Automated real time analytics of multi party dialogue

• Main Author: Chidambaram Murugan Senthil Nathan
• Other Authors: Justin Dauwels
• Format: Theses and Dissertations
• Language: English
• Published: 2017
• Subjects: DRNTU::Engineering::Electrical and electronic engineering
• Online Access: http://hdl.handle.net/10356/73068
• Institution: Nanyang Technological University

Schizophrenia is regarded as a mental disorder that is characterized by abnormal behavior and difficulty in perceiving the real and the fiction. Although some causes like the Genes, the working Environment and the brain chemistry & structure are known, still more research is required to identify the causes of schizophrenia. Till now research works have been done on manual methods and also audio data for the diagnosis of negative symptoms in schizophrenic patients. However manual methods could not produce significant results, the automated efforts through audio data analysis managed to yield significant results. But there has been very less research works carried out on video data for automated classification of negative schizophrenic patients. The primary objective of this dissertation work is to develop an automated real-time system based on video analysis that could automatically classify the schizophrenic patients and the healthier controls without the aid of psychiatrist. Visual data containing the interviews of schizophrenic patients and controls with psychiatrists acquired using Kinect Sensor is used as input data. This visual data is converted into text files each containing the 2D coordinate position of 19 skeletal joints of the participants using a GUI (Graphical User Interface) developed with Visual studio. In the case of video analysis, one of the important negative symptoms of schizophrenia is reduced responsiveness (reaction through body movements). The primary task is to classify the patients and controls by detection of reduced movements. So by using these coordinate positions, mathematical features like distance, position, velocity and angle are evaluated. Then these attributes are fed into Machine Learning algorithms and the accuracy of each of the classifier is observed and highly correlated features are ranked and the combination of highly correlated features which yield maximum accuracy are evaluated. Thus a real time automated system is developed to diagnose the negative schizophrenic patients.