Can depth data improve the accuracy when classifying mops?

Abstract

Since the emergence of low cost RGB-D cameras, a new world of possibilities have opened up in the field of Computer Vision. This projects focuses on both the practical and theoretical part of how depth data can improve the accuracy when classifying objects in an industrial environment. We have tested both classical machine learning methods and Google's Residual Neural Network: MobilNetV2. The goal was to achieve an accuracy that can match HF RFID-tag(95-97\%). The purpose of the research question was to find out more about RGB-D images and what methods that can be used best for classification.