抗遮挡的自适应运动目标跟踪方法

针对复杂场景中的目标遮挡问题,提出一种基于均值漂移(Mean shift)和轨迹校正的自适应目标跟踪方法.由于Mean shift迭代易陷入局部最优点,这里引入Kalman滤波器以预测和校正目标运动轨迹,并根据迭代轨迹误差校正协方差,使得跟踪器在多峰值非高斯分布的复杂环境下也能收敛到全局最优点.基于Bhattacharrya系数计算色彩x、y方向分量相似度,并根据邻帧分量相似度偏差自适应调整相似度融合权值.综合当前帧和前面帧作用更新目标运动状态、特征和尺度模型.实验结果表明提出的方法对于静态场景遮挡和目标间互遮挡、部分和全部遮挡下的目标跟踪均具有鲁棒的跟踪性能.

Adaptive method for tracking mobile object in occlusion

To track occluded object under complex scene,an adaptive method is proposed based on mean shift and trajectory rectifying.Since mean shift easily falls into local extremum,Kalman filter is utilized to estimate and rectify object moving trajectory with the noise covariances being corrected by iteration trajectory errors.It makes the tracker always converge at the global extremum even if there is multi-peak and non-Gaussian distribution under complex scene.Bhattacharrya coefficient is employed to compute the individual color similarity in x and y-directions,and fusion weights are obtained by calculating the similarity deviations in adjacent frames.The object state,feature and scale models are updated by integrating the current frame and foregoing one.Experimental results show that the proposed method is robust in tracking the objects under such cases as static scene occlusion,occlusion among objects,partial and total occlusion.