摘要
针对目前模型更新方法仅依赖于上一帧或最近帧定位到的目标信息,跟踪的历史信息未充分利用的问题,运用历史跟踪结果设计了多模板模型更新策略,并基于该更新模型结合卷积网络提出了一种新的运动目标跟踪方法。在运动目标跟踪测试视频集VOT2015下与目前热点运动目标跟踪方法对比实验表明:方法对于遮挡现象和目标自身形变具有较强的鲁棒性和较高的准确性。
Aiming at problem that current model updating method only relies on the target information from the previous frame or the recent frames,and the historical information is not fully utilized,a multi-template updating strategy is designed,and a new moving target tracking method is proposed based on this updated model and convolutional network. Compared with some state-of-art trackers in VOT2015 tracking benchmark dataset,the experimental results show that the method is robust and has high accuracy for occlusion phenomena and target deformation.
出处
《传感器与微系统》
CSCD
2018年第2期53-56,60,共5页
Transducer and Microsystem Technologies
基金
国家自然科学基金资助项目(61462052
31300938)
关键词
运动目标跟踪
多模板
卷积网络
遮挡
目标形变
moving object tracking
multi-template
convolutional network
occlusion
target deformation
