欢迎访问《图学学报》
分享到：
摘要：基于计算机视觉的连续手势识别因为其自然性和便捷性在大型互动娱乐、互动教
育等方面得到了广泛应用。在连续手势识别过程中，解决手势分割问题的已有方案多存在计算
量大效率低的缺点；解决独立手势识别问题的已有方案多存在训练参数设定过程复杂的缺点。
针对这两个问题做出两点优化：其一，提出一种基于隐状态模式归一化的方法对连续手势进行
分割，提高了手势分割的效率；其二，提出一种基于参数自反馈调节的独立手势训练和识别方
法，降低了独立手势训练的难度，并提高了识别的精度。实验证明，提出的基于双重优化的连
续手势识别方法与原有方法相比在精度和效率上都有较大提升。
Abstract: The continuous gesture recognition based on computer vision has been successfully applied#br# in the field of large interactive entertainment and large interactive education. During the process of#br# continuous gesture recognition, the first key problem is gesture segmentation, the existing solutions#br# of this problem mostly have the disadvantage of low efficiency; the second problem is independent#br# gesture recognition, the existing solutions of this problem mostly have the disadvantage of#br# complicated training problem. A double-optimization approach for these problems is proposed. Firstly,#br# in order to improve the efficiency of hand gesture segmentation, a hidden state mode normalization#br# method is given for continuous gesture segmentation. Secondly, to reduce the complexity of#br# independent gesture training and improve recognition accuracy, a training method is presented based#br# on parameter self-feedback regulation for independent gesture recognition. Experiments have shown#br# that our methods greatly enhance accuracy and efficiency.
