Position estimation by merging low cost imu and camera data using the extended kalman filter

Moura, Philipe Miranda de

Please use this identifier to cite or link to this item: http://hdl.handle.net/11422/17958

Type:	Trabalho de conclusão de graduação
Title:	Position estimation by merging low cost imu and camera data using the extended kalman filter
Other Titles:	Estimativa de posição combinando IMU de baixo custo e câmera Dados usando o filtro Kalman estendido
Author(s)/Inventor(s):	Moura, Philipe Miranda de
Advisor:	Lizarralde, Fernando Cesar
Abstract:	This work details the development of a positioning system that couples computer vision and inertial navigation in order to improve safety in a refinery. This combination of sensors was made since refineries typically are partially indoor – limiting the use of global-navigation-satellite-systems – and require a particular spark-free certified smartphone to be used. The studied refinery has about 20,000 assets – among flanges and valves – that need to be opened, closed or maintained. Ensuring intervention is performed in the correct asset is currently done by checking QR codes attached to each one of them. However, the current solution is expensive to be maintained and does not allow navigation. The navigation based on computer vision is provided by the QuickVision system developed by a company. It consists of a camera able to detect a preset pattern and determine its pose with respect to the pattern coordinate frame. The inertial navigation information was obtained using a MEMS IMU equipped with both a 3D accelerometer and gyroscope. In addition, the fact that the smartphone is hand-held by a human being walking through the plant allowed a Zero Velocity Update (ZUPT) to be implemented in the vertical component in order to reduce drift. The main contribution of this work was to investigate the accuracy over time of IMU-based position estimation when no visual patterns were in view. The results obtained in the office under perfect conditions show that the gyroscope’s noise causes random walk in the estimated orientation typically 1° off the true orientation, generating error in the estimated position in the order of a few meters after 10s without visual aiding. By comparing the results, it can be inferred that MEMS IMU are not suitable for dead reckoning applications without holonomic constraints, for amounts of time greater than a few seconds. As consequence, visual information should be used during pattern absence as aid by tracking features present on images.
Keywords:	Localization Kalman Filter ZUPT
Subject CNPq:	CNPQ::ENGENHARIAS
Production unit:	Escola Politécnica
Publisher:	Universidade Federal do Rio de Janeiro
Issue Date:	Dec-2018
Publisher country:	Brasil
Language:	eng
Right access:	Acesso Aberto
Appears in Collections:	Engenharia de Controle e Automação

Files in This Item:

File	Description	Size	Format
monopoli10029971-min.pdf		1.59 MB	Adobe PDF	View/Open

Show full item record Recommend this item View Statistics

Pantheon Institutional repository