Vision-based probabilistic absolute position sensor

Many industrial applications require to determine the absolute position of an extended surface without modifying or touching the target object. This contribution presents a concept for an optical absolute position sensor based on an off the shelf camera, operating perpendicular to an extended surface over long strokes, as necessary e.g. for piston actuators. The proposed sensor uses a Particle Filter to measure the absolute position within an once-only learned global feature map with low memory footprint, which is achieved by using an adapted feature detector. The probabilistic approach allows for certain robustness against false feature detection and enables fast recovery after power loss without the need for a referencing movement. The absolute position is detected with sub-millimeter accuracy over a stroke of 100 mm.