The minimum risk angle for automatic target recognition using the intersecting cortical model
2004 (English)In: Proc. Seventh Int. Conf. Inf. Fusion, 2004, 1014-1021 p.Conference paper (Refereed)
While kids easily find 3-D objects like animals in a scene (e.g. a photograph), this is still not the case for algorithms running on von Neumann computers or neural network chips. The present investigation has two goals: Finding "signatures" of the object in the scene and trying to find out at which observation angle the chance of correct identification is the best. By signatures we mean a vector of reasonable size (say 50 elements). Clearly a cow looks different from the back or from the side. A car is probably more easily identified viewed from the front than it is from above. For a plane it may be the other way around. Thus if we define a general but compact "signature" of the object, it will surely depend on the viewing angle. The problem of finding the most optimal viewing angle is dealt with in this paper.
Place, publisher, year, edition, pages
2004. 1014-1021 p.
, Proceedings of the Seventh International Conference on Information Fusion, FUSION 2004, 2
Conditional density estimation, Feature extraction, Image signature, Target angle, Viewing angle, Algorithms, Cameras, Electronic document identification systems, Matrix algebra, Neural networks, Parameter estimation, Risks, Object recognition
IdentifiersURN: urn:nbn:se:kth:diva-156943ScopusID: 2-s2.0-6344229556ISBN: 917056115XOAI: oai:DiVA.org:kth-156943DiVA: diva2:768930
Proceedings of the Seventh International Conference on Information Fusion, FUSION 2004, 28 June-1 July 2004, Stockholm, Sweden
QC 201412052014-12-052014-12-042014-12-05Bibliographically approved