AN ACOUSTIC SENSOR SYSTEM FOR OBJECT RECOGNITION

Object identification with application to industrial manufacturing processes using acoustic means is an alternative method to optical sensing. A sensor system consisting of a pair of broad-band transducers operating at wavelengths of some millimeters in air can recognize objects with sizes up to 15 mm. Arranged in a side-looking position with respect to the objects, the transducer pair is moved across a linear aperture to give spatial and spectral information. The echo data are processed to calculate an angle-dependent spectrum which allows object displacements to take place. A pattern is generated which is invariant for object rotations of several degrees. During a teach-in procedure, data are collected from an ensemble of interesting objects. A small set of discrimination vectors is calculated using the Karhunen-Loeve expansion. Objects belonging to the ensemble are identified by a k-nearest-neighbour classifier.