Research On Image Feature Extracting And Matching In Mobile Robot Vision

In order to complete the robot localization and navigation in the filed of mobile robot vision navigation,this' s applied to the 3D reconstruction from 2D image information which was obtained from the robot vision system.Matching was most important process in the whole technology.In this study,we made different research program and process according to the different texture information of the environment.Firstly,we analyzed the high texture image,and completed matching between them.But,the premise of matching was that images had high-gradient characteristic.However,for those images that included low-texture regions(such as flat desktop),it's difficult to obtaining a high-quality result of 3D reconstruction using above methods due to the low-gradient image pixels.In this study we used image segmentation technology to obtain superpixels in low-texture regions of the image,then we used iterative thought and mathematical statistics method in superpixels matching.Through iterative searching to filter matching object whether was satisfied with conditions.The research results can be applied to the robot vision navigation system and further to complete 3D reconstruction.Experiment result showed that,using superpixels matching in low-texture image can improve the accuracy of 3D reconstruction.In order to complete the 3D reconstruction of the environment,we needed to calibrate the camera to get the camera parameters firstly.This parameter was the bridge between 2D image coordinates and 3D image coordinates.After we completed high texture matching,then completed low texture matching.In the end,completed 3D reconstruction.Experimental stage,we compared the effects of reconstruction by chossing different environments,obtaining advantages and disadvantages of various matching methods.In the experiment process,research program also showed its shortcoming,and that's shortcoming is also the direction in future research.After reading a lot of references,we analysed the bottleneck.Based on the analysis of the traditional image feature matching method,so we had mastered the theory used in the method,and completed the experiment process step by step.Finally,we selected the different texture images and used this method to complete the matching experiment,so we found the defects in the texture image matching used this method.On this basis,we summarized a lots of previous experiences and achievements.We used the latest research results to complete the research objectives.we used the C++ language,OpenCV,PCL and other libraries to complete the experiment.Through analysing and comparing the experiment results,the method which used in this paper can solve low texture matching in the process of 3D reconstruction and visual navigation.