基于核相关滤波的飞行员手势跟踪研究

结合飞行员操作过程中手部的运动特点,针对核相关滤波器跟踪算法在目标尺度变化和快速移动时存在的问题,提出了一种结合了手势检测与核相关滤波跟踪算法的飞行员手势跟踪方法。首先,将肤色分割与背景减除进行有效的融合,对静态手势进行快速检测。然后,使用KCF算法对定位的手势区域进行跟踪以获取手势实时位置和轨迹。跟踪过程中比较手势检测与手势识别所得到的手部信息,当出现偏差时对搜索框进行更新。从而实现有效的手势检测与跟踪。实验结果表明,该方法可以实现对飞行员变形手势快速、准确的实时跟踪,对部分遮挡和尺度变化有很好的适应性,能够满足后期飞行员手部行为分析的要求。

Pilots gesture recognition and tracking based on Kernelized Correlation Filters

As Kernelized correlation filter is difficult to deal with the problems of the scale changes and target,considered of the characteristics of the pilot"s hand movement,we propose a pilot hand tracking method based on Kernel Correlation Filter combined with hand gesture recognition method. Firstly, this proposed algorithm effectively fuses the results from skin color detection and background subtraction based on Gaussian mixture model. Secondly, get the location information of gesture based on a kernelized correlation filter (KCF) tracker. In the course of tracking, the hand information given by gesture detection and gesture recognition is compared, and the search frame is updated when there is a deviation. Experimental results show that the proposed algorithm can achieve rapid, accurate and continuous recognition of the pilot gesture motion and can deal with the problems of scale changes of the target. It can be used for the analysis of pilot behavior.