多特征融合的行人目标优选算法研究
|
摘要
行人检测是目标检测研究与应用中的热点和难点,在车辆辅助驾驶、智能视频监控和人体行为分析等领域中占据重要的位置。针对行人检测工程化中相似目标的优选问题,提出一种基于相似度的联合多特征提取方法。首先对目标图像进行高斯预处理,然后将目标图像一分为二,分别提取HOG-HOC特征并计算相似度,最后通过支持向量机SVM确定相似度的阈值,比较目标图像相似度与阈值的大小以确定最优目标。实验结果表明,该方法能有效的解决目标图像的优选问题。
Pedestrian detection is a hot and difficult point in the research and application of target detection. It occupies an important position in driver assistance, intelligent video surveillance and human behavior analysis in the fields. Aiming at the problem of similar target optimization in pedestrian detection project, a similarity combine multi-feature extraction method is proposed. First,the objects in the image was in progress the Gaussian pre-treatment. Then, target image is divided into two parts that extrracted HOG-HOC features and calculated the similarity respectively. Finally, through support vector machine(SVM) is applied to determine the similarity threshold, comparing target image similarity and the threshold value of the size to determine optimal objective.Experimental results show that the proposed method can effectively solve the optimization problem of the targeat image.
Pedestrian detection is a hot and difficult point in the research and application of target detection. It occupies an important position in driver assistance, intelligent video surveillance and human behavior analysis in the fields. Aiming at the problem of similar target optimization in pedestrian detection project, a similarity combine multi-feature extraction method is proposed. First,the objects in the image was in progress the Gaussian pre-treatment. Then, target image is divided into two parts that extrracted HOG-HOC features and calculated the similarity respectively. Finally, through support vector machine(SVM) is applied to determine the similarity threshold, comparing target image similarity and the threshold value of the size to determine optimal objective.Experimental results show that the proposed method can effectively solve the optimization problem of the targeat image.
引文
[1]张春凤,宋加涛,王万良.行人检测技术研究综述[J].电视技术,2014,38(03):157-162.
[2]张旭东.行人检测技术研究[D].西安:电子科技大学,2012.
[3]任汉俊.基于特征融合与深度学习的行人检测方法研究[D].无锡:江南大学,2018.
[4]N.Dalal,B.Triggs.Histograms of oriented gradients for human detection.In IEEE,2015.
[5]饶钦,谢刚.基于颜色自相似性和HOG特征的行人检测[J].小型微机计算机系统,2014(11):2582-2585.
[6]林海波,柯晶晶,张毅.结合粒子群寻优与遗传重采样的RBPF算法[J].计算机工程,2016,42(11):295-299.
[7]郝辉,李雪瑞,李亚雄.基于遗传算法的面目标瞄准点寻优问题研究[J].兵器装备工程学报,2016,37(01):128-131.
[8]Z.Li,S.Chang.Learning locally-adaptive decision functions for person verification.In Computer Version and Pattern Recognition,2015.
[9]W.Zheng,S.Gang.Reidentification by relative distance comparison[J].IEEE Trans,2013,35(3):653-688.
[10]Wang Xiaogang.Joint deep learning for pedestrian detection[C].In ICCV,2013.
[11]林建伟.基于深度学习的图像相似度度量及应用[D].济南:山东师范大学,2018.
[12]普亚如.基于CNN特征学习和SVM的极化SAR影像地物分类[D].西安:西安电子科技大学,2015.
[13]倪恺,肖志涛.基于梯度方向直方图的行人检测方法研究[J].电视技术,2013,3(5):65-70.
[14]房乐楠,何腾鹏,刘宇红.一种改进型PSO算法在SVM参数寻优中的应用[J].电子科技,2018,31(06):17-19.
[15]李林,张丽红.基于改进梯度方向直方图的多尺度的行人检测[J].计算机应用,2012,10(52):36-42.
[2]张旭东.行人检测技术研究[D].西安:电子科技大学,2012.
[3]任汉俊.基于特征融合与深度学习的行人检测方法研究[D].无锡:江南大学,2018.
[4]N.Dalal,B.Triggs.Histograms of oriented gradients for human detection.In IEEE,2015.
[5]饶钦,谢刚.基于颜色自相似性和HOG特征的行人检测[J].小型微机计算机系统,2014(11):2582-2585.
[6]林海波,柯晶晶,张毅.结合粒子群寻优与遗传重采样的RBPF算法[J].计算机工程,2016,42(11):295-299.
[7]郝辉,李雪瑞,李亚雄.基于遗传算法的面目标瞄准点寻优问题研究[J].兵器装备工程学报,2016,37(01):128-131.
[8]Z.Li,S.Chang.Learning locally-adaptive decision functions for person verification.In Computer Version and Pattern Recognition,2015.
[9]W.Zheng,S.Gang.Reidentification by relative distance comparison[J].IEEE Trans,2013,35(3):653-688.
[10]Wang Xiaogang.Joint deep learning for pedestrian detection[C].In ICCV,2013.
[11]林建伟.基于深度学习的图像相似度度量及应用[D].济南:山东师范大学,2018.
[12]普亚如.基于CNN特征学习和SVM的极化SAR影像地物分类[D].西安:西安电子科技大学,2015.
[13]倪恺,肖志涛.基于梯度方向直方图的行人检测方法研究[J].电视技术,2013,3(5):65-70.
[14]房乐楠,何腾鹏,刘宇红.一种改进型PSO算法在SVM参数寻优中的应用[J].电子科技,2018,31(06):17-19.
[15]李林,张丽红.基于改进梯度方向直方图的多尺度的行人检测[J].计算机应用,2012,10(52):36-42.