Low-Cost Sensor Fusion Technique for Surface Roughness Discrimination With Optical and Piezoelectric Sensors

Abstract

Surface roughness is one of the critical properties of an object. The capabilities of tactile sensors based on various working mechanisms were demonstrated to discriminate surface roughness. In our previous work, we showed that a biomimetic fingertip with piezoelectric sensors was effective in recognizing rough surfaces (Ra $> 1~\mu \text{m}$ ) but performed poorly if the surface roughness value was small (Ra $ ). Given the advantage of optical sensor in measuring fine surfaces, in this paper, we make an attempt to incorporate optical sensor together with the piezoelectric tactile sensors to recognize surface roughness. Specifically, we present two sensor fusion approaches, including feature-level fusion and decision-level fusion for surface roughness discrimination. Experimental results show that both fusion methods can improve the recognition performance, i.e., the highest classification accuracy of 99.88% and 98.83% can be obtained with decision-level fusion and feature-level fusion, respectively.