Proposes a new method of active vision which recognizes the 3-Dshape of objects without knowing camera motion parameters. The motionparameters are calculated from the optical flows and the depth of objectpoints whose 3-D shape is already known. Then, using these calculatedmotion parameters and the optical flows, the 3-D position of unknownpoints are reconstructed, which, in turn, will be used as the knownpoints in the next frame of image. These processes are iterated for asequence of images to recognize the 3-D scene. In this method, theeffects of quantization errors are overcome by two approaches. Theerrors of camera motion parameters are compensated by using a largenumber of points to calculate them. Then, the Kalman filtering method isapplied to the sequence of images to reduce the 3-D position errors ofeach unknown point
展开▼
