Scene Classification in Indoor Environments for Robots using Context\n Based Word Embeddings

摘要

Scene Classification has been addressed with numerous techniques in computer\nvision literature. However, with the increasing number of scene classes in\ndatasets in the field, it has become difficult to achieve high accuracy in the\ncontext of robotics. In this paper, we implement an approach which combines\ntraditional deep learning techniques with natural language processing methods\nto generate a word embedding based Scene Classification algorithm. We use the\nkey idea that context (objects in the scene) of an image should be\nrepresentative of the scene label meaning a group of objects could assist to\npredict the scene class. Objects present in the scene are represented by\nvectors and the images are re-classified based on the objects present in the\nscene to refine the initial classification by a Convolutional Neural Network\n(CNN). In our approach we address indoor Scene Classification task using a\nmodel trained with a reduced pre-processed version of the Places365 dataset and\nan empirical analysis is done on a real-world dataset that we built by\ncapturing image sequences using a GoPro camera. We also report results obtained\non a subset of the Places365 dataset using our approach and additionally show a\ndeployment of our approach on a robot operating in a real-world environment.\n