| Autonomous driving is a new trend in the development of the automobile industry.One of the major problems to be solved is the positioning of vehicles.This subject mainly studies that how vehicles determine their own position and complete the construction of global map only with lidar point cloud in unknown environment.This is simultaneous location and mapping.Firstly,this research uses a self-driving test vehicle of Baidu Apollo platform to collect environment data.The lidar point clouds are sampled at equal intervals and filtered with VoxelGrid.The final step of pretreatment is to divide data of road surface in order to reduce the number of point clouds and reduce the burden of the algorithm.Secondly,the pre-processed point cloud data will be matched.The first step is to use the principal component analysis to register the point cloud roughly and provide an appropriate initial value for accurate matching.The second step is to use the iterative closest point algorithm for accurate registration,and estimate the rotation matrix R and the translation matrix T.The third step is to put new points near the estimated results according to the gaussian distribution and complete the particle updating.Finally,according to eigenvalues,point cloud structure features can be divided into discrete,linear and planar structures.Gaussian mixture model is established respectively,and the weight and other parameters are solved by EM algorithm.Then the overlapped areas of the local map and the global map containing particles are rasterized,and the ratio of the number of grids with the same structural characteristics to the total number of grids is used as the importance weight to measure the mass of particles.The local map containing the optimal particles is added to the global map to complete the map construction.Through experimental analysis,the algorithm studied in this paper can locate the unmanned vehicle in real time and construct a clear environment map. |
PDF Full Text Request