Vision System for the Safe Deployment of Air Bags

Student: Che Anton Caldwell

Supervisor: Dr. Brian Lovell

Category: Electrical Engineering Thesis Project

First installed in 1986 in some luxury cars, air bags are now standard equipment in most countries. Air bags, however, are also highly controversial because they have killed people and injured more with severe trauma to the neck and head. Although virtually all of the bag's safety problems would be eliminated if passengers wore seat belts, authorities are now calling for air bags to be made safer, requiring auto manufacturers to redesign air-bag systems around the fact that that not all passengers do. Most safety improvements revolve around sensor-based systems that detect the size and position of the passenger seat's occupant and prevent the air bag from deploying if there is the potential for harm. However, these systems are still in various stages of development and will not be incorporated into production vehicles for several years.

I have, in my thesis, tried to develop a cheap but effective vision based system to control the deployment of air bags to maximize occupant safety. Computer vision systems lend themselves ideally to just such a task. My thesis investigates the developed a system using cheap and readily available usb camera’s, configured in a stereo pair, to detect the presence and position of an occupant.

Although many algorithms exist to detect the presence of a person in an image, this task is much more difficult to do in real time. The described method uses a combination of colour, edge and elliptical head model matching to achieve real time detection. It then goes on to use the detected object presence in a binocular set of images to interpret the range of the detected object.

I have not produced an integrated prototype but preliminary testing proves that the concept is viable.

More text!

Poster Presentation (PDF)

Thesis Document (PDF)