Abstract
Object recognition usually includes colour, shape and material types. This paper presents a methodology for surface material recognition by a tool which is tapped on an object for robotic applications. Recognition of a surface material can be explored by scratching the tip of the tool over the surface. To classify surface types, many different sensors such as acceleration, force, reflectance, image and audio were used via automated robot movements. For this purpose, 28 different surface materials including such as metals and papers were used. It should be emphasized that the properties of surface materials are also different. 22 different classifiers were trained with these surfaces using Matlab Classification Learner Application. The data which is collected ten times from sensors were examined also in different combinations. First, all data (combination of acceleration, force and reflectance) except image and audio data was observed. Then; only image, only audio and dual combinations of all data subsets were evaluated. In the end, classification accuracy of fused data including all sensors was compared to the rest of the results. The proposed fusion of all features provides a classification accuracy of 98.2% in our experiments when combined with a Bagged Trees classifier.