Research On Compression Algorithm Of Vehicle-Mounted LiDAR Point Cloud

As a key component of point cloud acquisition technology,LiDAR(LiDAR,Light Detection and Ranging)acts as an eye in autonomous driving.The vehicle-mounted LiDAR point cloud system is the prerequisite foundation for the vehicle to perform environment perception,path planning,navigation and other behaviors to realize autonomous driving.However,in practical applications,LiDAR point clouds have a wide distribution range and a large amount of data,which poses a huge challenge to the existing transmission and storage technologies.Therefore,the demand for efficient point cloud compression schemes has become increasingly prominent.This paper is dedicated to the research of the LiDAR point cloud data compression framework in the field of autonomous driving.Based on the analysis of the characteristics of the vehicle-mounted LiDAR point cloud data,it focuses on the point cloud intra-frame compression and inter-frame compression algorithms,and explores point cloud segmentation and point cloud structured,sparse sampling,motion estimation and other solutions have the effect of reducing the geometric spatial and temporal redundancy of the point cloud and the impact on the compression performance.The main work and conclusions are as follows:First,for the spatial redundancy of LiDAR point clouds,a new point cloud intraframe compression framework based on morphological segmentation and non-uniform sparse sampling is proposed.The LiDAR point cloud is divided into two parts of ground and non-ground point clouds through a progressive morphological filter,and the two parts are de-redundant and sparsely sampled with different intensities.Then,the point cloud is divided by the general strategy of point cloud structuring proposed in this paper.The three-dimensional form is mapped to the two-dimensional form(represented as a distance image),and the pixel value distribution of the distance image is characterized by the form of an occupancy map.At the same time,combined with Morton code sorting,the two-dimensional form of the point cloud is expressed as a more compact one-dimensional distance vector form.Finally,the image coding method is used to further compress the occupancy map and distance vector.The quality evaluation of the proposed intra-frame compression scheme shows that the performance of the proposed scheme is far better than that of MPEG anchor and Google Draco method,slightly better than G-PCC,and can achieve higher reconstruction quality at larger bit rates.Secondly,aiming at the time redundancy of LiDAR point clouds,a point cloud inter-frame prediction scheme based on inter-class motion estimation is proposed,which together with the intra-frame compression scheme constitutes a point cloud hybrid compression framework.Based on the non-ground point cloud spatial distribution characteristics and time correlation of the point cloud sequence,the point cloud sequence is divided into key frames(I frames)and prediction frames(P frames)and the non-ground is further divided,respectively,I and P frames are divided After matching the point cloud,perform inter-class motion estimation,and calculate the motion information of each part.Furthermore,it only needs to perform intra-frame compression on the ground point cloud and a small number of unmatched and clustersegmented remaining point clouds,and encode the motion information.The evaluation of the point cloud compression hybrid framework based on inter-frame prediction shows that the compression performance is improved on the basis of the intra-frame compression scheme.Compared with the G-PCC and the emerging LiDAR point cloud spatio-temporal compression scheme,there is a significant performance gain.In summary,this thesis proposes a set of point cloud compression algorithms suitable for dynamic capture based on the temporal and spatial correlation of vehiclemounted LiDAR point clouds,which can achieve efficient compression of LiDAR point clouds.The research results will provide academic references for artificial intelligence industries such as intelligent driving,mobile robots,augmented reality,and mixed reality.