Localization Technology For Mobile Robot Based On Multi-information Fusion

With the rapid development of the domestic and international robot technology, study on intelligent robot has been growing inevitably. As the key technology to improve the intelligent level of robot, machine vision and image recognition have been the focus and difficulty of the study. Aiming at the problems of low accuracy and poor robustness of visual positioning, a technology of mobile robot localization based on information fusion of binocular stereo vision and odometer is introduced in this paper. After preprocessing the images obtained by the binocular camera, the feature point pairs acquired by the SIFT feature extraction and matching method are used to establish the natural landmark database, and then by using the EKF algorithm, the pose information obtained by the odometer and visual information are fused, so as to realize the real-time positioning of the mobile robot.Firstly, this paper provides an overall scheme design of the robot platform system, realized the construction of mobile robot platform, and then introduces the software and hardware platform of mobile robot respectively. The introduction of the hardware platform mainly includes the embedded development board, the motor driver / controller, the binocular vision module, etc. And the introduction of the software platform mainly includes the construction of wireless local area network, the writing of embedded network program, the acquisition of visual information, and the programming of the client control interface of the system.Secondly, the camera imaging principle and camera calibration method are introduced, and then by adopting the principle and steps of Zhang Zhengyou camera calibration method, two cameras are calibrated by which their internal and external parameters are obtained.Thirdly, this paper analyzes and compares several image feature extraction and matching methods with focus on introduction of the feature extraction and matching process of SIFT algorithm. In view of the defects of SIFT algorithm that it requires large amount of computation and doesn't take image color and shape into consideration, this paper employs a method based on combination of HSV color space component and SIFT feature extraction algorithm to determine the initial region of target objects and then to match the feature points. The 3D coordinate information of the feature points is then recovered according to the binocular stereo vision parallax principle and a ranging and positioning experiment is carried out under different distances afterwards. Finally, The improved algorithm is verified with image matching experiment and ranging experiment. And the results show that the improved algorithm has a great improvement in terms of time performance and accuracy of feature extraction, achieves higher robustness in image feature extraction and matching and improves the real-time performance and accuracy of image matching.Finally, after the fusion of binocular stereo vision information and odometer information by using the EKF algorithm, the real-time localization of mobile robot is studied. Simulation experiments show the rationality and effectiveness of the proposed algorithm. This method not only solves the problem of inaccurate information of the feature points obtained from single camera vision, but also avoids the problem of large amount of computation and poor positioning accuracy in binocular visual image processing.