Material classification with laser thermography and machine learning

Abstract

Robotic devices that can perform teleoperation work in complex and unstructured environments (e.g. healthcare, education, automotive, telepresence, defence, etc.) rely mainly on visual and/or auditory feedback to interact with a remote operator. There is a vital need for measuring the physical properties of ambient environments when performing precise manipulation tasks over a distance, to improve the quality of performance. In this paper, we propose an approach to combine infrared thermography with machine learning for a fast, accurate way to classify objects of different material. A laser source stimulates the surface of the object while an infrared camera captures its thermal signature, extracts features about these signatures and feeds them into a machine learning-based algorithm for classification. Results demonstrated that a classification accuracy of around 97% can be achieved with majority vote Decision Tree classifier, even in the presence of data from multiple data acquisition sessions.