Faster 3D Object Detection in RGB-D Image Using 3D Selective Search and Object Pruning

3D object detection in RGB-D image has received considerable attention recently. But potential searching space in testing image is large, and extracting handcraft feature for every candidate hounding box is computational expensive, In this paper we introduce 3D Selective Search(SS) to generate high quality cuboids that most likely to contain object. Besides, Object Pruning is proposed to speed up the testing process by cutting hypothesis cuboids. The result evaluated on SUN RGB-D dataset demonstrates that our method is able to speed up testing process more than 50% without performance loss, compared with exhaustive searching in 3D space.