Scene understanding for autonomous vehicles with deep learning

This project aimed to carry out algorithm which is able to generate semantic images and then fuse the semantic images with point cloud to obtain point cloud with semantic information for the scene understanding tasks. This project can be seen as a base for scene understanding task, which covers a...

Full description

Saved in:
Bibliographic Details
Main Author: Wang, Dongying
Other Authors: Wang Dan Wei
Format: Thesis-Master by Coursework
Language:English
Published: Nanyang Technological University 2022
Subjects:
Online Access:https://hdl.handle.net/10356/159406
Tags: Add Tag
No Tags, Be the first to tag this record!
Institution: Nanyang Technological University
Language: English
Description
Summary:This project aimed to carry out algorithm which is able to generate semantic images and then fuse the semantic images with point cloud to obtain point cloud with semantic information for the scene understanding tasks. This project can be seen as a base for scene understanding task, which covers a wide range of data designing and processing work. First of all, labeling strategies designed for semantic segmentation aimed to help with scene understanding tasks have been carried out. In addition, for semantic information generation, a semantic segmentation network integrating attention mechanism and all MLP layers upsampling technique was applied. Finally, to get abundant information, the semantic images are fused with the point cloud using an algorithm based on robotics geometric projection theories.