A Computational Framework for Attentional 3D Object Detection

We present a computational framework for the detection of unknown objects in a 3D environment. It is based on a visual attention system that detects proto-objects which are improved by iterative segmentation steps. At the same time a 3D scene model is built from measurements of a depth camera. The detected proto-objects are projected into the 3D scene, resulting in 3D object models which are incrementally updated. Finally, environmentand object-based inhibition of return enables to withdraw the attention from one object and switch to the next. We show that the system works well in cluttered natural scenes and can find and segment objects without prior knowledge.