多特征融合且抗遮挡的长时目标跟踪算法  

Multi-feature fusion and anti-occlusion long-term object tracking algorithm

作　　者：吴涛 袁亮[1] WU Tao;YUAN Liang(School of Mechanical Engineering,Xinjiang University,Urumqi 830047,China)

机构地区：[1]新疆大学机械工程学院,新疆乌鲁木齐830047

出　　处：《计算机工程与设计》2021年第1期226-231,共6页Computer Engineering and Design

基　　金：国家自然科学基金项目(61662075);新疆维吾尔自治区重点研发任务专项基金项目(2018B02011);自治区自然科学基金项目(2019D01C021)。

摘　　要：针对目标跟踪过程中的遮挡、形变以及长时跟踪等问题进行研究,提出一种多特征融合且抗遮挡的长时目标跟踪算法。以判别尺度空间(DSST)算法为框架,融合颜色空间特征,引入APCE指标,增强目标位置的预测和抗遮挡能力,提高算法的鲁棒性;增加随机蕨分类器检测机制,在跟踪失败时对目标进行重新检测定位;在模型更新阶段,利用帧差法调整模型的更新率。实验结果表明,改进算法在目标遮挡、形变以及长时跟踪等复杂情况下的跟踪性能均优于其它经典算法。Aiming at the problems of occlusion,deformation and long term tracking in the process of object tracking,a multi-feature fusion and anti-occlusion long term tracking algorithm was presented.The discriminating scale space(DSST)algorithm was taken as the framework.The color space features were integrated,the APCE index was introduced to enhance the prediction and anti-occlusion ability of the target position,and the robustness of the algorithm was improved.The detection mechanism of random fern classifier was added to redetect and locate the target when the tracking failed.The update rate of the model was adjusted using frame difference method.Experimental results show that the improved algorithm has better tracking performance than other classical algorithms under complex conditions such as object occlusion,deformation and long-term tracking.

关 键 词：相关滤波 抗遮挡 长时跟踪 特征融合 随机蕨分类器 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]