Vision system with depth obtaining capabilities

Obtaining of 3D information from image sequences or stereoscopic images is a key problem, not completely solved, and any improvement in its resolution would bring to light a huge quantity of real life applications. The main difficulties lie in the matching of different elements of every image, the so-called correspondence problem. According to the trend of analysing singular points of the scene, we opt for a system based on the labelling and matching of contours and corners of the scene. Those points have some properties that make the method usable for most kinds of images. In order to obtain the contour information, a new algorithm is developed which is hardware implementable, able to autoadapt its parameters to external characteristics (such as as focus, illumination, contrast, etc.), and provides high reliability to the depth obtaining system. A specialised hardware is designed for contour obtaining and labelling with a fast and highly efficient pipeline architecture that gives the system a fast response time.