強化立體視覺解析度以減少影像深度量測誤差之研究

Enhancing the stereo vision resolution to reduce image depth measurement error

黃孟賢(Meng-Sian Huang)

指導教授 ： 廖俊鑑

朝陽科技大學/資訊學院/資訊與通訊系/碩士(2015年)

摘要

立體視覺(stereo vision)是模仿人類雙眼的系統，透過左右兩張影像中，相同物件的成像位置不同，來判斷該物件在實際環境下的位置。亦有利用立體視覺來進行距離量測的方法，雖然方法本身簡單且準確，但是仍有一嚴重的缺點，就是當量測目標的距離越遠時，會因為深度量測解析度(depth measurement resolution)不足，而可能產生較大的誤差，所測得之結果準確度(accuracy)將會越低。
　　本文提出一改善立體視覺距離量測之方法，先將立體視覺量測的影像透過Census Transform特徵比對，使系統能夠對兩張影像進行匹配，找出不同影像中之相同目標點。接著再改變量測的焦距(focal length)或基線(base line)長度進行解析度改善量測，隨著焦距或基線改變，使得立體視覺模型(stereo vision model)亦跟著改變，並產生新模型。最後將立體視覺量測與解析度改善量測的立體視覺模型重疊，可發現深度量測解析度因重疊的線條而提升。且透過本研究之實驗結果證實，此方法可使解析度獲得改善，確實是可應用於遠距離影像距離量測系統上。也因為改善了長距離的量測解析度，使我們可以準確獲得較遠之影像深度資訊，因此得以實現立體視覺大氣能見度的量測。

關鍵字

能見度量測 ； 機械視覺 ； 立體視覺 ； 距離量測 ； 數位影像處理 ； Census Transform

並列摘要

The stereo vision system which imitates the dual-eye system of a human being depends on distinct image positions of an identical object on two separate images to determine the object’s position in realistic environment. Albeit simple and accurate, the method based on stereo vision for distance measurement is still an imperfect method which induces significant errors and low accuracy attributed to the improper depth measurement resolution when the target to be measured is positioned at a distance.
In this research, we offer a method for improvement of distance measurement based on stereo vision. First, two images which were measured in stereo vision are matched to each other through census transform feature comparisons for finding the same target points in different images. Secondly, the focal length or the base line for calibration measurement is modified for change of a stereo vision model and further generation of a new model. Finally, the stereo vision measurement and the stereo vision model for calibration measurement overlap each other for improvement of the depth measurement resolution in virtue of overlaps of lines. It can be seen from experimental results in this research that our method improving the resolution can be truly used in a remote image measurement system. With the remote measurement resolution improved, our method is advantageous to capturing depth information of remote images and achieving stereo vision measurement in atmospheric visibility.

並列關鍵字

Machine vision ； Stereo vision ； Distance measurement ； Digital image processing ； Visibility measurement ； Census Transform

參考文獻

[1] Y. Y. Huang and M. Y. Chen, "3D object model recovery from 2D images utilizing corner detection," 2011 International Conference on System Science and Engineering (ICSSE), pp. 76-81, 8-10 June 2011, Macau, China.

[2] M. Samadi, M. F. Othman and S. M. Amin, "Stereo vision based robots: Fast and robust obstacle detection method," 2013 9th Asian Control Conference (ASCC), pp. 1-5, 23-26 June 2013, Istanbul, Turkey.

[5] Y. Li, S. Wang, Y. Zhao and Q. Ji, "Simultaneous Facial Feature Tracking and Facial Expression Recognition," IEEE Transactions on Image Processing, vol. 22, no. 7, pp.2559-2573, July 2013.

[6] B. Mirmahboub, S. Samavi, N. Karimi and S. Shirani, "Automatic Monocular System for Human Fall Detection Based on Variations in Silhouette Area," IEEE Transactions on Biomedical Engineering, vol. 60, no. 2, pp. 427-436, Feb. 2013.

[7] S. D. Member, "Automatic License Plate Recognition (ALPR): A State-of-the-Art Review," IEEE Transactions on Circuits and Systems for Video Technology, vol. 23, no. 2, pp. 311-325, Feb. 2013.