A SLAM Algorithm Based On Enhanced Visual And Radio Feature

Feature registration is the first problem to be solved in SLAM(Simultaneous Localization and Mapping)system,in particular,the feature extraction of feature sparse,high noise environment and loop detection in high cumulative error are the key to feature registration.In this paper,a weighted gradient color enhancement operator(WGE-ORB)based on ORB(Oriented Fast and Rotated BRIEF)features is proposed for the gastrointestinal environment with a sparse visual feature point,which can be used to describe the surface texture and superficial blood vessel contour of different tissues.The WGE-ORB operator is a color enhancement model based on the gradient similarity correlation measure,and the normalized weighted color channel factor is introduced to enhance the color contrast.The augmented Lagrange algorithm is used to optimize the weighting factor of each channel in the gray image so as to get the gray image with enhanced contrast.Then the enhanced gray scale image is extracted and matched with the binary operator ORB.Compared with three mainstream enhancement algorithms,HE,Retinex and HSV,the correct number of WGE-ORB feature point extraction is the largest and the false matching rate is the lowest.WGE-ORB,ORB and SURF(Speeded Up Robust Features)operators are evaluated from three aspects: the extraction time,the root mean square error under rotation and contraction.WGE-ORB has the best performance in image rotation and retraction,and its computation time is equivalent to the ORB operator,which is superior to other mainstream operators to meet the needs of online SLAM.The results show that the operator can improve the robustness of feature extraction and registration under the feature sparse environment by enhancing the contrast of color image.Aiming at the problem of map drift with high cumulative error,a multi-path fingerprint signal subspace location algorithm based on fusion of enhanced visual features and radio fingerprint features is proposed.When the visual loop appears,the radio fingerprint signal which with the help of key frame timestamp is used to assist the visual location.The simulation experiment determines three factors that affect the positioning accuracy of the multi-path fingerprint signal and determines the best parameters.The low dimensional projection Glassman discriminant analysis can keep the error accumulative probability in the range of 1 meters high to 93.4%.Based on the assumption that the robot path is random,the wireless multipath signal model is used to simulate the online fingerprint location and location test of indoor scenes.The experimental results show that the robot's random trajectory will not reduce the positioning accuracy based on Glassman's Radio fingerprint location algorithm.