【摘要】 相较于依赖其他传感器的手势交互，基于视觉的手势交互不需要额外穿戴外部设备、成本较低，成为了最自然的人机交互方式之一。针对增强现实电子沙盘中特定的手势识别要求，论文使用深度相机获取视频帧，利用CNN提取的特征和Hu矩融合，并结合LSTM对动态手势进行识别。实验显示，方法较单特征手势识别方法有更高的准确性，并且在不同光照情况下具有鲁棒性。针对增强现实电子沙盘中特定的九种动态手势，在不同环境下平均实时识别率达到91.57%。更多还原

【Abstract】 Compared with gesture interactions that rely on other sensors,visual-based gesture interactions are one of the most natural ways of human-computer interaction because they do not require the need to wear additional external devices and are less expensive. In view of the specific gesture recognition requirements in augmented reality electronic sand table,this paper uses a depth camera to obtain video frames,combines features extracted by CNN with Hu moment,and identiies dynamic gestures in conjunction with LSTM. Experiments show that the method is more accurate than the single-feature gesture recognition method,and it is robust in different lighting situations. For the nine dynamic gestures in the augmented reality electronic sand table,the average real-time recognition rate has reached 91.57% under different environments.更多还原