Localization Algorithm of Mobile Robot Based on Colored ORB Feature and Parallax Angle Parameterization

Abstract

The main contribution of this paper is to improve the existing deficiencies of the current feature-based robot localization algorithms. Since most feature-based robot localization algorithms use only grayscale images for features extraction and matching, which ignore the rich color information in the image and result in a low correct matching rate of features. In view of this problem, this paper proposes a colored ORB feature extraction algorithm, which can effectively improve the matching rate of features. In addition, the method of parameterizing features using XYZ coordinates in European space may lead to the ill-conditioned of optimization algorithms. In this paper, the method of parameterizing features based on parallax angle can effectively avoid the ill-condition of the objective function. Even when the feature is far away from the camera of robot, the convergence speed of the optimization algorithm is quite fast. It is shown in the simulation experiment that compared with the traditional localization algorithm based on ORB feature, the matching efficiency of the features of this algorithm is improved by about 12.5%. The average absolute trajectory error of the robot localization is 0.01034m, which can meet certain practical needs.