3D Mapping Based On A Spinning 2D Laser Range Finder

Nowadays many robots cope with tasks in motion.A high-accuracy map is basic and crucial for mobile robots comprehend environment and take actions.This thesis illustrates a 3D mapping strategy based on a spinning 2D laser range finder,achieves the goal that accumulate laser data to construct a 3D cloud map in motion.We aimed at high-accuracy 3D cloud map construction,focused on system architecture,sensor data processing,pose estimation.The main contributions are as follows:(1)The hardware and software architectures of mapping system are designed.To solve the multi-sensors time asynchronous problem,we propose a time synchronization strategy based on smoothing spline.Meanwhile,our time synchronization strategy is differed among sensors.A typical sensor utilizes time synchronization based on linear interpolation method.But for stereo visual odometry which has time uncertainty,we take advantage of reprojection and keyframe info,utilize synchronization method based on smoothing spline.(2)A pose estimation strategy with multi-sensor fusion based on JDL Model(Joint Directors of Laboratories Model)and EKF(Extended Kalman Filter)is proposed.This method adopts JDL model as multi-sensor fusion architecture.And in JDL's situation refinement level,we use EKF to fuse visual odometry data from a modified ORB-SLAM2 version and inertial measurement units' data.The pose estimation accuracy is enhanced by the complementarity of the visual odometry's position estimation and the inertial measurement units' rotation estimation.(3)A multi local map fusion optimization strategy based on ICP(Iterative Closest Point)is implemented.On-line method:to avoid the accumulative error caused by directly accumulate local map data,we fuse local maps with ICP to generate the final laser map.In this way,we can eliminate accumulative error of real-time pose estimation as much as possible,in the meantime maintained basic real-time performance.Off-line method:based on on-line method,we refine local map itself based on ICP algorithm,and assemble them to produce final map.