Feature extraction from an infrared sensor array for localization and surface recognition of moving cylindrical objects

The paper presents a low-cost solution for the determination of characteristic properties of roughly cylindrical objects like size, location, trajectory and velocity of motion, and surface reflectivity. In particular, the problems of identifying different items on a conveyer belt and reliably determining the number of persons getting into or out of a room is addressed. Two arrays of several infrared (IR) emitter-receiver pairs, working on a reflection light scanner principle, are mounted on opposite sides of the area of observation. The emitters are driven successively in time, hence no signal overlapping and cross-talk occur. Two configurations of the IR array were compared. From simple light Intensity measurements, a variety of properties of (cylindrical) objects can be deduced. A surface modeling with elementary reflectors is proposed and results are shown proving its potential. The model parameters are obtained by an optimization procedure, Variants of the processing for object characterization are discussed.