Assessing the Computational Effort for Structural 3D Vehicle Recognition

Abstract

A model based structural recognition approach is used for 3D detection and localization of vehicles. It is theoretically founded by syntactic pattern recognition using coordinate grammars and depicted by production nets. The computational effort significantly depends on certain tolerance parameters and the distribution of input data in the attribute domain. A brief theoretical survey of these interrelations is accompanied by comparing the performance on synthetic random data to the performance on data from different natural environments.

Probabilistic Decisions in Production Net An Example from Vehicle Recognition

“…Background features are assumed to be equally distributed all over the picture. Such assumption is valid only in special situations or if nothing else is known about the background [9]. If additional information is given, e.g.…”

“…Such productions operate on sets of objects instead of graphs, strings etc. The organization of object concepts and productions can be depicted by a production net [9] which displays the part-of hierarchies of object concepts.…”