基于尺度不变特征和位置先验的行人检测算法
发布时间：2019-07-18 16:26
【摘要】：基于相邻和非相邻特征(NNNF)行人检测算法,提出了一种方法来解决行人特征对尺度变化敏感的问题以及窗口误检的问题.首先,在NNNF基础上,设计了一种尺度不变的特征.由韦伯定理启发,该特征表示为两个相邻或非相邻区域的差分值与这两个区域特征和的比值.这种新的特征具有很强的尺度不变性.此外,还提出了基于行人位置先验的上下文信息,作为一种简单有效的后处理方法.在行人场景中,行人的高度与位置存在一定的映射关系.利用SVM(support vector machine)训练了行人高度关于行人位置的回归模型.该模型能有效地滤除那些行人高度与位置信息不符合回归模型的检测窗口.实验表明,相比于NNNF-L2和NNNF-L4,本文提出的方法在Caltech数据库的检测性能分别提高了2.90%,和2.28%,.同时,本文提出的方法也在所有基于非深度网络的行人检测方法中具有最好的检测性能,平均漏检率为14.56%,.
[Abstract]:Based on (NNNF) pedestrian detection algorithm with adjacent and non-adjacent features, a method is proposed to solve the problem that pedestrian features are sensitive to scale change and window error detection. First of all, on the basis of NNNF, a scale invariant feature is designed. Inspired by Weber's theorem, this feature is represented by the ratio of the difference score of two adjacent or non-adjacent regions to the sum of the characteristics of the two regions. This new feature has strong scale invariance. In addition, a context information based on pedestrian location priori is proposed as a simple and effective post-processing method. In pedestrian scene, there is a certain mapping relationship between pedestrian height and position. The regression model of pedestrian height about pedestrian position is trained by SVM (support vector machine). The model can effectively filter out the detection window where the pedestrian height and position information do not conform to the regression model. The experimental results show that the detection performance of the proposed method in Caltech database is improved by 2.90% and 2.28%, respectively, compared with NNNF-L2 and NNNF-L4,. At the same time, the method proposed in this paper also has the best detection performance among all pedestrian detection methods based on non-depth network, and the average missed detection rate is 14.56%.
【作者单位】： 天津大学电气自动化与信息工程学院;
【基金】：国家自然科学基金资助项目(61472274)~~
【分类号】：TP391.41
本文编号：2515969
[Abstract]:Based on (NNNF) pedestrian detection algorithm with adjacent and non-adjacent features, a method is proposed to solve the problem that pedestrian features are sensitive to scale change and window error detection. First of all, on the basis of NNNF, a scale invariant feature is designed. Inspired by Weber's theorem, this feature is represented by the ratio of the difference score of two adjacent or non-adjacent regions to the sum of the characteristics of the two regions. This new feature has strong scale invariance. In addition, a context information based on pedestrian location priori is proposed as a simple and effective post-processing method. In pedestrian scene, there is a certain mapping relationship between pedestrian height and position. The regression model of pedestrian height about pedestrian position is trained by SVM (support vector machine). The model can effectively filter out the detection window where the pedestrian height and position information do not conform to the regression model. The experimental results show that the detection performance of the proposed method in Caltech database is improved by 2.90% and 2.28%, respectively, compared with NNNF-L2 and NNNF-L4,. At the same time, the method proposed in this paper also has the best detection performance among all pedestrian detection methods based on non-depth network, and the average missed detection rate is 14.56%.
【作者单位】： 天津大学电气自动化与信息工程学院;
【基金】：国家自然科学基金资助项目(61472274)~~
【分类号】：TP391.41
【相似文献】
相关期刊论文 前6条
1 黄小兰;;比较几种挑选“最优”回归模型的指标[J];中国卫生统计;1988年04期
2 李君;;基于路径的多元标准回归模型及其应用[J];湛江师范学院学报;2006年06期
3 韩立强,姜兴序;机床主轴结构回归模型近似重分析优化设计[J];吉林工学院学报;1990年04期
4 李海涛;;重回归模型在软件开发预估工作量中的应用[J];电脑与电信;2011年03期
5 潘宝骏,张文彤,张锡斌,吴思英;以SPSS软件包拟合条件logistic回归模型的探索[J];海峡预防医学杂志;2002年06期
6 朱珏钰;;利用EXCEL实现计酬分配预测的回归分析[J];长沙大学学报;2010年02期
相关硕士学位论文 前2条
1 邓健康;基于级联回归模型的人脸配准研究[D];南京信息工程大学;2015年
2 李昊鑫;三维手姿态建模及其交互系统研究[D];电子科技大学;2016年
，本文编号：2515969
本文链接：https://www.wllwen.com/kejilunwen/ruanjiangongchenglunwen/2515969.html
