Banca de QUALIFICAÇÃO: ORCILANO MOTA LUZ

Uma banca de QUALIFICAÇÃO de MESTRADO foi cadastrada pelo programa.
STUDENT : ORCILANO MOTA LUZ
DATE: 09/10/2020
TIME: 14:00
LOCAL: Teleconferência - https://meet.google.com/yca-bcha-pir
TITLE:

Simultaneous Localization and Mapping using ground images (Ground SLAM) with hybrid foundation: visual odometry based on direct image recording and mapping based on characteristics


KEY WORDS:

Ground SLAM; hybrid SLAM; dual quaternion posture representation; direct image registration; efficient second-order optimization.


PAGES: 50
BIG AREA: Engenharias
AREA: Engenharia Elétrica
SUMMARY:

A majority of tasks mobile robots have, such as path planning, obstacle avoidance or positioning, depend strongly on localization. One of the possible solutions that has gained popularity over the years is that of Simultaneous Localization and Mapping (SLAM), in which the robot creates a map of the surrounding environment while predicting the current pose (position and orientation) at the same time. In one of the SLAM variants, named Ground SLAM, the robot points a camera downwards to the floor and uses this visual information to measure the relative movement of the robot with respect to its environment.

Ground SLAM can be divided into appearance-based algorithms and feature-based methods. Appearance-based algorithms focus their efforts on information extracted from pixel intensities in a region of the image. Feature-based methods attempt to detect points, lines or other distinguishable and robust features between consecutive images. Both approaches have advantages and drawbacks.

In this work, we propose to combine appearance-based algorithms and feature-based methods in a hybrid proposal of Ground SLAM. The localization part will mainly be based on direct non-linear image registration, using a dual quaternion to represent the rotation and translation between consecutive images and an efficient second order optimization method to calculate the new posture. The mapping, however, would not store all previous images, but only the most distinctive features that appear on several images. We will use line segments on the floor as visual features and compare their efficiency with other classic visual features based on points and descriptors.


BANKING MEMBERS:
Presidente - 350751 - ADELARDO ADELINO DANTAS DE MEDEIROS
Interno - 1242315 - PABLO JAVIER ALSINA
Externo à Instituição - ANDERSON ABNER DE SANTANA SOUZA - UERN
Notícia cadastrada em: 22/09/2020 16:53
SIGAA | Superintendência de Informática - | | Copyright © 2006-2021 - UFRN - sigaa09-producao.info.ufrn.br.sigaa09-producao