結合立體視覺與Kinect深度應用於室內自動車導航之研究

本論文中，我們將雙眼相機取得的左右影像做地平面校正，將校正後的左右影像相減後得到地面視差為零。接著將校正後的左右影像做區塊匹配，取得視差圖，在圖中可以清楚的分辨出地面與地面上具有高度之障礙物。此外，我們搭配深度測量裝置Kinect取得深度資訊，並經由三角幾何關係推算出物體垂直於地面之高度、物體與自動車之間的距離與物體相對於自動車之橫向偏移量。最後，使用權重的方法結合雙眼立體視覺系統與Kinect深度資訊，進而獲得更加完善之環境資訊，並利用這些資訊實現自動車導航系統，其中Kinect的最佳偵測範圍為100公分至350公分，在0公分至80公分內則是無法偵測到深地的，而其他的範圍則會有較大誤差的不可靠性存在。

關鍵字

Kinect ；視差圖；深度圖；區塊匹配；地平面校準；自動車

並列摘要

In this paper, we take the left and right images of binocular camera to do the ground plane cabliration, the disparity values of ground plane is zero after subtracting left image from right image, and then the disparity maps can be obtained by block matching, we can clearly see the ground plane and the objects with certain height, calculate the height which is perpendicular to ground plane and the distance between objects and Autonomous Land Vehicle(ALV) by the method of triangle geometry. In addition, we use the depth measurement device Kinect as another vision system to get the depth information, calculate the height which is perpendicular to ground plane and the distance between objects and ALV, the lateral displacement value by the method of triangle geometry. Finally, combination of stereo-vision and Kinect depth information by the weights. Further, we get more complete information, use the information to achieve ALV system, where where the Kinecct's best detection range is between 100 cm and 350 cm, can not detect the depth between 0 cm and 80 cm, and others range is larger error and unreliability.

並列關鍵字

Kinect ； Disparity map ； Depth map ； Block matching ； Ground plane alignment ； Autonomous land vehicle

參考文獻

[1]王鈺達，倒傳遞神經網路與地平面立體視覺作用於自動車導航之障礙物偵測與道路分類，碩士論文，國立台北科技大學自動化科技研究所，台北，2012.

[2]R. A. Hamzah, A. M. A. Hamid and S. I. M. Salim, “The Solution of Stereo Correspondence Problem Using Block Matching Algorithm in Stereo Vision Mobile Robot,” Proc. of IEEE 2nd Int. Conf. on Computer Research and Development, 2010, pp. 733-737.

[3]J. Weber, D. Koller, Q.T. Luong, and J. Malik, “An integrated stereo-based approach to automatic vehicle guidance,” International Conference on Computer Vision, Boston, June 1995.

[5]H. W. Keat and L. S. Ming, “An investigation of the use of kinect sensor for indoor navigation,” in TENCON 2012-2012 IEEE Region 10 Conference, 2012, pp. 1-5.

[12]M. Dash, and H. Liu, “Feature selection for classification,” Intelligent data analysis, Vol. 1, 1997, pp.131-156.

國際替代計量

結合立體視覺與Kinect深度應用於室內自動車導航之研究

全文下載

主題瀏覽