特征融合与视觉目标跟踪 *

针对跟踪过程中各类图像特征分离背景和目标能力的变化 ,提出一种基于增量判别分析的特征融合算法。该算法首先计算各特征图像的似然图 ,然后通过增量判别分析计算各特征分类性能 ,得到相应权重 ,并在此基础上求取融合似然图 ,通过粒子滤波算法确定待跟踪目标状态。通过对可见光及红外成像视频序列的仿真表明,该算法对环境光照变化、视角变化以及局部遮挡等均具有一定的鲁棒性。

Likelihood map fusion for visual object tracking

For the ability of every kinds of features to separate the object from backgroud is dynamic during the process of tracking, this paper presented an ILDA based likelihood map fusion. Gave a tracking frame, calculated each likelihood map and obtained the weight of each through ILDA, then fused the likelihood map based on the weights. At last, acquired the state of the objecte through particle filter based on the fused likelihood. Examples based on optical and infrared frames show that the proposed tracking framework is robust to the partial occlusions, view-point and illumination variations.