Research On Handheld Multi-Sensor Fusion SLAM Algorith

SLAM(Simultaneous Localization and Mapping)technology has become a research focus in the fields of 3D reconstruction and automatic driving because it can construct high-precision maps in real time.Aiming at the problems that traditional SLAM has low precision in complex environments,it is impossible to construct a complete map,and the map has no color and texture information,this paper proposes a positioning and mapping method that integrates 3D lidar,IMU(Inertial Measurement Unit),servo motor,and visual camera.The main research contents are as follows:(1)The overall scheme of handheld map building method is designed,and the software algorithm is divided into device side and PC side for the problem that the low-cost motherboard used in handheld map building method cannot run the building algorithm online.The device side is used to drive the sensor and save the data,and the PC side is used to run the SLAM algorithm to build a high-precision map.The connection between the two sides is established by saving the data to the SD card on the device side,and then running the SLAM algorithm by reading and parsing the SD card data on the PC side.(2)Aiming at the problems of low accuracy and incomplete mapping,this paper constructs a complete and high-precision environmental map by lidar,IMU,and servo motors.First,the servo motor is used to drive the lidar to rotate to expand the scope of the map,and the servo motor and IMU are used to correct the point cloud rotation and motion distortion.Secondly,according to the laser radar measurement error,the uncertainty model of point cloud and plane features is accurately established,and the matching distance threshold is calculated to eliminate wrong point cloud matching.Finally,the adaptive probabilistic voxel map of hash table mapping is used to improve the robustness to apply to different environments,and the planar features and their related attributes are combined with voxels to improve the update and query efficiency of voxel maps.(3)In view of the lack of color texture in the reconstruction process of the point cloud map,this paper fuses the visual camera to output the image with pose,and uses the pose to project the image to the point cloud map to construct a color high-precision map containing texture information.First of all,in order to improve the pose accuracy and enhance the effect of projection coloring,a joint variable speed motion model and reverse optical flow method are proposed for feature matching,which improves the matching accuracy.Finally,the pose and image are combined,and the pixels are projected into the laser point cloud for coloring.(4)In order to verify the effectiveness of the handheld mapping method,a handheld mapping device was built first.Secondly,the proposed joint variable speed model and the feature matching method of the reverse optical flow method were tested on the TUM dataset.By comparing the running results of ORB-SLAM2,the accuracy has been greatly improved.Finally,a large number of mapping experiments were done in the real environment using handheld mapping equipment.The results showed that details such as railings,stairs,trees,and floors in the map were clearly visible and the texture was clear after coloring,indicating that the method proposed in this paper can be used in large-scale and complex indoor and outdoor scenes.high-resolution maps can be constructed.