Accurate Recognition of Lightweight Objects With Low Resolution Pressure Sensor Array

Abstract

Flexible pressure sensor array systems, being able to measure spatial contact force distribution and perform accurate object recognition, are demanded for a wide range of applications. Using low resolution passive matrix sensor arrays would be able to provide an economic solution by decreasing manufacturing complexity and cost. However, with the low resolution data and presence of crosstalk and noise, signal processing techniques need to be developed for accurate recognition of lightweight objects. A blurring and sharpening (BS) method is proposed in this work to process the sensor array data with presence of noise and crosstalk effect to obtain more recognizable pattern features for further classification. The method is applied to a $10\!\!\times \!\!10$ pressure sensor array system with non-ideal characteristics including noise, variation of the measured response among different pixels, and signal crosstalk. It is shown that, with the BS method, the recognition accuracy using different classification algorithms can all be obviously improved. The sensor array system is demonstrated to be able to recognize 10 low weight objects (pressure < 70 Pa) with different digit shapes from 0 to 9 with a high accuracy reaching 92.4%.