Research On Visual Navigation Technology Based On Feature Set

Visual navigation from the point of view of bionics,it is through the simulation of human vision function,through the image acquisition equipment to collect the information of surrounding environment,and use computer vision technology for image processing to obtain navigation information,so as to achieve navigation.With the improvement of navigation algorithm and computer performance,visual navigation has become a hot research field of robot navigation.This paper mainly studies the generation of offline path data,the use of feature triangles for fast image feature matching and the use of the most dense small areas for online and offline data retrieval methods.This paper mainly carried out the following work:Firstly,a method for generating offline path data of feature triangle set is proposed.In order to more conveniently realize the generation of the offline path data,the data generation GUI software interface is designed.Through this software,functions such as image import,feature point extraction,matching,filtering,attribute classification,and three-frame association were realized.Finally,the offline path was completed.In this paper,the effect of the method is tested in a real scene,and the trend of the offline path data is analyzed.Secondly,a hierarchical matching strategy based on feature triangle is proposed.Firstly,the feature points are extracted from the two matching images to form feature point set.Then the delaunay triangulation method is used to construct the feature point set with triangle network,and then the four-level matching is adopted.The strategy implements the matching of triangles in two networks.In this paper,simulation experiments and real-world experiments on KITTI are performed for the effect of the method.The experimental results show that this method improves the calculation speed compared with the traditional descriptor matching method when the accuracy is satisfied.Finally,when performing online and offline data retrieval matching,taking into account the problem of using the entire image information for slow retrieval speeds,a strategy based on the most dense area matching is proposed,and the case where the matching is not possible is processed.Only the local area information of the image is needed.Experiments show that this method is faster than the entire image matching method and meets the accuracy requirements.