Tensor block-wise singular value decomposition for 3D point cloud compression

A novel and efficient block-wise decomposition-based codec (BDC) for a three-dimensional (3D) light detection and ranging (LiDAR) point cloud (PCD) image (BDCPCD) has been introduced in this paper. The raw LiDAR data is cleansed and normalized by applying the axis outlier detection and circular differential cosine transformation methods, respectively. Then, the iterative dimensionality reduction approach is used to decompose and quantize the tensor structured signal data through block-wise singular value decomposition and signal block vectorization methods, respectively. The final single order tensor is considered as a compressed bitstream for efficient transformation. The proposed BDCPCD is applied on three different dense 3D LiDAR PCD data sets. The results demonstrate that it outperformed the four existing well-known compression techniques, such as WinRAR, 7-Zip, Tensor Tucker decomposition, and Random sample consensus (RANSAC) point cloud compression algorithm. This iterative compression algorithm constantly reduces the 66.66% of tensor blocks in each iteration. This research proves that the BDCPCD compresses different sizes of 3D LiDAR PCD spatial data to be reduced into six bytes and averagely increases the quality of the decompressed image by 1.6 decibels than the existing Tucker based algorithm.