Spatio-temporal Context-guided Algorithm For Lossless Geometry Point Cloud Compression

With the increasing performance of multi-platform,multi-resolution 3D acquisition equipment,3D laser scanning technology has been widely applied in scientific and engineering research,such as engineering measurement,cultural heritage,smart city,virtual and augmented reality,due to its characteristics of all-weather,high speed,high precision,and high density.However,the massive,unstructured,and uneven density characteristics of point cloud data bring great pressure to limited storage space and transmission bandwidth.Up to now,compression methods can hardly adapt to the exponentially increasing point cloud data capacity and achieve a balance between high compression ratio,low distortion rate,and computational cost.Therefore,achieving low bit rate and low distortion rate point cloud compression in limited storage space and network transmission bandwidth has important theoretical significance and practical value.In this context,therefore,we propose a lossless point cloud geometric compression algorithm based on spatio-temporal context,which is suitable for compression and decompression tasks of 3D point clouds in multiple scene types,single frames,and continuous frames.Specifically,the method includes the following two modules:(1)Analyzing the spatial structural characteristics of the point cloud model and dividing it into image-sliced structures of unit thickness,predicting the geometric information of each slice based on the traveling salesman algorithm and making full use of the spatial information correlation of adjacent layers in the frame,and designing a spatial context-based intra-frame prediction algorithm;(2)incorporating the spatiotemporal information correlation of adjacent layers between frames,designing a spatio-temporal context-based inter-frame prediction algorithm,and then constructing an upper and lower context dictionary to choose the current best context,selecting and updating the global optimal value based on entropy estimation,and subsequently utilizing the adaptive properties of the encoder to achieve fast calculation of probability upper and lower bounds.After these steps,the method realizes lossless point cloud geometric information compression at a low bit rate,outputting the compressed and decompressed results of 3D point clouds in multiple scene types,single frames,and continuous frames.The proposed method is validated with MPEG datasets and handheld scanning datasets collected from buddha of Mogao Grottoe.Qualitative and quantitative experiments demonstrate that the actual application effect of the algorithm has achieved the design goal and can meet the demand for lossless compression of point cloud geometric information in scenes of dynamic human images,static buildings,cultural relics,and heritage.In comparison experiments,the proposed method is significantly better than the baseline method,verifying the effectiveness of the method design.The main innovation of this article is that it not only inherits the advantages of mature algorithms for image compression but also avoids the loss of geometric feature information through hierarchical division,designing a traveling salesman algorithm prediction and an encoding mode that is applicable to both intra-frame and inter-frame scenarios,and studying adaptive binary arithmetic coding that considers global context information and entropy estimation values.This achieves efficient compression of 3D point clouds for multiple scene types,single frames,and continuous frames,providing stronger data support for engineering measurement,cultural relics and heritage,smart cities,virtual reality,and augmented reality research.