边界约束的核密度估计红外人体目标跟踪方法

基于核密度估计的均值漂移算法因其良好的实时性而被广泛地应用于目标跟踪,但传统的均值漂移算法极易因颜色等信息的缺乏而使跟踪不稳定,且目标尺度的变化也不利于目标位置的准确估计,为此,提出了一种具有边界约束的均值漂移红外人体目标跟踪新方法。该方法通过各向异性扩散,并联合红外图像的梯度与亮度信息来获取目标边界,自适应调整核窗宽,从而利用均值漂移策略进行红外人体目标跟踪。实验结果表明,该方法在红外人体目标尺度改变时仍能实现良好的跟踪。

Infrared pedestrian target tracking based on KDE with boundary constraint

As mean shift algorithm based on KDE has good performance of real-time,it has been widely used in target tracking.However,the tracking robust of traditional mean shift algorithm is often depended on such features like color,etc.Moreover,the tracked position is usually affected by the scale change of target during tracking procedures.To overcome these disadvantages,a new infrared pedestrian target tracking approach based on mean shift with boundary constraint is proposed.This method uses the gradient of infrared image processed by anisotropic diffusion.Particularly,the target boundary is obtained by its gradient as well as brightness information,kernel bandwidth is adaptively adjusted.At last,the infrared pedestrian target tracking is carried out by the strategy of mean shift algorithm,and the experimental results show that the proposed approach can achieve efficient tracking when the scale of the target changes.