We propose a new recognition method which minimizes the affect of the 3D error of stereo reconstruction or video reconstruction. After 3D structure of an object is found via any Structure from Motion (SfM) method, 2D recognizers will be implemented on...
We propose a new recognition method which minimizes the affect of the 3D error of stereo reconstruction or video reconstruction. After 3D structure of an object is found via any Structure from Motion (SfM) method, 2D recognizers will be implemented on the minimal error sub-space of the object"s 3D reconstruction. For each image frame from either stereo images or video clips, the best 2D sub-space recognizer is determined by projecting objects original 3D (CAD) models onto each individual image frame"s 2D sub-space. Compared to the previous work [1] where 3D reconstruction was directly used for recognition, sub-space method will use sub-space to minimized the influence of the unknown 3D triangulation error.