Abstract: The recognition algorithm based on outline feature can deal with the appearance changes of hand gestures very well, but the accuracy and robustness of the existing gesture recognition is low when the target is rotated and scaled. A new gesture recognition method based on finger tip and Hu invariant moments is proposed for the problem above. This method improves the fingertip detection based on fingertip curvature, which enhances robustness of feature extraction from the finger tips. It??also?improves the global description of gesture model by f-using hand fingertips and Hu invariant moments. Besides, it uses recognition algorithm in which new gestures can be added automatically to improve feasibility of the algorithm. Experiments show that the proposed method can effectively improve the accuracy and robustness of gesture recognition without reducing efficiency.

Key words: hand gesture recognition, fingertip detection, curvature calculation, Hu invariant moment

摘要： 基于轮廓的识别算法能够很好地处理手势的外观变化，但现有的识别算法在目标旋转和缩放时其识别率和鲁棒性较低，针对这一问题提出了一种基于指尖点和Hu不变矩的手势识别方法。该方法对基于曲率的指尖检测方法进行改进，增强指尖点特征提取的鲁棒性；融合Hu不变矩和指尖点特征，提高手势模型的全局描述性；利用能够自动添加新手势的识别算法，提高算法的实用性。实验表明，该方法在满足实时性的基础上有效地提高了手势识别的准确性和鲁棒性。

关键词: 手势识别, 指尖检测, 曲率计算, Hu不变矩