A Novel Approach to Improve the Performance of a Classifier Using Visual and Haptic Data

In haptic identification of objects, various factors like friction and acceleration between the object and the haptic tool, audio, and image information contribute a lot. This paper proposes a novel technique to improve the performance of a classifier using visual and haptic data. Both conventional machine learning and deep neural network have been used for the experiment. The dataset used here is the haptic data with image, friction, and acceleration as inputs. Individual predictions models are developed for each input modality and then combined using the weighted average technique. The results showed a considerable improvement when compared to conventional multimodal feature-based classifiers. Simulation and analysis are done using Python3.7/Jupyter Notebook.