整合具鑑別性及差異性的車輛局部影像特徵應用於後方車輛偵測

Visual Part-based Vehicle Detection by Integrating Different Discriminative Features for Rear Traffic Scenes

邱一航(Yi-Hang Chiu)

指導教授 ： 傅立成

共同指導教授 ： 蕭培墉(Pei-Yung Hsiao)

國立臺灣大學/電機資訊學院/資訊工程學研究所/碩士(2008年)

https://doi.org/10.6342/NTU.2008.01671

摘要

每年都有許多人因為交通意外而受傷或死亡。因此，本篇論文希望藉由影像視覺，偵測後方危險的車輛，以警告和保護駕駛者。這些危險的車輛，在不同的交通環境或天氣環境下的不同種類車輛，或因為相機的視角，造成後方車輛影像呈現部分可見或完全可見車輛的情況下，都必須偵測得到。本篇論文所提出的方法，以電腦視覺的方式，將車輛以一組車輛局部影像來描述，並且整合和選擇具鑑別性及差異性的車輛局部影像特徵，來進行後方車輛偵測。其中基於修改AdaBoost演算法，來整合差異性的車輛局部影像特徵，並找出具鑑別性的局部影像特徵。其演算法包含多個弱分類器(Weak Classifier)，弱分類器包含弱特徵描述分類器，對於描述局部車輛外觀的特徵之鑑別度做分類，以及弱特徵空間分類器，對於車輛局部區域與車輛中心的幾何關係之穩定度做分類。其中弱特徵描述分類器用來評估不同局部車輛特徵的外觀之鍵別性，以及特徵空間分類器來評估不同車輛局部特徵對於車輛中心的幾何關係之穩定度。由經修改過後的AdaBoost演算法來選擇不同車輛局部特徵。在影像空間下偵測車輛時，這些經學習與選擇的具差異性且具鑑別性的車輛局部區域，對於可能車輛的中心位置作投票，當圖像中為多數票數的區域即為車輛中心的位置。

關鍵字

車輛偵測 ； 後方車輛偵測 ； 物體局部偵測 ； 特徵選擇 ； 特徵整合

並列摘要

Every year many people are injured and killed in traffic accidents. In order to warn and protect the driver in the potential accidents, we develop a vision system to detect whatever style rear vehicles under various traffic or weather conditions. Otherwise, in the vision system, the incomplete rear vehicle is shown in the image due to the limitation of field of view of the camera. We propose a vision system detecting rear vehicles, whether occluded or not. The system selects and integrates different parts of the rear vehicle pattern; these parts are represented by complementary characteristics. There are two kinds of classifiers for the part selection. First, the appearances of each vehicle’s part with different feature types are evaluated by the representative classifiers. Second, the geometry of the parts is modeled as a spatial classifier; each spatial classifier evaluates the stability of vehicle location estimation of each part. Representative classifiers and the spatial classifiers are learnt and integrated according to the performance of each part, including the discriminabilities of appearance and the stabilities of the geometry estimation. The modified error rate function of a boosting algorithm selects parts of the rear vehicle with complementary discriminative features. These selected parts vote the possible vehicle’s location in the image; the majority votes locate the possible rear vehicles.

並列關鍵字

Vehicle Detection ； Rear Vehicle Detection ； Part-based Object Detection ； Feature Selection, Feature Integration

參考文獻

[2] Z. Sun, G. Bebis, and R. Miller, "Monocular precrash vehicle detection: features and classifiers," IEEE Transactions on Image Processing, vol. 15, pp. 2019-2034, 2006.

[5] X. Wen, H. Zhao, N. Wang, and H. Yuan, "A Rear vehicle Detection System for Static Images Based on Monocular Vision," in International Conference on Control, Automation, Robotics and Vision, pp. 1-4, 2006.

[6] S. Kim, S.-Y. Oh, J. Kang, Y. Ryu, K. Kim, S.-C. Park, and K. Park, "Front and rear vehicle detection and tracking in the day and night times using vision and sonar sensor fusion," in International Conference on Intelligent Robots and Systems, pp. 2173-2178, 2005.

[7] F. Jiang and X. Lin, "A Learning Based Approach for Vehicle Detection," in IEEE Region 10 Conference, TENCON, pp. 1-4, 2006.

[8] Y. Liu, Z. You, L. Cao, and X. Jiang, "Vehicle detection with projection histogram and type recognition using hybrid neural networks," in IEEE International Conference on Networking, Sensing and Control, pp. 393-398, 2004.