摘要
为解决复杂场景中目标检测实时性差和鲁棒性低问题,提出了一种基于软级联支持向量机(SVM)分类器的行人检测算法。该算法采用梯度方向二值模式(ORBP)为特征描述子,基于自适应特征选择与多级分类阈值构建软级联分类器。为确保样本选取的完备性,通过模糊估计随机构建正负样本集,结合快速特征点与中值流实现目标追踪。试验结果表明,在复杂场景中,该算法实时性优且鲁棒性高。
To solve the target detection problems of poor real-time performance and low robus t -ness in complex scene, an algorithm for the pedestrian detection is proposed based on soft cascade support vector machine （SVM） classifier. The oriented gradient rotated binary pattern （ORBP） feature is regarded as the image feature descriptor in the algorithm. Based on chosen adaptive fea-ture and multistage cascaded thresholds, a soft cascade classifier is constructed. To ensure the ra-tionality of sample selection, positive and negative sample sets are randomly constructed by fuzzy evaluating, and the target tracking is realized combined with the fast feature spot and the median- flow. Experimental results show that the algorithm has the excellent real-time performance and the high robustness in complex scene.
出处
《指挥信息系统与技术》
2017年第5期104-108,共5页
Command Information System and Technology
关键词
行人检测
支持向量机
软级联分类器
梯度方向二值模式
pedestrian de te c t ion
support vector machine （SVM）
soft cascade classifier
oriented gradient rotated binary pattern （ORBP）
