openTSST – An open web platform for large-scale, video-based motion analysis during acute psychosocial stress
Exposure to acute stress leads to characteristic changes in body posture and movement. Analyzing these changes can be a valuable extension of the existing markers for assessing an acute stress response, such as cortisol and inflammatory markers. It might provide novel insights into stress research. However, state-of-the-art motion capture systems require specialized equipment which comes with high costs, limited availability, and can interfere with natural behavior during stress induction. Video-based pose estimation can overcome these limitations but has not been widely adopted yet. For that reason, we propose openTSST as a platform for large-scale and end-to-end extraction of body posture and movements from video data and to enable a more holistic stress assessment.
The openTSST platform was designed to allow researchers to extract motion parameters from videos recorded during stress protocols, such as the Trier Social Stress Test (TSST), the gold standard for acute stress induction in the laboratory. It consists of a web interface for the user and the business logic for coordinating processing tasks. Videos uploaded to openTSST are stored in simple storage service (S3) instances and sent to a GPU worker cluster that performs human pose estimation using OpenPose, which is then used for feature extraction characterizing human posture and movements during acute stress.