Abstract: Aiming at the limitations of pedestrian detection using single feature under complex background, this paper proposes an algorithm which fuses several features by using the elastic models and part quadrat-ic weighting. Fusing Histogram of Oriented Gradient（HOG）, skin color, hair color and curvature efficiently, it est-ablishes kinds of feature models which can be adaptive to pedestrian detection. The first level which adopts the improved HOG feature and is combined with elastic models uses the SVM classifier for the first det-ection. The second level extracts the Region Of Interest（ROI） of partial models to detect head skin color, hair color and curvature of the legs. The experimental result shows that the algorithm makes up for the inade-quacy of single feature. Meanwhile, the algorithm detects the whole and part key features of pedestrian effe-ctively and it improves the recognition performance.

Key words: Histogram of Oriented Gradient（HOG）, skin color, hair color, curvature, elastic models, part quadratic weighting

摘要： 针对复杂背景下采用单一特征进行行人检测时的局限性，提出了一种融合多种特征并运用模板弹性模型与局部二次加权的算法，将梯度直方图（HOG）、肤色、发色与曲率有效融合，建立了适用行人检测的各特征模型。第一级采用改进HOG特征结合模板弹性模型利用SVM分类器初次检测;第二级提取局部模板感兴趣区域（ROI）进行头部肤色、发色与腿部曲率检测。实验表明，该算法弥补了单一特征的不足，有效检测了行人整体与局部关键特征，提高了识别性能。

关键词: 方向梯度直方图, 肤色, 发色, 曲率, 弹性模型, 局部二次加权