SLAM Map Recover Algorithm Based On Submap

Recently with the booming development of unmanned aerial vehicles,robots,autonomous driving,indoor measurement,SLAM technology has been widely concerned by many researchers and become a research hotspot in computer vision.According to the data collected by the sensor,such as images,laser-scanning point cloud,etc.,it can obtain the position and orientation of the sensor and the map of the scene in real time,so as to provide the fundamental data of positioning and navigation for robots and vehicles.In general,the process of visual-based SLAM includes 4 steps: tracking,mapping,loop detection and global optimization.System obtains the pose of current video frame captured by sensor through tracking(the current pose of the sensor is determined),and then perform forward intersection with several key frames and local optimization to construct map.When the closed-loop trajectory is detected,system starts global optimization to reduce the cumulative error.In the above workflow,due to the blurred image and lack of texture of the scene,tracking lost occurs very often.In case of tracking failure,most current vSLAM systems adopt the method of relocalization,which has a fatal disadvantage.If a similar scene is not detected after tracking lost,the system will remain in the lost state and cannot continue mapping until the relocalization is successful,and the map information cannot be recovered from the failure of the system tracking until relocalization.This paper proposes a lost map recovery algorithm based on submap and undirected connected graphs.In this algorithm,the global map is divided into several submaps by where the tracking loss occurs.Each submap is regarded as the node of the undirected connected graph,and the common map points between the sub-maps are regarded as edges,and all the submaps are merged into a whole by finding the transformation path among each node.At the same time,this paper also explores the selection of reference submap for coordinate transformation when merging submaps and the selection of transformation path among submaps.Results of experiments show that the method proposed in this paper works better than current mainstream SLAM methods in terms of tracking lost and map completeness.