Research On Robot’s Simultaneous Localization And Mapping Based On Panoramic Vision

Mobile robot simultaneous localization and mapping based on the visual sensor has attracted attentions of many researchers in mobile robot navigation areas. Because the panoramic vision sensor has a360degree comprehensive perspective and information-rich so it is ideal for use in mobile robot navigation. To the problem that panoramic vision can hardly capture the environment information fully and evenly, and that the map features are not stable enough, this paper has studied the mobile robot simultaneous localization and mapping based on the panoramic vision by the united model of multi-camera system and multi-view geometry.First of all, this paper used the multi-camera system fixed on the mobile robot for acquisition of the360degree panoramic images. According to the unified model of multi-camera system, the multi-camera viewpoints have been unified to the model center, so that the multi-camera coordinate system realized the consistency. And the problem:(1) ambiguity mapping;(2) spherical pixel expression, when the computer generate the discrete panoramic image, have been solved.Secondly, the panoramic images feature extraction and matching methods have been researched. And the commonly used panoramic image feature extraction methods have been discussed and compared. Through the comprehensive performance contrast, the PCA-SIFT method has been eventually selected, to extract panoramic image characteristics. After the PCA-SIFT descriptor was established, the feature points are matched roughly by the nearest NN method. Then, the matched feature points will be further processed by RANSAC methods. Through the experiment contrast, it was verified that the PCA-SIFT algorithm is superior to SIFT algorithm in comprehensive performance.Thirdly, the mobile robot SLAM was decrypted using mathematical method. And the motion model and observation model based on the multi-camera system was established, which can be updated through the EKF algorithm at the real-time. And the system structure of the SLAM based on the panoramic vision was put forward. In this paper, the ordinary SLAM simulation and the panoramic vision based SLAM simulation was realized, through setting different observation range to the visual sensor. The results show that the observation range of the sensor get larger, the characteristics can be tracked more effectively, and the uncertainty’s convergence is obtained faster. Then the SLAM experiment on the mobile robot was conducted. The experimental results verified the feasibility and effectiveness of the mobile robot SLAM solutions based on the panoramic vision, which was put forward in this paper.Finally, the mobile robot SLAM solutions based on the panoramic vision was summarized. The research of next phase and forward ideas was discoursed.