Researches For Robot Touch Gesture Recognition Based On Three-way Decisions

Touch behavior is one of the important non-verbal forms of social interaction like visual cues such as facial expressions and body gestures,which is important in interpersonal interaction.In HRI,touch can transfer additional informations merged with other modalities,such as audio and visual.One of the application is the robot therapy which has great social significance.However,the current researches on the affective robot focuse on vision and audition,tactile touch has not caused enough attention.In this paper,the research of robot touch gesture recognition is working on the data set: CoST.The work is carried out from three aspects: data preprocessing,feature extraction and classification model.The main contents are as follows:First of all,based on the analysis of the data set CoST,two preprocessing methods-“cutout” and “removing background” are proposed,and features are extracted from six perspectives,including basic features,histogram-based features,sequence features,gradient-based features,contact area features and channel-based features.Experiment is carried out on the training set with ten-fold cross validation using random forest as classifier.The results show that different preprocessing has different effects on the recognition of different gestures.Secondly,in order to combine the advantages of different preprocessing for different gesture recognition,an algorithm of touch gesture recognition based on three decision is proposed.According to the idea of sequential three-decisions,the classifier trained by different data sets is used as the decision model under different granularity,and each layer is decided according to the rules of three-way decisions.The algorithm includes three steps: changing an m-category classification problem into an m two-category classification problem and calculating the thresholds of three-way decisions for every two-category classification,choosing the final predictive value in the decision results.The experimental results show that the algorithm improves the recognition rate to some extent.Finally,based on the improvement of the previous algorithm,an ensemble classifier based on three-way decisions is proposed.According to the idea of the ensemble classifier,the prediction results of the classifiers trained by different data sets are combined with Bayes voting.In the process of the combination,three-way decisions is introduced and let the vote base on three-way decisions.Simulation results show that current algorithm has better classification effect than the previous algorithm,and then compared with other researchers’ work,the algorithm in this paper achieves a more accurate classification to some extent.