Study On Mobile Robot Localization Algorithm With Vision-odometry Fusion

In the past decade,human beings are starting to be replaced by mobile robots in daily life and industry to undertake some simple handling tasks.To implement autonomous tasks successfully,an accurate localization of mobile robots is always needed in practical applications.To obtain the objective,researcher are trying to equip a variety of sensors on the robotic systems,combined with algorithms for precise localization of mobi le robots.Unfortunately,most of the localization algorithms are based on single kind of sensor,while other algorithms are based on multi-sensors with non-unified coordinates.The dissertation considers the problem of non-unified coordinates of multi-sensors,and aims to design an algorithm to unify the coordinate between monocular camera and odometry.We also want to realize the localization of a mobile robot with absolute visual scale.In our algorithm,the visual system need to a visual initialization process at first;a coarse extrinsic transformation of the camera and encoder coordinate frames and the scale of visual system will be calibrated with a novel hand-eye calibration approach.Due to the measurement noise,the coarse calibration result and sc ale of visual system cannot satisfy the requirement of robotic localization,a graph-optimization considering the bundle adjustment error and odometry measurement error is constructed in the algorithm to refine the robotic poses,the feature positions in t he map and the extrinsic sensor transformation.To ensure the real-time performance,the system is decomposed into three threads for parallel computing,including tracking thread to track the mobile robot pose and match the features in the map,local mappi ng thread to optimize the extrinsic calibration,and loop-closing thread to search the visual loop-closing during robot motion.The scale of visual system will be refined once the visual closing has been found.To verify the effectiveness of the proposed approach,some typical algorithms is compared in the dissertation.RMSE is used to evaluate the accuracy and robustness of the algorithms.Furthermore,to verify the necessity and reliability of the nonlinear optimization in our algorithm for external calibration,the coarse calibration and the refine calibration procedures are verified with KITTI dataset.The experimental results show that the accuracy of the calibration is improved prominently with the proposed approach.We also use the motion capture system in our laboratory to evaluate the performance of our approach.The result shows the accuracy of our algorithm for mobile robot localization.The proposed approach are useful for the development of mobile robots in civil and industrial applications.