作为前视红外成像末制导的关键部分, 红外目标跟踪是一个极具挑战性的课题。本文针对前视红外建筑物目标, 提出了一种基于协方差描述子稀疏表示的红外目标跟踪框架。首先, 提取红外建筑物目标的协方差描述子特征; 其次, 由于协方差描述子属于黎曼空间, 采用 log-Euclidean变换将其转换到欧式空间; 最后, 在粒子滤波的理论框架基础上, 采用目标在字典中的稀疏表示作为观测模型, 对红外建筑物目标进行表示, 通过贝叶斯状态推理框架进行目标跟踪。对前视红外建筑物目标的跟踪实验表明, 该方法在跟踪准确度及鲁棒性方面体现出了优良的特性。

As the key component of forward-looking-infrared(FLIR) image terminal guidance, infrared object tracking is a challenging task. In this paper, a FLIR building object tracking framework based on sparse representation of covariance descriptor(Cov) is proposed. First, the Cov of FLIR building is extracted and then transformed to Euclidean space due to the reason that Cov lies in Riemannian space. Then, based on particle filter theory, the observation model of object is represented through sparse representation of template dictionary, and object tracking is continued by using a Bayesian state inference framework. Experiments on FLIR building object show that the proposed method obtains effectiveness in tracking accuracy and robustness.