3D Point Cloud Reconstruction and Semantic Segmentation: Application to Indoor, Outdoor and Heritage Datasets

Sahithi, Veggalam (2023) 3D Point Cloud Reconstruction and Semantic Segmentation: Application to Indoor, Outdoor and Heritage Datasets. Masters thesis, Indian Institute of Technology Hyderabad.

Text
CE20MTECH12001.pdf - Submitted Version
Restricted to Registered users only until 22 June 2028.
Download (5MB) | Request a copy

Abstract

Recent advancements in data acquisition systems have led to an increase in deep learning applications on 3D datasets, with a particular focus on point clouds. 3D point cloud reconstruction and semantic segmentation techniques have emerged as crucial tools in computer vision and robotics, enabling a comprehensive understanding of complex environments. However, there is a lack of available point clouds for Indian heritage structures, which limits the application of computer vision techniques like semantic segmentation in this domain. In this thesis, we address this limitation by generating 3D point clouds of Indian heritage structures from 2D images using the structure from motion technique. We employ state-of-the-art semantic segmentation models, including PointNet and DGCNN, and train them using diverse datasets that encompass indoor, outdoor, and heritage scenes. The trained models are then evaluated using various metrics and benchmarks to assess their performance.

[error in script]

IITH Creators: