| Anxiety is a negative emotion when people realize that things will have bad consequences or face potential threats.Traditional anxiety and emotion analysis technology mainly adopts scale analysis and artificial evaluation,but all of them lead to subjective deviation due to the need to rely on human to evaluate and analyze,and mostly focus on the overall anxiety and emotional changes of the subjects.However,there are few studies on individual long-term anxiety analysis and follow-up evaluation for different individuals.Compared with traditional technology,the method of anxiety emotion analysis based on computer technology has the advantages of high efficiency and fast,but it is limited in the practical application due to the extraction of behavior information and the unclear quantitative mapping relationship between behavior information and anxiety emotion.In this paper,a method of speaker anxiety analysis and evaluation based on multiple action characteristics is established for the performance of the subjects in the speech situation.The quantitative mapping relationship between multi-dimensional action characteristics and anxiety level is constructed to realize the long-term anxiety analysis and tracking evaluation of the subjects.The specific research contents are as follows:(1)The quantitative mapping relationship between anxiety and multi-dimensional action was established.By integrating the psychological data of action behavior related to anxiety and emotion,the continuous emotion evaluation model of anxiety degree was constructed by using two kinds of anxiety evaluation indexes,i.(2)The intelligent recognition algorithm of limb and hand movements is designed.Based on Kinect SDK toolkit,the human skeleton information extraction code is compiled to extract the limb feature points.Through feature engineering design,16 kinds of limb motion features are obtained,and then a "1-to-1" SVM algorithm is established.Six limb movements are identified and classified.The average recognition accuracy rate is 98.9%;The contour of the hand was obtained by two skin color extraction and 5-layer convolution neural network was designed to identify 7 kinds of hand movements with obvious discrimination.Then,the hand skeleton data output by openpost was analyzed to identify three kinds of actions with high confusion.Finally,the joint algorithm is constructed to realize the recognition of 10 kinds of hand movements,with an average recognition accuracy of 98.4%.(3)Integrate and optimize long-term four-dimensional personalized data flow.Based on kinectsdk and Dlib development package,the identity information of the tested group is extracted,and a long-term personalized data stream is formed,which includes four dimensions:time information,individual identity,body anxiety and hand anxiety.The key frame is automatically screened by analyzing the long-time motion information,and the data flow is further optimized by voting.The verification shows that the key frame automatic extraction and optimization reduction of personalized data flow can be realized effectively when the key frame data loss rate is less than 3%.(4)The test system is built and the effect of anxiety analysis is verified.In the simulated speech environment,a test system is established to analyze the anxiety of speakers,and the evaluation results of this paper are compared with those of trained personnel.The actual test shows that the average deviation rate of this method between the anxiety and emotion analysis and the human analysis is only 3.36%.The results show that the method based on multiple action features is practical.In the simulated speech scene,it can effectively analyze the speaker’s anxiety level in a long-term scale and form a personalized report. |
