基於視訊點雲壓縮器(V-PCC)之三維點雲前處理改進

Improved 3D Point Cloud Preprocessing for Video-Based Point Cloud Compression(V-PCC)

林敬軒(Ching-Hsuan Lin)

指導教授 ： 林鼎然 ； 邱奕世

中原大學/電機資訊學院/電子工程研究所/碩士(2022年)

https://doi.org/10.6840/cycu202200047

摘要

視訊點雲的壓縮(Video-based Point Cloud Compression)現有標準利用分割方式GRS (Grid Refine Segmentation)，將三維空間的點雲資訊投影到不同的二維平面，再透過現有二維影像壓縮技術標準HEVC(High Efficiency Video Coding)進行二維影像壓縮。本論文根據適用於三維點雲場景語意分割之深度學習網路為基礎，改善V-PCC將點雲資訊投影到多個不同二維平面的投影分類，並在幾何BD-PSNR平均提升0.02dB。另外，在GRS處理過程至少有90%以上的執行程序並未改變投影面結果，因此本論文針對這點進行演算法的加速優化，預測該點後續投影平面是否有變化的可能，減少不必要的計算步驟。BD-Rate增加幅度不大的情況下，在GRS階段可降低15%的執行時間。

關鍵字

V-PCC ； GRS ； HEVC ； 加速 ； 投影平面 ； 深度學習

並列摘要

Video-based Point Cloud Compression (Video-based Point Cloud Compression) is the existing standard that uses the segmentation method GRS (Grid Refine Segmentation) to project the 3D point cloud information into different 2D planes, and then use the existing mature 2D image compression technology standard HEVC (High Efficiency Video Coding) performing 2D image compression. Based on a deep learning network suitable for semantic segmentation of 3D point cloud scenes, this paper improves the projection classification of V-PCC onto multiple different 2D planes, and improves Geomotry BD-PSNR by an average value of 0.02dB. Also, in the GRS process, at least 90% of the cases did not change the projection surface results during the refinement. Based on this point, we aim to accelerate the optimization of the algorithm, and predict whether the subsequent projection plane may change, and reduce unnecessary computations. The results show that the increase in BD-Rate is not large, and the execution time can be further reduced by 15%.

並列關鍵字

V-PCC ； GRS ； HEVC ； Accelerate ； Projection plane ； Deep learning

參考文獻

[1] Li Cui, Rufael Mekuria, Marius Preda, and Euee S. Jang, "Point-cloud compression: moving picture experts group's new standard in 2020," IEEE Consumer Electronics Magazine, vol. 8, pp. 17-21, July 2019.

[2] Charles R. Qi, Hao Su, Kaichun Mo, Leonidas J. Guibas, "PointNet: deep learning on point sets for 3D classification and segmentation," 2017 IEEE Conference on Computer Vision and Pattern Recognition (CVPR), 2017.

[3] Charles R. Qi, Li Yi, Hao Su, Leonidas J. Guibas, "PointNet++: deep hierarchical feature learning on point sets in a metric space," 31st Conference on Neural Information Processing Systems (NIPS 2017), Long Beach, CA, USA.

[4] Yangyan Li, Rui Bu, Mingchao Sun, Wei Wu, Xinhan Di, Baoquan Chen, "PointCNN: convolution on X-transformed points," 32nd Conference on Neural Information Processing Systems (NeurIPS 2018), Montréal, Canada.

[5] Francis Engelmann, Theodora Kontogianni, Bastian Leibe, "Dilated point convolutions: on the receptive field size of point convolutions on 3D point clouds," 2020 IEEE International Conference on Robotics and Automation (ICRA), 2020.