Human gait type classification

Classification of different types of gait (walking, jogging, and running) can not be done reliably based solely on the speed of motion. We investigate representations of the human gait that allow us to classify gait types without knowledge about the scene or the camera. We typically work with surveillance-like videos which put a natural emphasis on robust methods.

Human gesture recognition

Our work on human gesture recognition aims at developing a small set of primitives that can be used to describe a set of gestures without modeling the whole trajectory of the movement. By identifying a few primitive in a video sequence we can classify the gesture being performed. We base our approach solely on motion and hence avoid advanced modeling of a human.

Robust foreground segmentation

In surveillance videos there is a need for robust foreground segmentation that can handle the natural changes that happen in a scene during a day. Overall illumination changes, shadows, and changing backgrounds are all thing that need to be handled by foreground segmentation without loss of detection power. We have been working with such a method and have tested the performance of the method extensively.