Eventos Anais de eventos
COBEM 2021
26th International Congress of Mechanical Engineering
Efficacy test of the visual odometry method on TerraSentia robot
Submission Author:
Jorge Id Facuri Filho , SP
Co-Authors:
Jorge Id Facuri Filho, Vitor Akihiro Hisano Higuti, Marcelo Becker
Presenter: Jorge Id Facuri Filho
doi://10.26678/ABCM.COBEM2021.COB2021-1051
Abstract
In the agricultural scenario, some of the biggest challenges that farmers suffer are the identification of trails, navigation and mapping of remote and hard to access regions. In this context, visual odometry is one of the methods to obtain positioning and orientation information, which when applied to a mobile robot, removes the need for people to move around in these difficult access environments. Since it relies on a single monocular camera, visual odometry (VO) is often cheaper and more flexible in the robot assembly when compared to common localization sources such as GPS. Additionally, visual odometry recovers the robot’s trajectory only by analyzing the changes of feature points in the captured images. Thus, a visual odometry algorithm is implemented using the images captured by TerraSentia robot to enable the estimation of its trajectory in an agricultural environment. FAST detector with the Kanade-Lucas-Tomasi tracker for extraction with tracking is implemented to compare the trajectory from VO with the trajectory from EKF data. In a scenario without interferences that disturb the tracker, only drift errors are present in the estimated trajectory. The path recovered by VO gets to follow the ground truth, with less than 1 m of translation error.
Keywords
Agriculture, Mobile Robots, Computer Vision, Visual Odometry
