Classification of vehicle occupants using 3D image sequences

The deployment of vehicle airbags for maximum protection requires information about the occupant's position, movement, weight, size etc. Specifically, it is desirable to discriminate between adults, children, front or rear-facing child seats, objects put on the seat or simply empty seats. 2D images lack depth information about the object and are very sensitive to illumination conditions. Occupant position classification techniques are developed based on low resolution 3D image sequences. The proposed methods are of low complexity and high reliability, allowing real time implementation and meeting the rigorous requirements for passenger safety systems. Features are extracted from the 3D image sequences and a sequential forward search (SFS) feature subset selection algorithm is employed to reduce the size of the feature set. Two classification techniques are evaluated, the Bayes quadratic classifier and the polynomial classifier. We present the classification results based on a large set of measurements from low resolution 3D image sequences. Full scale tests have been conducted on a wide range of realistic situations (adults/children/child seats etc.) which may occur in a vehicle.