局部特征与全局特征结合的HMM静态手势识别

HMM Static Gesture Recognition Algorithm Based on Fusing Local Feature and Global Feature

针对静态手势识别问题,提出了一种综合考虑局部形状与全局轮廓的隐马尔科夫模型(HMM)静态手势识别算法。该算法提取局部形状熵特征与上层轮廓特征分别作为训练数据训练每类手势的HMM参数。测试时,先凭借局部形状熵特征得出初步识别结果,然后根据初步识别结果的模糊性,附加与局部特征互补的上层轮廓特征进行再识别,得出最终识别结果。实验结果表明,该算法对于形状差异占主导地位的手势库有很好的效果,并且将静态手势的空间序列模拟成时间序列使得静态手势识别具有空间尺度不变性;同时该算法合理控制特征维数,一定程度上弱化了HMM训练时间长的弊端,加快了识别的速度。

Focusing on the issue of static gesture recognition,a hidden markov model(HMM)static gesture recognition algorithm based on local and global contour shape was proposed.It extracts local features and upper contour shape entropy as training data of each type of gesture respectively to train its HMM parameters.While testing,the algorithm works with local shape entropy to obtain preliminary identification results,and then according to the fuzziness of preliminary identification,chooses whether it needs to work with upper contour feature,which is a kind of global characteristic,and complementary to the local characteristic to get the final result.The experimental results show that the algorithm has a good effect for gesture library in which shape difference is dominant.And the ideal simulating static spatial feature data into time series makes static gesture recognition have the space scale invariance.At the same time,reasonable data dimension has shortened the training time,and accelerated the speed of recognition.