A new method of position estimation for visual servoing

A position estimation scheme in visual servo system with single camera mounted on a robot is proposed. The scheme combines the techniques of depth-from-motion and depth-from-defocus to satisfy the real-time requirement. For long distance moving objects, the position estimation of the object is obtained by Plucker expression in Grassmann-Cayley algebra. For close distance moving object, a multilayer feedforward neural network is employed to recover the object position from defocused image. A criterion is used to switch between these two methods. The improvements of estimation accuracy by using the proposed method are verified by experimental results.