基於三維模型和部位資訊增強之車輛搜尋系統

Augmenting Vehicle Retrieval Using 3D Model and Part Information

蔡旻光(Ming-Kuang Tsai)

指導教授 ： 徐宏民

國立臺灣大學/電機資訊學院/資訊工程學研究所/碩士(2011年)

https://doi.org/10.6342/NTU.2011.00665

摘要

對於現今的監視器系統來說，在一般未受限制的環境中辨認車輛是一件很重要的事，但是因為監視器拍攝的地點、角度多變，有白天、夜晚等光影變化，加上背景環境的干擾，傳統的車輛辨識能力有限，所以我們提出了一個車輛搜尋、辨識的方法，這個方法是利用三維模型取出帶有較多特徵及資訊的部位，像是前車柵、車燈和輪胎等來進行搜尋。我們利用三維模型貼齊到圖片上，進一步擷取出這些特徵部位，這些部位經過圖片校正到特定的角度後，再與資料庫中的圖片進行比對。在實驗中，我們比較了不同的三維模型貼齊方法，並且確認了基於部位資訊的方式對於車輛搜尋的效果提昇是十分顯著的。

關鍵字

三維模型建立 ； 三維模型貼齊 ； 車輛搜尋 ； 車輛辨識 ； 部位校正

並列摘要

The capability to recognize vehicles in unconstrained environment plays an important role in surveillance system. However, due to large variations in viewing angle/position, illumination, occlusion, and background, traditional vehicle recognition is extremely challenging. We approach this problem in a different way by fitting 3D vehicles to a 2D image and searching the vehicles based on their informative parts such as grille, lamp, and wheel. These parts are extracted from the fitted 3D model constructed using Active Shape Model. In order to get consistent features from the parts of the same vehicle, the rectification technique is applied to transfer the parts from disparate view into the same reference view. In the experiments, we compare different 3D model fitting approaches and verify the impact of part rectification on the vehicle retrieval performance is significant.

並列關鍵字

3D model construction ； 3D model fitting ； vehicle retrieval ； vehicle recognition ； part rectification

參考文獻

[1] M. Arie-Nachmison and R. Basri. Constructing implicit 3D shape models for pose estimation. In ICCV, pages 1341-1348, 2009.

[2] T. Ahon, J. Matas, C. He, and M. Pietikainen. Rotation invariant image description with local binary pattern histogram fourier features. In SCIA, pages 61-70, 2009.

[3] A. E. Beaton and J. W. Tukey. The fitting of power series, meaning polynomials, illustrated on band-spectroscopic data. Technometrics, pages 147-185, 1974.

[4] J. M. Ferryman, A.D. Worrall, G. D. Sullivan, and K.D. Baker. A generic deformable model for vehicle recognition. In BMVC, pages 127-136, 1995.

[5] Y. Guo, Y. Shan, H. S. Sawhney, and R. Kumar. PEET: Prototype embedding and embedding transition for matching vehicles over disparate viewpoints. In CVPR, pages 1-8, 2007.