本文爲葡萄牙里斯本技術學院（作者：Miguel Branco RoqueNazaré Ferreira）的碩士論文，共110頁。
點雲是通過三維座標系中的一組點對全光函數進行採樣而獲得的表示模型，具有可選的相關屬性，例如顏色、表示對象或表面。近年來，在視覺信息的獲取和顯示方面取得了一些新的進展，如3D沉浸式臨場感和3D體育廣播等。因此，一項具有挑戰性的任務是以有效的方式表示和編碼必要的視覺信息，以便於存儲、傳輸和處理這些數據。對於動態點雲，即隨時間變化的點雲，編碼解決方案的開發仍處於非常初始的階段，還有很大的改進空間。最近，MPEG集團在這個領域發起了一個標準化項目。
考慮到現有的動態點雲編碼解決方案不多，本文提出了一種同時利用動態點雲時空相關性的新方案。該方案利用點描述符估計當前幀和以前幀之間的運動，建立兩幀中點之間的對應關係，用幾何變換（homography）對運動進行建模，然後生成一個具有一定扭曲步進的運動補償幀。再對運動補償幀和目標幀之間的差異進行編碼和傳輸，如基於八叉樹的雙緩衝點雲庫（PCL）編解碼器。計算結果表明，當兩個連續幀之間存在高速運動時，運動補償幀比前一幀具有更好的預測性能。然而，RD性能評估表明，這種改進不足以使所提出的解決方案在速率失真方面優於使用歷史編碼幀的原始PCL編解碼器。
Point clouds are a representation model obtained from the samplingof the plenoptic function through a set of points in a 3D coordinate system,with optional associated attributes, e.g. the colours, representing objects orsurfaces. Recent advances in capturing and displaying of visual information areenabling more immersive and realistic user experiences, such as 3D immersivetelepresence and 3D sports broadcasting. Thus, a challenging task consists inrepresenting and coding the necessary visual information in an efficient way,to facilitate the storage, transmission and processing of such data. For dynamicpoint clouds, i.e. point clouds changing through time, the development ofcoding solutions is still at a very initial phase, with large room forimprovements. Recently, the MPEG group has launched a standardization projectin this field.
Considering the few dynamic point cloud coding solutions available in theliterature, this Thesis proposes a novel solution to exploit both temporal andspatial correlations in a dynamic point cloud. The proposed solution estimatesmotion between the current and previous frames by using point descriptors, establishingcorrespondences between points in the two frames, modelling the motion with ageometric transform (homography) and then producing a motion compensated framewith some warping step. Then, the differences between the motion compensatedand target frames are encoded and transmitted, as in the octree-based doublebuffering Point Cloud Library (PCL) codec. The evaluation shows that the motioncompensated frame has a better prediction than the previous frame, when thereis high motion between two successive frames. However, the RD performance assessmentshows that this improvement is not enough for the proposed solution tooutperform in rate distortion the original PCL codec, which uses the past codedframe.
- 引言
- 點雲：基本和主要編碼解決方案
- 基於單應的點雲編解碼器：體系結構與工具
- 性能評估
- 結論與未來工作展望
附錄A 文獻中關於點雲描述符的綜述
附錄B PCL中可用聚類技術綜述
附錄C 時間預測Rmse評估圖
更多精彩文章請關注公衆號：