BMVC 2004, Kingston, 7th-9th Sept, 2004
Real-time occupant detection system in an active illumination
J.J. Yoon (City University) and T. J.Ellis (Kingston University)
A single grey-scale camera based object classification system for vehicle
airbag deployment control in wide and frequent illumination variations is
introduced. Image sequences are acquired using an active illumination systems
that is used to minimise the effects of the widely varying levels of ambient
illumination, combined with a means of shadow suppression. Twodimensional
information of the object is extracted by employing the active
contour model, based on a priori knowledge of the passenger behavior. A
triplet of images, of which each image is illuminated from a different direction,
are sequentially used by the photometric stereo method to recover
the three-dimensional shape of the object. Utilizing both the two and threedimensional
properties of the object, a 29-dimensional feature vector is de-
fined for the training of a neural network designed to solve a three-class problem,
with the classes being forward-facing child seat, rear-facing child seat,
and adult. The system is tested on a database of over 84,000 frames collected
from a wide range of objects in various illumination conditions. A classification
accuracy of 98.9% was achieved within the decision-time limit of three
seconds.
(pdf article)