Multi-Source Data Fusion For Consistency Optimization Of Helmet Mobile Mapping Point Cloud

Point cloud,a kind of new fundamental surveying and mapping data,has been widely used in the electric power industry,forestry,railway,etc.Current laser scanning systems can acquire large-scale outdoor point clouds quickly.In contrast to the outdoor environment,the indoor environment is not convenient to access,and not easy to acquire high precision point cloud.The helmet-based mobile mapping system is of small volume and easy operation and is a convenient tool for 3D data acquisition which can help deal with extracting 3D information from both above and underground,indoor and outdoor scenarios.However,on the one hand,it’s hard to receive reliable GNSS signals indoors or where there are lots of buildings,and correcting trajectories globally is not convenient.On the other hand,due to the limitation of cost and the weight the device can bear,putting on a high-precision IMU is sometimes unrealistic.These two reasons lead to the different layers of point clouds when mapping over long distances and time,which is hard to satisfy the needs of automatic driving,management and monitoring of natural resources,3D real scenes of China,etc.Therefore,it is meaningful to develop a method fusing multi-source data and then optimize the consistency of the point cloud on a helmet-based mobile mapping system for dealing with the inconsistency of large-scale mobile mapping point clouds.The main work is as follows:(1)To deal with high-dimension parameters while optimizing the local consistency of the point cloud,this dissertation proposes a bundle adjustment-based method for the local point cloud consistency optimization by constructing a plane feature-related 1D variable.This method optimizes point cloud and Li DAR poses simultaneously and can improve the quality of the local point cloud effectively.(2)To deal with the high computational complexities and a slow convergence rate while optimizing the consistency of the global point cloud,this dissertation proposes a hierarchical strategy by splitting mobile mapping data into sub-maps,matching pairs,building fine matching constraints concerning local features,and combining with IMU pre-integration constraints,GNSS constraints,and odometry constraints to construct the residual equations.This method can solve the problem of different layers of global point clouds and drifts of global poses.(3)To validate the effectiveness of this optimization method,this dissertation has done experiments based on a helmet-based mobile mapping system WHU-Helmet.After checking the trajectory accuracy,the pose error has reduced from 0.98 m to 0.27 m and the mean attitude error has reduced from 2.8° to 2.1 °.After checking the point cloud quality,the mean checkpoints offsets have reduced from 2.07 m to 0.37 m,the mean global point cloud offsets have reduced from 1.14 m to 0.18 m and the plane fitting RMSEs have reduced from 0.17 m to 0.07 m.