改进线性外推法预估器的手势跟踪

Gesture Tracking Using Improved Linear Extrapolation Predictor

针对动态手势识别系统手势跟踪问题,提出了一种基于改进线性外推法预估器的手势跟踪算法.该算法用前两帧的平均位移作为未来帧的位移预测,提高了预测精度;采用5点直线拟合,根据拟合直线斜率判断目标遮挡和重叠状态下的运动方向,克服了由于手势目标质心变化引起的预测位置偏离实际目标的缺陷.实验结果表明:所提出的算法能准确稳定地跟踪手势目标,平均预测偏差缩小到3.374像素,并且能在手势被遮挡和手势重叠的情况下实现有效跟踪.

To improve accuracy of gesture tracking in dynamic gesture recognition system,a gesture tracking algorithm using an improved linear extrapolation predictor is proposed.Specifically,to improve prediction accuracy,the algorithm uses the average displacement of two previous frames as the future-frame predictor.Besides,to deal with occlusion and hands-overlap,the target movement direction is determined based on the slope of fitting line with five points.Thus deviation between the predicted and actual positions caused by the changing gesture centroid is reduced.Experimental results show that efficiency of the proposed gesture tracking method with the average prediction deviation is reduced to 3.374 pixels.In addition,even in the case of occlusion and hands-overlap,the gesture target can also be tracked effectively.