This paper describes a 3-D vision system based on a new 3-D sensor technology for the detection and classification of occupants in a car New generation of so-called "smart airbags" require the information about the occupancy type and position of the occupant. This information allows a distinct control of the airbag inflation. In order to reduce the risk of injuries due to airbag deployment, the airbag can be suppressed completely in case of a child seat oriented in reward direction. In this paper we propose a 3-D vision system based on a 3-D optical time-of-flight (TOF) sensor for the detection and classification of the occupancy on the passenger seat. Geometrical shape features are extracted from the 3-D image sequences. Polynomial classifier is considered for the classification task. A comparison of classifier performance with principle components (eigen-images) is presented. This paper also discuss the robustness of the features with variation of the data. The full scale tests have been conducted on a wide range of realistic situations (adults/children/child seats etc.) which may occur in a vehicle.