一种基于梯度方向直方图的俯视行人的检测方法

Zenithal Pedestrian Detection Based on Histogram of Oriented Gradient

目前已有很多关于行人检测方面的研究,这些研究基本建立在行人竖直站立或行走的平视图上,主要应用于视频监控和车载辅助驾驶等领域,但在实际应用中,有时需要从不同的视角检测行人.文中提出一种针对俯视行人检测方法,该方法将俯视行人头部的梯度方向直方图统计信息作为检测目标的特征.通过训练样本提取的特征向量在支持向量机中进行训练得到分类模型参数,然后提取检测样本的特征向量输入分类模型进行判别.与现有行人检测的梯度方向直方图算子相比,文中特征描述算子突出目标的区域与轮廓特征,在目标分块、特征计算和特征统计方法上均有变化.实验证明算法有效且处理速度明显提升.

There are extensive researches on pedestrian detection, which mostly suppose visible humans observed in fiat view and are applied in video surveillance, driving assistance etc. However, sometimes pedestrian detection from another perspective should be considered in practice. In this paper, a histogram of oriented gradients （HOG） descriptor is introduced for the zenithal pedestrian head detection. The vectors extracted from training samples are trained in the support vector machine to get the classifier parameters, and then the vectors of test samples are input into the classifier to discriminate which targets are. Compared with the existing methods, the proposed descriptor highlights both the region and the contour of the object. Partitioning blocks are reformed, and the feature calculation and statistical method are changed adaptively to the task. The experimental results show that the proposed method is effective and can be applied to count pedestrians in vertical view with a faster processing speed.