GENERAL ENQUIRIES: Tel: + 27 12 841 2911 | Email: callcentre@csir.co.za

Show simple item record

dc.contributor.author Govender, Natasha
dc.contributor.author Warrell, J
dc.contributor.author Keaikitse, M
dc.contributor.author Torr, P
dc.contributor.author Nicolls, F
dc.date.accessioned 2014-12-22T07:36:17Z
dc.date.available 2014-12-22T07:36:17Z
dc.date.issued 2015
dc.identifier.citation Govender, N., Warrell, J., Keaikitse, M., Torr, P. and Nicolls, F. 2015. Probabilistic active recognition of multiple objects using Hough-based geometric matching features. In: New Development in Robot Vision: Springer Publishing: Berlin Heidelberg, pp 89-109 en_US
dc.identifier.issn 978-3-662-43858-9
dc.identifier.uri http://link.springer.com/chapter/10.1007%2F978-3-662-43859-6_6
dc.identifier.uri http://hdl.handle.net/10204/7823
dc.description Copyright: Springer Publishing, Berlin Heidelberg en_US
dc.description.abstract Bayesian methods have proved effective in many active sensing scenarios, including object recognition from a mobile platform, tracking and medical diagnostics. Despite its importance though, models for active object recognition have concentrated on simple scenarios in which a single object is present, and there is no clutter or occlusion. In this paper, we investigate models for active object recognition in circumstances which are both more complex and realistic, in which multiple objects must be recognized simultaneously, and occlusion and clutter (through distracter objects) is common. We propose a representation for object viewpoints using Hough transform based geometric matching features, which are robust in such circumstances. We show how these features can be incorporated into Bayesian object models which can be used for single and multi-object active recognition tasks. Further, we investigate an efficient active viewpoint selection algorithm based on vocabulary-tree clustering and a Term Frequency Inverse Document Frequency (TFIDF) uniqueness metric. We test our fully Bayesian approach on challenging data containing multiple objects, and show it to give excellent results compared to approaches which do not incorporate geometric matching. Further, we show our viewpoint selection algorithm to be both faster and more accurate than alternatives in both Bayesian and discriminative contexts, including methods based on mutual information in the former case. en_US
dc.language.iso en en_US
dc.relation.ispartofseries Workflow;13812
dc.subject Active vision en_US
dc.subject Bayesian modelling en_US
dc.subject Object recognition en_US
dc.title Probabilistic active recognition of multiple objects using Hough-based geometric matching features en_US
dc.type Book chapter en_US


Files in this item

This item appears in the following Collection(s)

Show simple item record

Search ResearchSpace


Advanced Search

Browse

My Account