Pose Estimation Based On Visual Sensors For Mobile Robots

The mobile robot received wide attention, because of its good flexibility. The mobile robot often needs to present its position and orientation in order to determine its behaviors or movements. The fast and effective pose-estimation that determines the robot's position and direction in the work environment are key steps to the completion of its tasks. Therefore, based on the robot-configured sensor, the online pose estimation is important to achieving this flexibility and acclaim.This paper presents a method of pose estimation based on visual sensors for mobile robots. Experiments showed that the precision and speed of the method is suitable for mobile robots. At the same time, this method overcame the defect of high cost and poor precision of other traditional methods based on inertial sensors. In this paper, we adopted SIFT algorithm and implemented image feature extraction and matching. After the moving of the robot, we searched for a matching point between the image sequences, and the feature extracted had the properties of scale-invariants and rotation-invariants. In addition, we used Zhang Zhengyou's calibration method to solve the internal and external parameters. These parameters express the relationship between image coordinates and world coordinates of the feature point. By this method, the variation of the feature point's image coordinate caused by robot moving can be used to calculate the variation of the world coordinate. The quaternion algorithm is introduced to solve the matrix of rotation and translation, and thus pose estimation is strong and robust. Finally, the simulation process of comparing the real mobile robot trajectory and the estimated value of the pose verifies the feasibility of the method and overall accuracy.The pose estimation method this paper proposes has the advantage of using only the robot's vision sensors to complete the robot pose measuring, which is significant for improving the operational capabilities of mobile robots.