Human activity recognition from video sequences

Contact: Dr Adrian Bors

The first part of the project involved the development of an observational video analysis system that would learn the activities taking place in a given space based on their movements’ directions and intensities. This would allow for a dictionary of activities to be formed that any new activity could be compared with to identify it as a known activity or anomalous.

During the second part of the project, a classification system for human group interaction activities was developed. Human group activities display complex movement, where people either replicate movement or interact softly and violently with each other. The software being developed was applied for assessing video data showing various activities taking place outdoors by groups of four to six people.