing tasks can be significantlyenhanced.