基于粒子滤波的目标图像多特征融合跟踪方法

Method of Tracking Target Images Based on Multi-feature Fusion and Particle Filter

研究了序列图像中红外弱小目标的检测跟踪问题。基于多特征融合的小目标检测算法具有较好的检测性能和适应性,而粒子滤波则是一种处理非线性和非高斯动态系统状态估计的有效方法。结合两种算法的优点,提出了一种基于粒子滤波的目标图像多特征融合跟踪方法。从红外序列图像中提取了局部灰度均值对比度、局部梯度均值对比度、局部熵和灰度分布四个典型特征,根据各个特征对弱小目标检测的贡献,自适应地进行特征融合。在粒子滤波的框架下,将融合后的特征信息转化为粒子的权值,对红外弱小目标进行跟踪。仿真试验表明,该算法有着良好的检测与跟踪性能。

Detecting and tracking small targets in a infrared images sequence is studied. The small target detection algorithm based on multi-feature fusion is robust and efficient, and the particle filter is an effective method for the state estimation of non-linear and non-Gaussian dynamic systems. Therefore, a method of tracking target images based on multi-feature fusion and particle filter is proposed. Four features are extracted from image se- quences including local contrast mean difference, local average gradient strength, local entropy and grey-level distribution which fuse adaptively according to their contribution to small target detecting,The fused feature information is transformed into particle weights under the framework of particle filtering. And the proposed algorithm based on particle filter can successfully detect and track small target in non-Gauss and non-linear case. Simulation results show that algorithm has good performance of detecting and tracking.