Please use this identifier to cite or link to this item: https://scholarbank.nus.edu.sg/handle/10635/231417
Title: IMPROVING INDOOR OBJECT RECOGNITION USING 3D CONTEXT
Authors: YE QIYUAN
Keywords: Indoor Object Recognition, 3D Context, Adjacent Objects, Conditional Probability
Issue Date: 14-Jul-2022
Citation: YE QIYUAN (2022-07-14). IMPROVING INDOOR OBJECT RECOGNITION USING 3D CONTEXT. ScholarBank@NUS Repository.
Abstract: This dissertation focuses on the problem of indoor object recognition to help people with visual impairments (PVI). We envision a system where a PVI has a wearable device, equipped with a camera, that can identify the objects in a scene and transcribe it to audio to help the PVI in orientation and mobility. A naive approach toward indoor scene recognition is to use object recognition algorithms and label each object in the scene independently. In this dissertation, we wish to exploit 3D information of the indoor scene to improve the accuracy of object recognition. We assume that the 3D layout is published by venue owners (e.g., hotels, malls) to make the places more accessible to PVI. Using YOLOv4 as the core, we integrated information from the 3D layout of a scene into the indoor object recognition pipeline. Particularly, we include the distance between (fixed) objects in the scene, so that co-occurrence of objects contributes to the likelihood of an object being identified correctly. Using the AI2Thor indoor scene dataset, we showed that our method improve object recognition accuracy from 81% to 90%.
URI: https://scholarbank.nus.edu.sg/handle/10635/231417
Appears in Collections:Master's Theses (Open)

Show full item record
Files in This Item:
File Description SizeFormatAccess SettingsVersion 
YeQY.pdf3.13 MBAdobe PDF

OPEN

NoneView/Download

Page view(s)

26
checked on Dec 1, 2022

Download(s)

6
checked on Dec 1, 2022

Google ScholarTM

Check


Items in DSpace are protected by copyright, with all rights reserved, unless otherwise indicated.