摘要
针对行人被遮挡、汽车与行人相对速度较大使得汽车行人跟踪算法无法继续跟踪的问题,设计实现了一种改进的抗遮挡汽车行人稳定跟踪方法。即在核相关滤波器(KCF)跟踪算法的基础上引入方向梯度直方图(HOG)和尺度不变特征变换(SIFT)特征。当视频帧中的目标行人没有出现或部分出现时,提取当前帧的SIFT特征与HOG特征,采用主成分分析(PCA)降维技术进行多特征降维,通过串行组合的方式来构造HOG和SIFT融合特征集,生成HOG-SIFT特征模板。利用形成的特征模板与下一帧中的HOG-SIFT特征进行匹配,对行人进行快速重新定位跟踪。视频序列对比实验结果表明:改进算法较跟踪-学习-检测(TLD),CSK和KCF算法跟踪准确度分别提高了24.1%,15.8%,2.2%。
Aiming at the problem that the automobile and pedestrian tracking algorithm can not continue to track because of the pedestrian is sheltered and the high relative speed between automobile and pedestrian,an improved anti-occlusion automobile and pedestrian tracking method is designed and realized,which introduces the histogram of oriented gradient(HOG)and scale-invariant feature transform(SIFT)feature on the basis of kernel correlation filter(KCF)tracking algorithm.When the target pedestrian in the video frame does not appear or partially appears,the SIFT feature and HOG feature of the current frame are extracted.Principal component analysis(PCA)dimension reduction technology is used to reduce the dimension of multi-feature.Then the fusion feature set of HOG and SIFT is constructed by serial combination,and the feature template HOG-SIFT is generated.By matching the generated feature template with the HOG-SIFT feature in the next frame,pedestrians can be quickly repositioned and tracked.The result of video sequence comparison experiment show that,compared with TLD,CSK and KCF,the tracking accuracy of this algorithm is improved by 24.1%,15.8%and 2.2%,respectively.
作者
张传伟
李波
杨萌月
曾虹钧
王睿
ZHANG Chuanwei;LI Bo;YANG Mengyue;ZENG Hongjun;WANG Rui(School of Mechanical Engineering,Xi’an University of Science and Technology,Xi’an 710054,China)
出处
《传感器与微系统》
CSCD
2020年第10期42-44,共3页
Transducer and Microsystem Technologies
基金
中国博士后科学基金面上资助项目(2018M633537)
陕西省自然科学基础研究计划资助项目(2018JQ5086)
陕西省教育厅科学研究计划资助项目(2018JK0510)。
关键词
行人跟踪
特征融合
主成分分析
核相关滤波
抗遮挡
pedestrian tracking
feature fusion
principal component analysis(PCA)
kernel correlation filtering
anti-occlusion
