The automatic focusing of a camera is frequently necessary. Closed loop techniques that relate measurements of image blur to focus distance provide self-contained mechanisms for autofocus. Making the focuser adaptive gives the controller the ability of improving its performance over time. To achieve a focused image search of the space of focus positions is required. Treating the camera as a non-linear and time varying plant the input is the image distance and the output is a blur measurement a control law used to drive the blur measurement to a minimum is adapted over time. A trial and error process is employed by the focuser to expedite the adaption. From the adaption process a model of camera dynamics is identified. To achieve real-time adaption a single nonlinear neuron is employed. Sufficient adaption speed is achieved by the use of Parker''s second order LMS learning algorithm and by the adaption of the learning rates. Stability of the closed loop focuser is insured by maintaining a bracket on the optimum focus distance. Bracketing provides a way of incorporating a lower bound on the error from the optimum focus distance into the adaption error function. The performance of the adaptive closed loop focuser is illustrated by simulations using a synthetic camera. Depending on both the quality of the blur measurements and the size of the object distance step the focuser''s asymptotic performance typically requires only

Date Published: 12 August 1992
PDF: 11 pages
Proc. SPIE 1656, High-Resolution Sensors and Hybrid Systems, (12 August 1992); doi: 10.1117/12.135920

Published in SPIE Proceedings Vol. 1656:
High-Resolution Sensors and Hybrid Systems
Morley M. Blouke; Winchyi Chang; Laurence J. Thorpe; Rajinder P. Khosla, Editor(s)