利用霍夫森林建構行人偵測技術

Hough Forest Based Human Detection

陳仕儒(Shr-Ru Chen)

指導教授 ： 黃仲陵 ； 林嘉文

國立清華大學/電機資訊學院/電機工程學系/碩士(2012年)

https://doi.org/10.6843/NTHU.2012.00696

摘要

在本篇論文中，我們提出了一種利用樣本比對的方法產生相似特徵，並結合了霍夫森林的分類演算法，透過霍夫森林產生出來的每一個葉點皆可視為一個行人身上的區域部位偵測器，這些葉點除了是行人的區域部位偵測器之外還同時扮演著一個行人中心位置的密碼書，因此一張區域圖片掉到一個葉點後可以被判斷為跟葉點裡的訓練資料為相同區域部位，之後利用此葉點的密碼書對行人中心位置進行投票，透過不同葉點不同部位的投票結果，在霍夫平面上擁有較多票數的地方為行人中心位置的機率就越高。
為了避免錯誤的投票位置，我們使用了一種滑動窗口的偵測策略，對每一個視窗裡的影像進行第一次投票，之後將這個視窗內區域最大值位置重新轉換到原本影像上，並於整張影像裡的相對位置進行第二次投票，透過這個方法可以大幅的降低兩行人中間的false positive情形。

關鍵字

行人偵測 ； 霍夫森林

並列摘要

In this paper, we propose a method for the human detection. This method use similarity feature as our feature and hough forest as our classifier. We calculate the similarity between input path and example patch, and this similarity is one bin of our similarity feature. This similarity feature makes the split of the node in hough forest more meaningful. Through the hough forest, it collect the similar patch in the same leaf node, so that the leaf node can be seen as the part detector. Besides the leaf node also severs as a codebook recording the possible locations of the human center. Based on the codebook, the patch falling in this leaf node can cast probabilistic votes for possible locations of the centroid of the human. After all the path cast their votes, more votes indicate more likely the centroid of the human is.
In order to reduce the error voting, we use the sliding window strategy to prevent casting the votes outside the window. The local maximum in the hough space of the window is then filtered by the threshold. The position with remaining local maximum is then transformed to the coordinate of the image, and cast the second vote in the hough space of the original image. By using the sliding window strategy, the filter of the threshold, and the second vote method, it can reduce the false positive in the image.

並列關鍵字

human detection ； hough forest

參考文獻

[1] N. Dalal, B. Triggs. “Histograms of oriented gradients for human detection,” IEEE Conf. Computer Vision and Pattern Recognition, vol. 1, pp. 886-893, July 2005.

[2] Xiaoyu Wang, Tony X. Han, Shuicheng Yan. “An HOG-LBP human detector with Partial occlusion handling,” IEEE Conf. Computer Vision, pp.32-39, 2009.

[3] Timo Ahonen, Abdenour Hadid, and Matti Pietikainen. “Face description with local binary patterns: Application to face recognition,” IEEE Trans. Pattern Analysis and Machine Intelligence, vol. 28, pp.2037-2041, 2006.

[4] Y. Freund and R. E. Schapire. “A Decision-Theoretic Generalization of On-Line Learning and an Application to Boosting,” Journal of Computer and System Sciences, vol. 55, pp. 119-139, 1997.

[6] A. Bosch, A. Zisserman, and X. Mu˜noz. “Image classification usingrandom forests and ferns”. ICCV, pp. 1–8, 2007.