This paper proposes a high resolution video sensor for the 3D reconstruction of 
architec-tural models from multiple image sequences. The hybrid system unifies 
triangulation methods of spatial stereo with tracking methods of temporal 
stereo. We describe an efficient spatial image match-ing algorithm, which is 
based on trinocular image rectification and semi-global optimization. The motion 
of the video sensor is estimated using temporal feature tracking and allows the 
integration of dense point clouds. First experimental results are shown for 
images of a real scene.