Detection of moving objects in video is a very important task in numerous applications of computer Vision, video processing, surveillance and vision-based human-computer interaction. Most systems that attempt to detect and analyzer events in live or recorded video rely heavily on an early step, commonly called “background subtraction”, that segments the scene into novel (“foreground”) and non-novel (“background”) regions.

Trespasser detection example

Detection of moving objects in video can be difficult for several reasons. The algorithm needs to account for possible changes in illumination of a scene, constantly moving background objects such as waving trees, shadows and inter-reflections from moving objects, background changes when objects are added or removed from scene, etc.

Our method is based on modelling the background color distribution with adaptive Gaussian Mixtures (GMM), coupled with color-based shadow detection. The developed algorithm posesses several favourable properties:

• Can handle non-stationary background;
• The shadows are detected and separated from the moving objects;
• Is robust to gradual changes of illumination;
• Can adapt to radical background changes (objects adding/removal);

Examples

Moving objects an shadow detection videos 1
Moving objects an shadow detection videos 2
Moving objects detection with slight camera shift videos

The project team

Contacts