基于时空上下文的手势跟踪与识别

Gesture tracking and recognition based on spatio-temporal context

针对复杂背景下手势运动过程中出现的手势形态变化、遮挡、光照变化等问题,提出了一种基于时空上下文的手势跟踪与识别方法。使用机器学习方法离线训练手势样本分类器,实现对手势的检测和定位;利用时空上下文跟踪算法对动态手势进行跟踪,同时为了避免跟踪过程中出现的漂移、目标丢失等情况,使用手势检测算法对手势位置信息进行实时校准;根据手势运动轨迹对手势运动进行跟踪与识别。实验表明,提出的方法可以实现对手势运动快速、准确、连续识别,满足人机交互的要求。

A gesture tracking and recognition method based on spatio-temporal context is proposed to solve the problems such as gesture posture variation, occlusion, illumination changes under the complex background. Firstly, in order to get the location information of gesture, machine learning is used for off-line training a gesture classifier; then spatio-temporal context algorithm is used to track dynamic gesture. At the same time, the gesture detection algorithm is used for real-time calibration to the gesture position information in order to avoid drift, target lost in the tracking process. Finally, according to the gesture trajectory tracking, hand movements can be identified. Experimental results show that the proposed algorithm can achieve rapid, accurate and continuous recognition of the gesture motion and meets the requirements of the humancomputer interaction.