A Precise Visual Control Method for Micromanipulator

A feature-based object recognition algorithm, which is robust against occlusion, is presented. The subpixel accuracy can be obtained by minimizing the sum of the distance between corresponding pixels of template and image. In order to avoid the iterative adjustment and complicated inverse solution of image jacobian, based on CMAC (Cerebellar Model Articulation Controller) neural network, a feedforward visual servo scheme is used to implement the mapping from the error signal in the image space to the control signal in the task space. Experimental result shows that precise positioning accuracy can be obtained by using this method.