As more and more safety vision applications begin to
integrate into vehicle platforms it becomes necessary to
choose a general purpose object detection algorithm that
encompasses the detection of all that is present in the road
ahead. It is not just the collision prone targets that we are
interested in; we also must understand the landmarks that
signify a particular region where the vehicle is present.
Therefore, instead of investigating many different object
detection algorithms present in the current literature we
evaluate a robust real-time algorithm [2] called the Viola and
Jones object detection algorithm. The conventional approach
for object detection with real-time results would be to perform
flesh toning of the entire image region to focus on a particular
region of interest and return this area to be tracked continually
[3]. One of the advantages of employing the Viola and Jones
over this conventional approach is its success in terms of
running its detector on the entire image and yet achieving realtime
results of about 15 fps on a conventional 700 MHz
Pentium III [2].
Although our ultimate goal would be to implement this on
embedded hardware, given the utilization of a GPU we would
definitely obtain an accelerated performance. Another
advantage of this algorithm is its ability to construct a
comprehensive data model of the scene ahead comprised of
different types of targets and landmarks commonly
encountered on an urban road scenario. This is achieved using