應用於立體電視之結構光法近物深度量測重建演算法與晶片實現

Algorithm and Chip Implementation of 3DTV Depth Map Reconstruction Based on Structure Light Scheme

黃品綱(Pin-Kang Huang)

指導教授 ： 范育成

國立臺北科技大學/電資學院/電腦與通訊研究所/碩士(2012年)

摘要

近年來3D相關產業蓬勃發展，3D內容的需求量相對的也越來越大。3D內容的產生主要是仰賴著DIBR(Depth Image Based Rendering)的技術來產生，其僅需要藉由一張圖像以及其相對應的深度圖輸入，經過處理後即可產生虛擬多視角影像，接著就可以藉由3D顯示器達到立體的視覺效果。因此，深度資訊的萃取對於3D產業而言是很重要的。以往對於物體的深度資訊我們僅能利用相機擷取並搭配各種深度演算法計算出其深度，但是這樣的效果不彰，直到微軟推出了Kinect，其內建深度感測元件，讓我們在物體的深度萃取上多了更好的選擇。但是Kinect在設計上，主要是拿來做體感控制，所以在深度精確度上較為粗糙，對於物體細節沒辦法有效的描繪出來。
本論文實現一結構光法近物深度量測重建系統，運用投影機投射出結構光圖形，並利用數位相機接收，經過一連串邊緣萃取演算法，其中包含了雙向Sobel水平邊緣、型態濾波器、邊緣細線與修復演算法，對欲解碼的結構光編碼圖形進行處理及配對，計算出樣本產生的變化量與梯度值。並搭配建立的梯度與實際深度距離轉換資料庫，取代以往量測所使用的高複雜度三角演算法，可以快速且精確的得到實際深度值。此外，考慮到人眼對於深度感知的非線性變化，也將梯度值進行量化，使其符合人眼立體視覺感受。我們不僅可以輸出實際深度資訊也可以輸出量化後的深度圖，並可經由DIBR技術播放到3D顯示器觀看到立體效果。

關鍵字

結構光法 ； 深度量測 ； 深度圖 ； 2D對3D轉換

並列摘要

3D industry vigorous growth in recent years, it leads to the requirement of 3D contents becoming more and more important. 3D contents production rely on the DIBR(Depth Image Based Rendering) technology, it only need to input a color image and its related depth map. After proceeding, we can get a virtual multi-view images and view with auto-stereoscopic 3D display. In the past, we used multi-camera to capture images and computed depth distance value via algorithm, the result quality is poor. Until Microsoft launched the Kinect, its built-in depth sensor, let us have more choice to extract the depth information. But its depth precision is rough, can not describe the detail of object. As a result, we can know the depth information is important for 3D issues.
In this thesis, we proposed an algorithm and chip implementation of 3DTV depth map reconstruction based on structure light scheme. We use a projector to project the structure light pattern and a camera to capture the pattern. After a series of edge extraction algorithm, which contains bi-direction sobel horizontal edge detection, morphology filter, edge thinning algorithm, edge recovery algorithm, pattern matching, gradient calculation, etc. Instead of triangulation algorithm, we build the gradient database to convert gradient to physical distance, it is fast and accurate. In addition, we take human’s perception into consideration, not only output the physical distance depth map but also quantificational depth map, through DIBR technology to 3D displays to view three-dimensional effects.

並列關鍵字

Structure Light ； Depth Measurement ； Depth Map ； 2D to 3D

參考文獻

[6] R. J. Woodham, “Photometric Method for Determining Surface Orientation from Multiple Images,” Optical Engineering, vol. 19, no. 1, pp. 139-144, January 1980.

[7] 陳國源，以影像處理為基礎的三次元量床自動量測策略設計，大同大學碩士論文，2005。

[9] Emad Attalla, Pepe Siy, “Robust Shape Similarity Retrieval Based on Contour Segmentation Polygonal Multiresolution and Elastic Matching,” Pattern Recognition, vol. 38, 2229-2241, 2005.

[12] S. B. Gokturk, H. Yalcin, and C. Bamji,“A Time-of-Flight Depth Sensor – System Description, Issues and Solutions,”IEEE Computer Vision and Pattern Recognition Workshops, Washington, USA, June 27-July 2, 2004, pp. 35-44.

[14] Chu-Song Chen, Yi-Ping Hung, Chiann-Chu Chiang and Ja-Ling Wu, “Range Data Acquisition Using Color Structured Lighting and Stereo Cision, “Image and Vision Computing, pp.445-456, 12 November, 1996.