一种基于字典学习的动态手势识别方法

A Dynamic Gesture Recognition Method Based on Dictionary Learning

为提高动态手势的实时性和识别率,提出一种基于字典学习的识别方法,该方法直接对跟踪到的手势轨迹进行操作,省却了传统方法的特征提取过程。其核心是将动态手势的分类识别问题转化为训练样本的字典构造和学习问题,并基于类别的字典学习方式获得一个经过优化的过完备字典,实现了可随时根据需要进行手势样本类别的添加或删除,从而降低了计算成本和识别时间,提高实时性。考虑到在识别中不同的动态手势甚至是相同动态手势的不同实例都有可能具有不同的动作幅度和速度这一难点,为每一类动态手势添加一个唯一的类别标签并通过预先设定的判别性稀疏编码将标签信息和字典的原子项紧密的联系在一起,使得来自同一类别的动态手势具有相似的稀疏编码,而来自不同类别的动态手势具有不同的稀疏编码,从而提高了动态手势识别的准确率。在自定义的10类动态手势集上进行了测试,实验结果表明,与现有的一些识别方法相比,本方法具有更高的识别率和更快的识别时间。

In order to improve the real-time performance and recognition rate of dynamic gestures,a recognition method based on dictionary learning is proposed,which directly operates on the tracked gesture trajectory and saves the feature extraction process of the traditional method.Its core is to transform the classification and recognition of dynamic gestures into dictionary construction and learning of training samples,and to obtain an optimized over-complete dictionary based on category-based dictionary learning,which can add or delete gesture sample categories as needed at any time,thus reducing the calculation cost and recognition time and improving the real-time.Considering the difficulty that different dynamic gestures and even different instances of the same dynamic gesture may have different motion amplitudes and speeds in recognition,a unique category tag is added to each category of dynamic gestures and the tag information is closely linked with the atomic items of the dictionary through pre-set discriminatory sparse coding,so that dynamic gestures from the same category have similar sparse coding while dynamic gestures from different categories have different sparse coding,thus improving the accuracy of dynamic gesture recognition.Tests were carried out on a custom set of10types of dynamic gestures,and the experimental results show that this method has higher recognition rate and faster recognition time than some existing recognition methods.