KEYFRAME-BASED GLOBAL LOCALIZATION IN LARGE-SCALE SCENES

The purpose of scene recognition is to find keyframes from a known map that match the current frame. In large-scale scene positioning, the cumulative error of SLAM increases with the increase of the mileage of the unmanned vehicle due to the inevitable small error of inter-frame registration. Aiming at the localization problem of mobile robots in large-scale scenes, a global positioning method based on keyframes in large-scale scenes is proposed. First, the three-dimensional point cloud is transformed into single-line laser data, and key frames are selected from the single-line laser data to obtain a keyframe sequence. The trained laser spots then detect features and build databases and glossaries for efficient retrieval and location recognition. In order to verify the feasibility of the proposed method, multi-line lidar is used to conduct global positioning experiments in a wide outdoor environment. The experimental results show that the method can achieve a localization success rate of 61.2%, and the fastest speed is within 2ms. This method basically meets the accuracy and efficiency requirements of global positioning for autonomous driving in practical applications.