Machine learning based shape classification using tactile sensor array

Abstract

Contact shape recognition is an important functionality of any tactile sensory system as it can be used to classify the object in contact with the tactile sensor. In this work we propose and implement an affine transformation invariant 2D contact shape classification system. A tactile sensor array gives the contact pressure data which is fed to an image enhancement system which sends the binary contact image to the global, region and contour based feature selection block. The five element feature vectors is used as input to a voting based classifier which implements a C4.5 algorithm and a naïve Bayes classifier and combine the results to get an improved classifier. The system is designed and tested offline in a Pressure Profile system based tactile array using both cross validation and separate dataset. Results indicate that combining simple classifiers increases the accuracy of the system while being computationally efficient. © Springer International Publishing Switzerland 2014.