Abstract: In order to locate pedestrian ahead of vehicle faster and more accurately, this paper proposes a pedestrian detection method based on Histograms of Oriented Gradients（HOG） of Region of Interest（ROI）. The feature is extracted by Sobel masks in the ROI where the pedestrian’s legs may exist, which is helpful to decrease the dimension of feature vector and simplify the calculation. Meanwhile, through the camera calibration only half image is scanned and the HOG features is constructed in blocks of the whole image. In this way, both the interference of complex background and the detection process are simplified significantly. Finally, multiple detection results are fused by an algorithm based on the symmetry of vertical edges of legs. Experimental results indicate that this method can achieve an effective accuracy with lower process time.

Key words: vehicle collision safety, pedestrian detection, Histograms of Oriented Gradients（HOG）, Region of Interest（ROI）, edge symmetry

摘要： 为快速定位车辆前方的行人，提出一种基于腿部感兴趣区域梯度方向直方图（HOG）特征的行人检测方法。将可能存在行人腿部的区域作为感兴趣区域，采用Sobel算子增强腿部垂直边缘特征，并提取梯度方向直方图特征，有效地降低了特征向量的维数；在检测过程中仅扫描可能存在行人腿部的图像下半部分，并在整幅图像的块内计算HOG特征，减少了复杂背景对行人检测干扰，进一步简化了检测过程；基于垂直边缘对称性特征对检测结果进行融合。实验结果表明，该算法能在保持检测率的同时提高检测速度。

关键词: 汽车碰撞安全, 行人检测, 梯度方向直方图, 感兴趣区域, 边缘对称性