Position Estimation and Correction of Mobile Robot by Model-based Scene Matching and Optimization Method

This paper presents a method to estimate and correct the positions of an indoor mobile robot. Scene matching is based on geometric hashing[9, 101. An optimization method[l2] of maximizing cross-correlation of image regions is used to estimation correction. We assume that the robot is equipped with two or more video cameras, a 2 0 laser range finder and an odometry system. The 3 0 CAD model of indoor environment is known. First, the indoor environments are remodeled in a form convenient for scene matching. Some selected features, which are the model edge positions on the horizontal plane of the 2 0 laser range finder; are transformed into geometric invariants by system calibration and range data fusion. These geometric invariants of features are modeled off-line in hashing tables according to the proposed basis constraints[l3]. They are then used as indices to make scene matching by geometric hashing with weighted voting rule. After scene matching, the self-localization of mobile robot can be completed by coordinate transformation and a least square method. Furthermore, we use a two-step optimized descent method in a number of search directions to make estimation correction. Finally, the next position estimation can be made more accurate and eficient by combining uncertainty analysis[l] with geometric hashing.