Enhancing vision data using prior knowledge

A significant problem which arises when assembling or reworking surface mount devices is that of linear and angular misalignment between the device and its mating part on the printed circuit board. With the expectation that devices will get smaller and of even finer pitch, people will not be capable of this work and for automation the problem of; alignment will become more severe. The manufacture of the device and the PCB is such that dead reckoning is not adequate. Usually a vision system is used to determine the position of the surface mount pads relative to the manipulators datum and to cross relate this information to achieve a successful assembly. This is done by looking at discrete pads on the device and fiducial marks on the PCB. The problem which arises is that either the whole picture is viewed and the resolution obtained using even the best vision systems is poor or parts of the scene are viewed individually where the resolution is better but where the necessary manipulator motion can introduce errors. The paper describes various techniques which have been tried to improve the resolution of a vision system based on the known and plentiful knowledge about the relative locations of the assembly features.