The aim is to analyze different activities and behaviors from tracked objects (e.g. humans, faces, machine tools) from videos. Some typical examples of applications are in: elderly care centers or out-patient care units, office environments, vehicle drivers, and automations.
The study includes the machine recognition of different (or, specific) activities/behaviors, analysis of individual activities to obtain a range of long-term (or short-term) statistics on normal/abnormal activities/behaviors, for improving elderly care, detecting abnormality, improving office environment, and reducing driving risk. Main methods we investigate in this project are:
• Novel machine learning and pattern classification methods, e.g., domain-shift classifiers on manifolds, AdaBoost, SVMs;
• Effective feature descriptors for activities/behaviors;
• Statistical modeling and parameter estimations for normal/abnormal activities;
• Generate recommendations or trigger actions based on objective criteria.