Monocular Simultaneous Localization And Mapping with Moving Object Tracking using Conditional Independent submaps

From AIRWiki
Revision as of 14:58, 21 March 2009 by RobertoRigamonti (Talk | contribs)

(diff) ← Older revision | Latest revision (diff) | Newer revision → (diff)
Jump to: navigation, search

Realistic, unstructured environments constitute a treacherous playground for robotic systems in charge of mapping a scene whilst simultaneously localizing in it. Limited computational resources and low-cost sensing devices make these tasks more involved. Whilst state of the art algorithms rely upon the highly restrictive static environment posit, and might experience consistency issues when estimations are confined to a single map, this thesis introduces a novel solution that extends covariance form submapping techniques to dynamic environments, identifying and tracking moving objects by means of a single camera, and thus preventing them from corrupting estimations. This result has been achieved by coupling a pure monocular EKF-SLAM algorithm, working on conditional independent submaps expressed in the inverse scaling parametrization, with a Bearing-Only Tracker. The most recent proposals from the SLAM research community are thus coalesced in a single system, which aims to provide on-line, accurate, and reliable navigation in a real, unstructured terrain. Albeit traditional submapping solutions impose statistical independence among estimates in different maps, the conditional independent SLAM algorithm adopted allows to exchange information between neighboring submaps, improving past estimations with current measurements by means of a back-propagation mechanism. The inverse scaling parametrization presents a measurement equation which is provably more linear than the commonly adopted inverse depth parametrization, while at the same time demanding fewer computational resources (since the number of required parameters for each element decreases for six to four). Geometrical reasoning has been exploited to detect dynamical objects in the scene by properly casting maps' elements in the Uncertain Geometry setting. Finally, an adaptive threshold over the number of features in each submap has been established by means of a fuzzy controller, which takes into account maps' fill-in to avoid wasting computational resources. A full Matlab implementation of the developed ideas is presented, validating theoretical results by means of extensive experiments in both simulated and real environments.

Part 1: project profile

Project name

Monocular Simultaneous Localization And Mapping with Moving Object Tracking using Conditional Independent submaps


Start date: 2008/08/01

End date: 2009/04/20

People involved

Project head(s)
Other Politecnico di Milano people
Students currently working on the project
External personnel:

Laboratory work and risk analysis

This project does not include laboratory activities.