Object classification with aggregating multiple spatial views using a machine-learning approach

Abstract

The article proposes a solution for object classification using multiple views generated from 3D data rendering and convolutional neural networks. For presentation purposes and easier verification of the solution, an application was developed to create views of 3D objects, classify them using the selected CNN, and evaluate the performance of the CNN. The evaluation is based on metrics and characteristics described in the article. Seven testing objects were used to verify the proposed solution; five CNNs were tested for each.