An occupant sensor incorporates a 3-D imaging system that acquires a 3-D image of an object. The image is segmented to remove unwanted portions and to identify a region-of-interest, and the content thereof is classified responsive to a plurality of 3-D features. In one embodiment, the 3-D image is transformed to a second 3-D image from a second viewing perspective. A two-dimensional projection of the second 3-D image is classified, and a presence, size and position of occupant can be identified from features thereof. A safety restraint system is controlled responsive to the detected scenario, including the presence, position and size of an occupant.