Extraction of moving obstacles from a moving observer is vital for many ITS applications like Forward Vehicle Collision Mitigation (FVCM) and Adaptive Cruise Control (ACC) systems. Single camera based Global Motion Model (GMM) like optical flow is commonly used in the past decade. This paper addresses a relatively new problem in the literature: GMM based on stereovision which involves the depth (disparity) information in the model, thus global motion can be analyzed in 3D space rather than traditional 2D image plane. An efficient algorithm is presented which parameterizes the GMM based on the 3D camera motion analysis within U-V-disparity domain. Combining the geometric and motion information, regions that do not match the GMM will be extracted as moving obstacles. For the covering abstract see ITRD E134653.
Samenvatting