單眼視覺的行人偵測與追蹤

Monocular-vision pedestrian detection and tracking

郭志宏(Chi-Hong Kuo)

指導教授 ： 曾定章

國立中央大學/資訊電機學院/資訊工程學系/碩士(2011年)

摘要

隨著經濟的成長，機動車輛越來越多，交通事故也越來越頻繁；因此發展車輛防撞系統就變得更加的重要。特別是於市區的交通狀況，由於行人走動相當頻繁，因此行人偵測是一個重要的議題。在本研究中，我們提出了一個使用單眼相機的行人偵測與追蹤系統，並應用於複雜的市區環境中，以避免行人遭受到車輛撞擊。
在行人偵測中，我們先計算影像中每個像素的梯度，並且利用內插的方式將各種不同的梯度向量量化為九個固定方向的向量，將其累積以形成 Histograms of oriented gradients ( HOG )特徵，以做為偵測行人的特徵。當我們取出欲偵測區域的 HOG 特徵時，首先使用較為快速的 AdaBoost 分類器來篩選輸入的資料做篩選，通過篩選的區域將會由精準的 SVM 分類器來進行分類判斷是否為行人。最後為了減少因外在因素使得分類器失效對系統的影響程度，我們使用 camshift 的方式對偵測的結果做追蹤，主動去找尋偵測失敗的行人。最後整體系統能夠於背景單純的情況下達到 89% 的偵測率；於複雜背景的情況下可以達到 70% 的偵測率。

關鍵字

行人偵測 ； 特徵辨識 ； 機器學習 ； 影像處理 ； 輔助駕駛系統

並列摘要

Flowing the growth of economics, the amount of vehicles is rapidly increased and then the traffic accidents and consequentially piled up. Thus the development of vehicle collision avoidance system becomes more and more important. In urban areas, there are lots of pedestrians, bicycles, and motorcycles, thus the detection of the pedestrian-like bikes is the most important task. In this study, we proposed a pedestrian detection and tracking system using monocular camera to help drivers avoiding pedestrian traffic collision.
In the proposed system, we first compute the gradients of image pixels; then decompose every gradient to two adjacent two directions of nine fixed directions. Third, we construct Histograms of oriented gradient (HOG) features from the processed gradients to detect pedestrians. We have constructed thousands of HOG features; we use an AdaBoost strong classifier composed of sixteen weak classifiers to filter out the background and non-pedestrian features, and then use a precise SVM classifier to detection pedestrians based on the remain features. Finally, we use camshift method to find the failed detected pedestrians to achieve a higher detection rate. Overall system’s detection rate can achieve 89% in the simple background case and 70% in the complex background case.

並列關鍵字

pedestrian detection ； pattern recognition ； AdaBoost ； SVM ； tracking ； HOG

參考文獻

[1] Agarwal, S., A. Awan, and D. Roth, ''Learning to detect objects in images via a sparse, part-based representation,'' IEEE Trans. Pattern Analysis and Machine Intelligence, vol.26, no.11, pp.1475-1490, 2004.

[3] Ahonen, T., A. Hadid, and M. Pietikinen, ''Face description with local binary patterns: application to face recognition,'' IEEE Trans. Pattern Analysis and Machine Intelligence, vol.28, no.12, pp.2037-3041, 2006.

[4] Alonso, I. P., D. F. Llorca, and M. ?. Sotelo, ''Combination of feature extraction methods for SVM pedestrian detection,'' IEEE Trans. Intelligent Transportation System, vol.8, no.2, pp.292-307, 2007.

[6] Bertozzi, M., A. Broggi, M. Del Rose, M. Felisa, A. Rakotomamonjy, and F. Suard, ''A pedestrian detector using histograms of oriented gradients and a support vector machine classifier,'' in Proc. IEEE Conf. Intelligent Transportation Systems, Seattle, Washington, Sept.30-Oct.3, 2007, pp.143-148.

[7] Cao, X.-B., H. Qiao, and J. Keane, ''A low-cost pedestrian-detection system with a single optical camera,'' IEEE Trans. Intelligent Transportation Systems, vol.9, no.1, pp.58-67, 2008.