ACTION RECOGNITION FOR DEPRESSION ASSESSMENT USING DEEP LEARNING

Abstract

Depression is the most common mental health mood disorder worldwide, which affects functionality and well-being of individuals having significant effect on their personal, family and social life. The objective ofthis project is to create a system for pose estimation using deep learning and to make a system of assessment of risk factor of depression patients depending on pose estimation. In this study, a small survey was conducted on those people who either have tried to commit suicide or have had thoughts on attempting suicide due to depression, as a result, 47.1% participants were found who either had tried or had thoughts of suicidal attempts. Three postures were found from the survey which were adapted by the individuals in those situations, being: hair pulling, looking down and slouching postures. In this project, the Dataset was collected through Google Images and augmentation applied to reach the required amount. The project is done using Convolutional Neural Network (CNN). CNN is a type of deep learning model for processing data and i . : : i ; was '■ , TensorFlow was used on the backend. Different stages involved in this project comprising of data-augmentation, data pre-processing, training and testing through CNN resulting in the overall accuracy of . ! 83%, video framing, classification/detection of postures done by the subject, and generation of depression flowchart which indicates the amount oftimes/ no. ofseconds a posture is done.