以分割式直方圖及卡爾曼濾波器為基礎並結合遮蔽處理之多目標影像追蹤

在此論文中，我們提出了一個結合分割式彩色直方圖及以卡爾曼濾波器作位置估測的多目標影像追蹤方法。遮蔽問題也在此方法中得到適當的處理。我們利用蘇恕德之兩段式背景相減法來得到移動物體的區域，接著找出各個相連的部分及把面積較小的去除。由於彩色直方圖在大小及姿態改變時有著相對穩定的優勢，因此被選為對物體外觀的一種描述。考慮到直方圖往往會損失空間的資訊，因此我們把每個物體分成上下兩部分，也就是說，每個物體被上下的紅、綠、藍，總共六個直方圖所描述。同時，隨著時間的前進，外觀模型會以指數加權移動平均的方式進行更新。另一方面，每個物體的位置都以卡爾曼濾波器進行估測。最後，我們結合外表模型及位置條件以實現追蹤。對於極具挑戰的遮蔽問題，我們也提出了一個包含合併、分開、被物體遮蔽及重新出現的新穎方法。我們也利用歷史資訊去解決嚴重甚至完全遮蔽的狀況。總的來說，這個方法可以處理物體間遮蔽及被背景物遮蔽中的長、短遮蔽，完全、部分遮蔽等問題。從眾多不同的實驗可以看到所提出的方法的強健度。

關鍵字

分割式彩色直方圖；卡爾曼濾波器；背景相減；外觀模型；模型更新；遮蔽處理

並列摘要

A multi-target visual tracking approach, which consists of segmented color histograms, position estimation by Kalman filter and occlusion handling, is proposed in this thesis. The moving object regions are extracted by using Su’s Two-Staged Background Subtraction approach. Connected components are located and those of small area are discarded. Color histograms are used as descriptors of an object’s appearance due to their intrinsic benefits of being relatively scale-invariant and posture-invariant. In order to resolve the problem of losing spatial information due to histogram, an object is segmented into the upper and lower parts. Thus, the objects’ appearances are described by six histograms, namely, R, G, B histograms corresponding to the abovementioned two parts. The appearance model is then updated through Exponentially Weighted Moving Average across time. Furthermore, Kalman filter is used for position estimation for each object. We combined the appearance model and position condition to implement the tracking approach. In view of the challenging occlusion problem, a novel approach consists of Merge, Split, Obstructed and Reappeared is proposed. Historical information is utilized such that severe and even full occlusions can be handled. By and large, this approach can handle long, short, full, partial occlusions efficiently for both inter-object occlusion and occlusion by obstructors. Various experiments verify the robustness of our approach.

並列關鍵字

segmented color histograms ； Kalman filter ； background subtraction ； appearance modeling ； model update ； occlusion handling

參考文獻

[1] A. Yilmaz, O. Javed, and M. Shah, "Object tracking: A survey," Acm Computing Surveys, vol. 38, 2006.

[2] S. T. Su, "Moving object detection based on two-staged background subtraction approach," Master Thesis, Department of Electrical Engineering, National Taiwan University, Taipei, 2009.

[3] D. G. Lowe, "Distinctive image features from scale-invariant keypoints," Computer Vision and Image Understanding, vol. 60, pp. 91-110, 2004.

[4] I. Haritaoglu, D. Harwood, and L. S. David, "W4: Real-time surveillance of people and their activities," IEEE Transactions on Pattern Analysis and Machine Intelligence, vol. 22, pp. 809-830, 2000.

[5] S. Belongie, J. Malik, and J. Puzicha, "Shape matching and object recognition using shape contexts," IEEE Transactions on Pattern Analysis and Machine Intelligence, vol. 24, pp. 509-522, 2002.

國際替代計量

以分割式直方圖及卡爾曼濾波器為基礎並結合遮蔽處理之多目標影像追蹤

全文下載

主題瀏覽