一個新的疊接分類器用於行人偵測

A New Cascade Classifier for Pedestrian Detection

詹定懋(Ding-Mao Jhan)

指導教授 ： 鄭文昌

朝陽科技大學/資訊學院/資訊工程系/碩士(2011年)

摘要

在此論文中我們提出一個結合自適應提升演算法與支持向量機的疊接分類器，並將它應用於行人偵測。在我們的行人偵測系統中首先利用固定尺寸的視窗影像由左至右，由上至下依序擷取行人候選區域以此進行特徵抽取，最後利用提出的疊接分類器完成行人偵測。本文主要有三個貢獻，首先透過對自適應提升演算法的分析改善挑選弱分類器的方法，使得疊接自適應提升演算法分類時能夠更著重於正例樣本的分類，其次，此提出的疊接分類器能自適應選擇自適應提升演算法或支持向量機，因此能有效依據訓練樣本建構疊接分類器，進而提升分類效果以及降低訓練時間，最後此方法也解決支持向量機對大的訓練樣本資料庫進行分類會非常耗時且不易收斂的問題。為了驗證我們所提出的方法，因此透過我們擷取的行人訓練樣本資料庫、PETs行人資料庫、INRIA行人資料庫以及MIT行人資料庫完成行人偵測實驗。實驗結果顯示，在我們的樣本資料庫及PETs資料庫的比較中我們的方法與現有的方法有同樣的分類效果，但是我們的方法有最短的訓練時間，而在較為複雜背景的INRIA與MIT行人樣本資料庫的比較上，我們的方法除了有最短的訓練時間之外還有最佳的結果。

關鍵字

行人偵測 ； 支持向量機 ； 自適應提升演算法 ； 分類器組合 ； 疊接分類器

並列摘要

In this paper, we propose a cascade classifier combined by AdaBoost and SVM, which will be applied to pedestrian detection. In the pedestrian detection system, we first adopt window image of fixed size to extract the selected pedestrian area in the sequence from left to right and top to bottom so as to conduct feature extraction and finally complete the pedestrian detection by our proposed cascade classifier. There are three contributions made in this paper. Firstly, it improves the method of selecting weak classifier through the analysis of AdaBoost algorithm which enables Cascade-AdaBoost to focus on the classification of positive samples. In addition, the proposed cascade classifier can adaptively choose AdaBoost and SVM. Thus, it can effectively construct cascade classifier based on the training samples so as to improve the classification effect and reduce training time. Finally, this method can also solve the problems of large time consumption and difficult convergence when SVM conducts classification to big training sample database. In order to verify our proposed method, we use our extracted pedestrian training samples database, PETs pedestrian database, INRIA pedestrian database and MIT pedestrian database to complete the pedestrian detection experiment. The experiment result shows that when comparing our samples database and PETs database, we find although our approach has the same classification effect as same as the present approach, it has the shortest training time. Moreover, in the comparison of the complicated INRIA and MIT pedestrian samples database, our approach not only has the shortest training time but also the best classification accuracy.

並列關鍵字

Pedestrian Detection ； SVM ； AdaBoost ； Combination of Classifiers ； Cascade Classifier

參考文獻

[2]. L. Malagon-Borja and O. Fuentes, “Object Detection Using Image Reconstruction with PCA,” Image Vision Computing, vol. 27, no. 1-2, pp. 2–9, 2009.

[4]. T. Gandhi and M. M. Trivedi, “Pedestrian Protection Systems: Issues, Survey, and Challenge,” IEEE Transactions on Intelligent Transportation Systems, vol. 8, no. 3, pp. 413-430, 2007.

[5]. M. Enzweiler and D. M. Gavrila, “Monocular Pedestrian Detection: Survey and Experiments,” Pattern Analysis and Machine Intelligence, vol. 31, no. 12, pp. 2179-2195, 2009.

[6]. D. Gerónimo, A. M. López, A. D. Sappa and T. Graf, “Survey of Pedestrian Detection for Advanced Driver Assistance Systems,” Pattern Analysis and Machine Intelligence, vol. 32, no. 7, pp. 1239 - 1258, 2010.

[8]. P. Viola and M. J. Jones, “Detecting Pedestrians Using Patterns of Motion and Appearance,” International Journal of Computer Vision, vol. 63, no. 2, pp. 153-161, 2005.