在线时空关联的人体检测跟踪一体化方法

Integration Person Detection and Tracking Method Based on Online Spatial-Temporal Correlation

视觉人体检测跟踪一体化技术在很多领域都有着重要的应用价值,其关键技术有:鲁棒的目标检测技术、稳定的路径关联技术。主流的路径关联技术只考虑了目标的时间相关性,没有同时兼顾目标的时空关联属性,当相似目标距离太近时,容易导致错位跟踪。本文提出基于深度神经网络学习和结构化在线学习算法联合的目标检测跟踪一体化方法,通过构建结构化在线学习模型,建立目标时空关联关系。实验验证了所提方法可以有效抵制相似目标的干扰问题,并可提升对严重遮挡目标的跟踪能力。

The person detection and tracking integration technology has important application in many fields. Its key technology include: robust target detection and path correlation technology. The mainstream path correlation technology only considers the time correlation of the target, and does not take into account their spatial-temporal association property. When similar targets are too close, it is easy to lead to misalignment tracking. This paper proposes an integrated method of target detection and tracking based on deep neural network learning and structured online learning algorithm. The method can establish the target spatial-temporal relationship by constructing a structured online learning model. Experimental results show that the proposed method has resisted the interference problem of similar targets, and improved the ability of resisting serious occlusion.