基于自学习稀疏表示的动态手势识别方法

针对加速度传感器的手势采集方式提出一种基于自学习稀疏表示的动态手势识别方法。该方法将分类识别问题转化为求解待识别样本对于训练样本的稀疏表示问题，直接对原始加速度信号进行操作，省去了特征提取过程，可方便地添加新的手势类别和删除已有的手势类别；利用面向类别的字典学习，来寻求一个较小的并经过优化的超完备字典来计算待识别样本的稀疏表示，从而大大缩减算法的计算复杂度，满足实时性要求。在包含18种手势的3 000多个样本的公开数据集上进行测试，实验结果验证了该方法的有效性。

Gesture recognition approach based on learning sparse representation

An approach of robust accelerometer-based hand gesture recognition based on self-learning sparse representation was proposed. This method operated directly on the original acceleration signals by sparse representation without feature extraction and used the class-specific dictionary learning for sparse modeling to reduce the computing cost and time of recognition. The proposed system can easily add a novel gesture category as well as remove existing ones. Experiments on real-world database of 18 hand gestures validate the availability of the proposed algorithm.