Multi-view 3D sense retrieval approach based on dynamic Bayesian networks

A new dynamic 3D scene retrieval approach is proposed based on a novel dynamic Bayesian network (DBN) lightfield descriptor. To overcome the disadvantages of the existing 3D scene retrieval methods, we explore dynamic Bayesian network for building a new lightfield descriptor. Firstly, dynamic 3D object is put into lightfield, and many gray-views can be obtained along a sphere, and then features can be calculated based on gray-views. DBN graph model would be built based on learning of feature sequences. Secondly, a new 3D scene retrieval method is proposed based on graph model measurement. Beneficial from the statistical learning, our descriptor is robustness as compared to the existing methods. Experimental results demonstrate that our proposed approach is with better performance than the existing methods.