Font Size: a A A

Real-time Remote Detection Of Human Stress Based On Respiration Signals

Posted on:2017-03-29Degree:MasterType:Thesis
Country:ChinaCandidate:Y H ShanFull Text:PDF
GTID:2308330503483836Subject:Signal and Information Processing
Abstract/Summary:PDF Full Text Request
Human psychological stress is an un-homeostasis state that is triggered bypsychological stressors. Moderate stress can help people concentrate more and thus perform better. However, long-term stress may lead to various diseases. Therefore, a system that can continuously monitor the psychological stress state will be helpful for people to understandtheir stress states and manage their stress accordingly. However, traditional stress monitoring systems use physiological signals measured by contact way, i.e. the sensors have to be attached to the users. Inconvenience and discomfort may be introduced if the sensors are worn for a long time. Moreover, the traditional systems only discriminate the psychological stress from the calm state without considering the effect of the physical stress state, which shares many physiological features with the psychological stress.In this thesis, a methodology forremote sensing psychological stress in real-time by using Microsoft Kinect is proposed. This method features non-contact respiration signal measurement, real-time classification of psychological stress, physical stress, and relaxing states,and would be helpful for everyday stress management.Four parts of work has been done in this master project, they are described as below:(1) Development of a system for remote measuring respiration signals. The Microsoft Kinect was employed for the respiration signals measurement. Algorithm was designed for a robust measurement independent of body movement. The signal measured by our system is compared with that measured by contact sensor, BIOPAC MP150.(2) Design of stress experimentand establishment of database. 86 participants’ relaxing, psychological stress, and physical stress state were induced, which were stimulated by music, Stroop Test, and moderate exercise, respectively. Their respiratory signals were obtained under the three states. After preprocessing, these signals and their corresponding affective labels formed the stress database for future analysis.(3) Recognition and classification of the stress. Two hundred and seventy six features of the respiration signals in time domain and sequence domain were extracted. Hypothesis test was employed to select 60 features from the 276 features. Three “one to many” fisher classifiers andone random forest classifierwere used to distinguish three affective states. Cross-validation results show that average accuracy of the three fisher classifiers were 94.19%, 93.02%, and 86.05%, respectively. The average accuracy of the random forest classifier was above 90%.(4) Development of a system for real-time stress monitoring. The Kinect was used as the sensor. The signal processing algorithm and classifier were implemented by using C# in a PC that controlling the Kinect. The system developed updateaffective states(relaxing, psychological stress, and physical stress) every 10 seconds.
Keywords/Search Tags:Psychological Stress Detection, Physical Stress Detection, Stress Classification, Real-Time Stress Monitoring, Remote Detection of Stress
PDF Full Text Request
Related items