EMBEDDED VISION-BASED SIDE COLLISION WARNING SYSTEM FOR MONITORING THE BLIND SPOT AREA

摘要

In this paper, we proposed a vision-based side collision warning (SCW) system, which features the combination of lane detecting function and vehicle detecting function to monitor the vehicles in the blind spot area by using the CCD camera mounted on the side mirror on a vehicle. Our proposed lane detecting function not only detects the being of lane but also classifies it as continuous line or dotted line for judging if vehicle detection function should be activated. As for the function of vehicle detection, our proposed algorithm detects vehicles with the feature-based approach and verifies the vehicle candidates by the appearance-based approach. Besides, we define three ROIs with different sizes and sampling rates. Only one of the three ROIs will be processed for detecting vehicle each time by applying motion vector (MV) estimation. When a vehicle in the adjacent lane is overtaking, the vanishing of the strong symmetry makes the continuously detecting and tracking difficult. For continuous monitoring, we proposed another mode of tracking by comparing the variance of the blind spot area before and after vehicle is overtaking. Our proposed algorithm achieves high detecting rate and low computing power and is successfully implemented in ADI-BF561 600MHz dual core DSP.