Study Of Feature Matching Algorithm Based On Color Invariant

Object recognition technology has been used in many applications, especially invisual SLAM, the object recognition technology plays an very important role. It isnecessary for robot to recognize objects efficiently with a perfect algorithm, at thesame time, for the real-time and flexibility SLAM system. The most common way tosolve the object recognition problem is local feature matching, which recognizesobjects by matching the target object features with the reference object.BRISK algorithm is a relatively new local feature matching algorithm whichperforms key-points detection in the scale space, the descriptors of key-points andHamming distance matches, it is suitable for applications of real-time and mobileplatforms. However, the algorithm is based on gray-scale, ignoring the colorinformation of the local area with similar characteristics of the color images whichmay causes false matches and dramatic changes in the lighting conditions of poormatching results. Therefore, the main goal of this paper is that increasing therobustness BRISK for illumination without destroy other performance of the originalalgorithm.Firstly, in order to solve the problem of BRISK operator ignoring the colorinformation, this paper proposes a colored BRISK algorithm based on invariants.Replacing the input image with color invariants and constructing scale space on thecolor invariant plane, then detecting key-points on the scale space, the use ofsampling point size to describe the relationship between RGB color values and colorinformation from the original local descriptors cascade BRISK then get color-BRISKdescriptors, and the Hamming distance completed by matching the improvedalgorithm for illumination changes have robust.Secondly, in order to solve the problem of the improved algorithm in the overallconsumption of time is not ideal and the key-points detection costs much time, thispaper propose a robust Haar features corner detection algorithm, by determining thecorner methods to improve BRISK candidate regions with corner detection algorithm detects the accuracy of FAST real corner. The improved FAST operator effectivelyrejects the influence of texture corners and noise, improves the efficiency of cornerdetection, thereby further reducing the overall time cost of the proposed algorithm.Finally, this paper describes the performance evaluation criteria of local featurematching algorithm, and performs a comprehensive comparative analysis ofexperiments with the algorithm proposed in this paper, the original BRISK and theC-SURF algorithm in the real scene images, the results demonstrate the effectivenessof the proposed algorithm.