Kalman Filter-Based Coarse-to-Fine Control for Display Visual Alignment Systems

A coarse-to-line two-stage control method is investigated for the display visual alignment systems. The proposed visual servo is with hierarchical loops, where the original line but slow vision loop is necessary for the exact localization of alignment marks while the coarse but fast vision loop of exploiting pruned image data is to compensate for the mask-panel misalignment. The degraded resolution of the reduced images is recovered in terms of the Kalman filter which tracks the mark centroids in near realtime. In order to construct the recursive estimation algorithm, the motion model for the moving alignment marks is determined by solving the forward kinematics of positioning mechanism and the measurements from vision sensors are given by means of the geometric template matching (Kwon and Hwang, “Kinematics, pattern recognition, and motion control of mask-panel alignment system,” Control Eng. Practice, vol. 19, pp. 883-892, 2011). Compared with the conventional alignment methods, this approach enables a fast and line alignment control. Experimental results are followed to validate the proposed control framework. Note to Practitioners-In order to successfully apply the developed alignment control to any display manufacturing equipment, it is necessary to well understand the principle of the geometric template matching (GTM) as an alignment mark specific fast algorithm, the details on which can be consulted in our preceding works (Kwon and Hwang, “Kinematics, pattern recognition, and motion control of mask-panel alignment system,” Control Eng. Practice, vol. 19, pp. 883-892, 2011). The new approach has the goal of updating the pose of an alignment mark as fast as the capturing rate of a frame grabber by utilizing pruned image data but recovering the lost resolution in terms of the Kalman filter. For example, in using a common 30 fps grabber, the reduced image of 320 240 pixels is a proper choice to finish the image processing and Kalm- n filtering within 30 ms under GTM. The proposed algorithm can be implemented in the current industrial display aligners by modifying the control software so that the reference inputs for the distributed joint servos follow the error compensation trajectory in Fig. 4 with the Kalman filter estimates.