Improvement Of Semi-Global Matching Algorithm And Research On Dynamic Image Preprocessing For Fire Scene Feature Matching

The binocular stereo vision technology combines scene information from different perspectives,and realizes the calculation of the disparity map through the image matching algorithm,then calculates the scene depth according to the three-dimensional geometric parameters,finally restores the three-dimensional spatial model.In the actual image acquisition process,due to the influence of factors such as illumination,transmission,foreground occlusion,low texture,etc.,image matching has become a key step in stereo vision technology,which is related to the accuracy and efficiency of the three-dimensional model.At the same time,with the development of autonomous driving and other fields,there are increasingly higher requirements for the accuracy and speed of image depth information acquisition.In response to the above problems,this paper designs an improved semi-global stereo matching algorithm,which takes into account the matching accuracy while improving the matching efficiency;In addition,this article applies the stereo matching technology in the fire environment to test the accuracy and robustness of the matching algorithm in the actual industrial field environment.The main innovations of this article are as follows:(1)Improve the LBP operator in the local matching algorithm.The LBP operator has good robustness to the lighting environment and low computational complexity,but it performs poorly in weak texture areas and object edges,because there are more redundant pixels and are easily affected by noise.This paper improves the LBP operator by selecting points at intervals,constructs an initial matching cost space,and obtains a larger perception range while filtering out redundant information.(2)Improve the cost aggregation path in the SGM semi-global stereo matching algorithm.Matching cost aggregation is the soul of the SGM algorithm.The traditional matching cost aggregation has 8 paths or 16 paths and is given the same weight.The aggregation of some paths depends on the matching cost of the previous pixel or the next pixel,and requires a large amount of space to cache data.In this paper,the SGM aggregation path is improved.According to the physical access rules and a priori disparity constraint,the aggregation path is reduced to 5,and the gray level constraint and the distance constraint principle are combined to realize the adaptive weight distribution.Experiments verify that the improved dense matching algorithm in this article has an average matching accuracy of 95.94% in the Middeval2 data set,which is3.5 percentage points higher than the traditional SGM algorithm,and the calculation speed is increased by more than 50%;in the Middeval3 data set,the average matching accuracy is86.6 %,compared with the traditional SGM algorithm,the accuracy rate is increased by 5percentage points,and the calculation speed is improved.(3)Application of binocular stereo vision technology in industrial field.This article uses video frame extraction and multi-frame image difference technology to achieve the blurring of flames and the enhancement of the edges of burning objects,and reduce the influence of flame jump on the stereo matching algorithm;explain the principles of several feature point matching algorithms,and choose SURF algorithm for feature matching,and the matching algorithm is verified in the actual scene.The experimental results indicate that the combination of the preprocessing technology in this paper and the SURF algorithm can achieve an average matching accuracy of about 90% and a calculation speed of 2.5s,which can meet the project requirements of feature dynamic matching and achieve a more robust matching effect.