融合两类跟踪框架优点的红外目标跟踪算法

Infrared Target Tracking Based on Particle Filter and Mean Shift

摘要:

基于粒子滤波的算法模型框架,针对红外的目标成像与背景对比度低、背景复杂等问题,进行提取目标灰度特征,通过对系统概率的密度函数的采样集进行了预测和更新,来逼近系统的后验密度概率,初步确定目标位置,再融合均值漂移算法进行小区域精确搜索,确定目标位置；同时,通过调整均值漂移的算法和函数的带宽,对于红外目标有阻挡情况下的识别能够得到有效处理；通过仿真得到该模型具有算法高效、去除粒子的退化,并且对于有遮挡的红外目标能够进行实时稳定地跟踪。

Abstract:

According to problems of cluttered environment or the low contrast between infrared target imaging and background, we extract the gray characteristics of target and use the particle filter algorithm which recursively forecast and update the sample sets of the state space to approximate the posterior density probability. In order to further increase the accuracy of tracking, we fuse the Mean Shift algorithm to search for target in local area. The bandwidth of kernel function in Mean Shift algorithm can also deal with the target occlusion problem effectively. Experimental results indicate the proposed method can not only improve the efficiency of the algorithm and overcome the particle degeneracy phenomenon, but also has good performance in occlusion.