人机交互手势的超声波检测及其HMM融合SVM识别算法

Ultrasonic detection of human-computer interaction gesture and its recognition algorithm based on fusion of HMM and SVM

随着智能装备的性能提升和普及应用,经常采用手势进行人机交互。为了更有效地解决复杂手势识别精度不理想的问题,提高对复杂手势的识别准确率,提出利用超声波多普勒频移的SVM-HMM手势识别算法对提取到的手势特征序列进行识别分类。该算法采用支持向量机SVM改进隐马尔可夫模型HMM中的状态转移概率矩阵,并经由Sigmoid函数处理状态序列中各个隐状态的输出概率,对HMM的分类性能进行优化。利用设置的三种不同分类器的对比实验结果表明,该改进算法具有较好的提升效果,尤其对于复杂手势的识别效果提升更为明显。通过实验结果表明,SVM-HMM的算法能够准确进行手势识别,总体手势的识别率为94.625%,相比未改进的HMM平均识别率提高了10.75%,对比其他改进HMM算法对复杂手势的识别准确率提升4%左右。

With the performance improvement and popularization of intelligent equipments,gestures are often used for human-computer interaction.In order to improve the recognition accuracy of complex gestures,an SVM-HMM gesture recognition algorithm based on ultrasonic Doppler shift is proposed to identify and classify the extracted gesture feature sequences.In this algorithm,the support vector machine(SVM)is used to improve the state transition probability matrix in the hidden Markov model(HMM),and the output probability of each hidden state in the state sequence is processed by Sigmoid function to optimize the classification performance of HMM.A contrastive experiment of three different classifiers was set up.The experimental results show that the improved algorithm has a better recognition effect,especially for the complex gestures,and can accurately recognize the gestures.Its overall recognition rate for gestures is 94.625%,which is 10.75%higher than the average recognition rate of the unimproved HMM.Its recognition accuracy of complex gestures is increased by about 4%in comparison with other improved HMM algorithms.