Vision-based 3D Spatio-Temporal Context Generation for Mobile Robot Applications

Abstract

A vision-based context generation system is proposed, where a point of the object associated with locations of SIFT features of the object model is utilized to describe an object as a representative point. And, SIFT features are also used to identify objects. Primitive spatial data and contexts are obtained based on our point-based description of objects as well as the use of some axiomatic rules. Finally, ST contexts are generated by classifying these primitive-ST contexts. Our proposed method is experimentally shown to be valid.