A simple and parallel algorithm for robot position estimation by stereo visual-inertial sensor fusion

The stereo visual-inertial sensor fusion is very popular in the near decade for the robot position estimation, because it is effective in the environments without the global position information. Fusing the stereo vision and the AHRS sensor (Attitude and Heading Reference System, three axial gyroscopes, accelerometers and magnetometers), and using the perspective projection property of the camera, we propose a new adaptive algorithm to estimate the robot position. Through the novel real-time method, the robot can be positioned in the environments without the global position information, and we can theoretically prove that the estimated positions converge to their true values. The new algorithm can be simply implement and parallel processed by GPU (Graphics Processing Unit), to compute the real-time positions. The performance of the real-time position estimation algorithm is validated by an experiment.