Abstract: Dynamic gesture recognition is one of the most important topics in human-computer interaction. Compared with static gestures, the change of dynamic gestures is more complicated. The full extraction and description of dynamic gesture features is the key to accurately identify dynamic gestures. In order to solve the problem of insufficient description of dynamic gesture features and low recognition efficiency, this paper proposes a feature sequence that can fully describe dynamic gestures and realize effective description of dynamic gestures. High-precision Leap Motion sensor is used to acquire hand parameters diametrically. A sequence feature including fingertips angles, fingertips distances, fingertips elevations and palm speed is proposed. In addition, the sequence is combined with Long Short-Term Memory（LSTM） network model for dynamic gesture recognition. And by analyzing the experimental results on the dynamic gesture dataset, a dataset containing 16 dynamic gestures is collected by Leap Motion, which contains complex finger changes and hand movements. The dataset is named LM-16. The experimental results show that the recognition accuracy of this method for the LM-16 dataset is 98.50%. Furthermore, experiments are carried out by using other feature sequences to compare the influence of feature sequences on dynamic gestures recognition.

Key words: dynamic gesture recognition, Leap Motion sensor, feature extraction, Long Short-Term Memory（LSTM）

摘要： 动态手势识别作为人机交互的一个重要方向，在各个领域具有广泛的需求。相较于静态手势，动态手势的变化更为复杂，对其特征的充分提取与描述是准确识别动态手势的关键。为了解决对动态手势特征描述不充分的问题，利用高精度的Leap Motion传感器对手部三维坐标信息进行采集，提出了一种包含手指姿势和手掌位移的特征在内的、能够充分描述复杂动态手势的特征序列，并结合长短期记忆网络模型进行动态手势识别。实验结果表明，提出的方法在包含16种动态手势的数据集上的识别准确率为98.50%；与其他特征序列的对比实验表明，提出的特征序列，能更充分准确地描述动态手势特征。

关键词: 动态手势识别, Leap Motion传感器, 特征提取, 长短期记忆网络（LSTM）