摘要
本文所提算法是一种卷积神经网络与时空上下文结合的目标跟踪算法。将卷积神经网络算法融入时空上下文算法框架下,使得跟踪系统整体的鲁棒性有显著提高。引入Kalman滤波来处理目标被严重遮挡时,跟踪框容易漂移的问题。此外,整个跟踪系统采取由粗到精的双重目标位置定位方式,由时空上下文算法实现目标初定位,由卷积神经网络进行目标位置的精确定位。经实验验证,算法不仅稳定性和鲁棒性较好,而且实时性的条件也基本满足。
In this paper, an algorithm of the target tracking combining convolution neural network with the temporal and spatial context is proposed. In the framework of the context-based algorithm, the convolutional neural network algorithm is integrated to improve the stability and robustness of the tracking system. The Kalman filter is introduced to deal with the problem that the target is obscured. In addition, the whole tracking system adopts a coarse-to-fine target location method, and the target localization is achieved by the temporal and spatial context algorithm, and the target location is accurately located by the convolution neural network. Experimental results show that the proposed algorithm is stable and robust for real-time performance.
出处
《红外技术》
CSCD
北大核心
2017年第8期740-745,共6页
Infrared Technology
关键词
目标跟踪
时空上下文
卷积神经网络
target tracking, spatio temporal context, convolution neural network
