Robust Vehicle Detection under Poor Environmental Conditions for Rear and Side Surveillance

The target accuracy is an error of 10% or less in the measured position. Assuming a mean relative speed A rear side system for vehicles has been between vehicles of 30 kmlh and that it takes the driver 0.4 developed that uses image processing. ~t uses two stereo s to avoid a collision, the minimum distance for avoiding a cameras to monitor the areas to the rear and sides of the collision with an approaching vehicle is 9 m. Adding a vehicle, i.e., the driverts blind spots, to detect the safety margin of 50% increases the distance to 14 m. The positions of other vehicles and their relative speeds. During system must therefore have a rmas~~rement a ~ o r no greater the period from dusk to dark, when it is difficult for a than 10% at a distance of 14 m to the rear and to the side. driver to see other vehicles, or when visibility is poor due This surveillance system is particularly helpful under to rain, snow, etc., the contrast between nearby vehicles poor conditions, such as the time between dusk and dark and the background is lower. Under such conditions, and during rain, snow, etc-9 b ~ ~ a u s e during these times it is conventional surveillance systems have difficulty detecting difficult for the driver to monitor the vehicle's surroundings. the outline of nearby vehicles and may thus fail to The environmental changes that occur at dusk, for example, recognize them. The proposed system can estimate the can reduce the contrast between nearby vehicles and the shape of a vehicle from a partial outline of it, thus enabling background in a video image, resulting in frequent failures the vehicle to be identified by filling in the missing parts. by ~ ~ e i l l a n c e methods to detect vehiclesThe stereo Testing of the system under environmental methods that have been proposed for detecting vehicle conditions showed that the recognition failure rate was position ([I] [41) require precise vehicle outlines to detect reduced to less than lo%, even at dusk and during rain or vehicles, SO they fail to detect vehicles when part of the snow, conditions that are problematic for conventional outline is missing due to poor environmental conditions, processing. which change a vehicle's shape in the image. To cope with this problem, we developed an