Japanese / English

Detail of Publication

Text Language Japanese
Authors Katsufumi Inoue,Hiroshi Miyake,Koichi Kise
Title A Memory Reduction Method for 3D Object Recognition Based on Local Descriptors—An Approach by Selecting Local Descriptors—
Journal Proceedings of MIRU 2008
Presentation number OS15-3
Pages pp.363-370
Reviewed or not Not reviewed
Month & Year July 2008
Abstract 3D object recognition methods based on local descriptors have been investigated. These methods em- ploy local descriptors extracted from a lot of images to construct object models and recognize 3D objects with these models. An important problem of these methods is extensive memory requirement due to a large number of local descriptors that constitute the models. In this report we propose a simple method to reduce the amount of memory by selecting local descriptors stored in the models. The proposed method is characterized by the mechanism of selection which employs an estimate of the ”contribution” of each local descriptor to object recognition. From experimental results for 11 objects, we achieved the recognition rate over 98% with about 1% of local descriptors extracted from all images for models. Experimental results with COIL-100(Columbia Object Image Library-100), show that 1/6 of the total local descriptors allowed us the recognition rate over 96%.
Back to list