移动摄像机环境下基于特征变换和SVM的分级行人检测算法

Feature Transformation and SVM Based Hierarchical Pedestrian Detection with a Monocular Moving Camera

本文提出了一种应用于单目移动摄像机的基于特征变换和支持向量机(Support Vector Machine,简称SVM)的coarse-to-fine行人检测方法.首先,用查找表(Look-Up Table,简称LUT)Gentle AdaBoost Cascade训练一个粗级的行人检测器.接着把粗级的行人检测器的每一段分别作为一个特征,并用能通过粗级行人检测器的正负样本来训练基于SVM的精密级行人检测器.最后,采用基于颜色和空间信息的时序分析方法进一步提高行人检测率和降低误报率.从实验结果可以看出我们算法的高效性.

In this paper, we propose a coarse-to-free pedestrian detection based on feature wansformation and SVM with a monocular moving camera.in this method,a coarse pedestrian detector is learnt by Look-Up Table （LUT） Gentle AdaBoost cascade. Then each stage classifier in coarse detector is taken as a feature, and a fine pedestrian detector based on those features is learnt with SVM from those training data which pass through the coarse pedestrian detector. The detection results by this detector are refined by temporal analysis using the color and spatial information of each detection results to improved pedestrian detection rote and decrease the false alarm rote. Experimental results show our method high performance.