Car Rear-view Mirror Blind Spot Detection And Warning Key Technology Research

Vision-based moving vehicle detection is an significant application of machine vision in driver assistance system. Compared with other non-visual sensors, visual sensor can determine the type of target accurately, but holds a disadvantage which is hugely time-consuming. However, with the improvement of computer hardware and image processing algorithms, the shortcoming is weakening and the advantage is highlighted.Blind spot of vehicle rear mirror caused startling traffic accident, especially when the vehicle is changing lane. This dissertation proposes a vision-based solution aiming to alert a driver about blind spot and possible collision. The blind spot detecting system (BSDS) processes image sequences taken from a backward-looking video camera mounted on rear mirrors and report the result to the driver. Then the BSDS gives the distance estimated between the target vehicle and the camera-assisted vehicle.This dissertation follows a widely used two-step method to detect vehicle in blind spot: 1) hypothesis generation (HG) where the location of possible vehicle in an image is hypothesized and 2) hypothesis verification (HV) where tests are performed to verify the presence of the vehicle. The facial feature of vehicle is used as a prior knowledge. First, Prewitt operator is used to obtain edge information and these edges are accumulated in horizontal and vertical direction separately to locate the potential vehicle. Second, a symmetrical feature is used to verify the existence of the vehicle hypothesized previously. The author develops the template-based matching method, proposes a "self-template matching" method, amends the matching algorithm. This will not only obtain effective results, but realize the BSDS in the embedded platform without the use of template library. Before that, Lane and road boundary are detected as preprocessing to eliminate the scene beyond the borders and reduce the complexity of vehicle detection. It is proved by PC test that the approach could meet real-time requirement.Subsequently, a pinhole camera model is used to estimate the inter-vehicle distance. The camera internal parameter is obtained by calibrating the actual distance and its corresponding pixels in image. And then the distance is calculated.