A multi-view stereo approach generates an inventory of objects located on an object holder. An object may be a sample tube and an object holder may be a tube rack as used in lab automation for healthcare diagnostics. A processor performs 3D tracking of the object holder and the geometric analysis of multiple images generated by a calibrated camera. A homography mapping between images is utilized to warp a second image to a viewpoint of a first image. Plane induced parallax causes a normalized cross-correlation score between the first image and the warped second image of a location on the holder that has an object that is significantly different from a normalized cross-correlation score of a location that has not an object and enables the processor to infer tube inventory and absence or presence of a tube at a location in a rack.