基於多演算法的行人追蹤

Pedestrian Tracking with Mixed Algorithm

楊凱筑(Wen-Shiung Chen)

指導教授 ： 陳文雄

國立暨南國際大學/科技學院/電機工程學系/碩士(2010年)

摘要

行人目標追蹤是計算視覺領域的一個重要分支，在視覺導航、智能監控、醫療診斷等方面被廣泛的應用。它們可以用來取代傳統的偵測方式，提高系統的效能，降低時間成本。但現實中普遍存在的遮擋問題，會引起運動目標的形變，導致目標檢測、提取和辨識的不穩定，產生目標資訊的遺失或者誤判，導致跟蹤系統失效。由此可知，如何在遮擋下保持目標跟蹤的有效性，是目標跟蹤問題當中，一個極待解決的重要問題。
本論文所提的行人偵測與追蹤系統，在偵測階段，總共分成背景偵測、前景切割、陰影去除等部分。系統追蹤部分，是採用顏色特徵的提取，利用建立直方圖，找出區域分布最廣的色彩，利用mean shift的概念，給予較高的權重，進行目標物的追蹤。基於色彩直方圖的色彩信息，對遮擋下的行人目標，具有比較好的跟蹤效果。但在環境中有相似色彩的物體時，容易造成行人匹配錯誤，所以另外提出利用隨機霍氏轉換法，求得每個行人走路的平均跨幅角度，當作是另一種特徵依據。當目標因為顏色相似導致判斷錯誤時，能夠依靠額外的步伐角度訊息進行資料庫比對。進一步找出正確的行人。

關鍵字

行人追蹤 ； 平均位移 ； 隨機霍氏轉換

並列摘要

Pedestrian Tracking is an important branch of computer Visual field, it's widely used in visual navigations、intelligent monitoring、medical diagnosis etc. They can be used to replace the traditional detection methods、improve system performance、reduce the time cost. But in reality, occlusion often occurred, it’s will cause moving target deformation, then lead to target detection, extraction and identification are unstable, produce objective information loss or miscarriage, and tracking system failure. From this, how to maintain the effectiveness of tracking under the occlusion is a very important issue to be resolved.
In this thesis, we propose pedestrian counting system. In the detection stage, it can be divided into background detection、foreground segmentation, and shadow removal etc. The parts of tracking are using color extraction, build histogram to find the colors which are widely distributed in the region, and use the concept of mean shift, given a higher weight for tracking. Based on color information of color histogram, it has better tracking efficacy for pedestrian target under occlusion. But in the condition of objects with similar colors, it easily cause pedestrian matching error, so proposed another method “Random Hough Transform” to obtain each average cross-site angle of pedestrian walking, as another characteristic. When the target is misjudged by the similar colors, we can compare additional step angle information to the database, and to find the right pedestrian further.

並列關鍵字

Pedestrian Tracking ； Mean Shift ； Randomized Hough Transform

參考文獻

[1] C. Wren, A. Azarbayejani, T. Darrell, and A. Pentland, “Pfinder: Real-Time Tracking of the Human Body,” IEEE Transactions on Pattern Analysis and Machine Intelligence, vol. 19, no. 7, pp. 780-785, 1997.

[2] J. Barron, D. Fleet, and S. Beauchemin, “Performance of Optical Flow Techniques,” International Journal of Computer Vision, vol. 12, no. 1, pp. 42-77, 1994.

[3] C. Anderson, P. Burt, and G. van der Wall, “Change Detection and Tracking Using Pyramid Transformation Techniques,” in Proceedings of SPIE - Intelligent Robots and Computer Vision, vol. 579, pp. 72-78, 1985.

[4] D. Comaniciu, V. Ramesh, and P. Meer, “Kernel-Based Object Tracking,” IEEE Transactions on Pattern Analysis and Machine Intelligence, vol. 25, no. 5, pp. 564- 577, 2003.

[5] J. Allen, R. Xu and J. Jin, “Mean Shift Object Tracking for A SIMD Computer,” in Processing IEEE Int’l Conference Information Technology and Applications, vol. 1, pp. 692-697, 2005.