Occupant classification by boosting and PMD-technology

This paper discusses a system for classification of a vehicle passenger for air bag control, using 3D-images from a time-of-flight measuring sensor. Radial depth is determined using a photonic mixer device (PMD) with suppression of background illumination for near infrared wavelengths. The occupant region is determined by fitting a gradient based model for the seat shape and surface features are extracted for classification into one of the classes empty, child seat and adult occupant. It presents a novel approach for classification by supervised online data-boosting, using AdaBoost and support vector learning. Both methods show highly accurate classification results, comprising 2.0% and 0.15% error rate, for a boosted and a representative set of training samples, respectively.