This project answers the need of following the front of human figures, in an environment of multiple cameras.
Abstract
This project answers the need of following the front of human figures, in an environment of multiple cameras. The system that has been developed operates in the following stages: Initially, an identification of a figure is performed inside a frame, by two perpendicular cameras. Afterwards, the location of the center of the mass is found, for each camera. This way, the figure’s location in the room can be found, and also its advancement and velocity vectors can be calculated, and indicate its direction of advancement, and thus, the optimal camera to capture its front.
The system has been tested by six videos, under the following assumptions: a static background, a single person in the room, a “natural” movement (no walking sideways, or backwards is allowed), and also, the person shouldn’t be holding or waving a relatively large object (such as a banner in a protest). There is room for future development in order to perform the required processing, without these assumptions.
Flowchart
Determine optimal camera
Results
Future research
- Use face detection or other body parts
- Turning while standing
- Dynamic background