Computational proto-object detection in 3D data G Martín García, S Frintrop |
|---|
Institute of Computer Science III, University of Bonn, Germany
|
| For humans as well as for machines, object detection is an essential task to understand the world and interact with it. The situated vision theory of Pylyshyn (Pylyshyn, 2001, Cognition 80, 127-158) states that in human vision, the detection of visual objects preceeds the categorization and investigation of their properties. This is in contrast to the standard approach in computer vision that usually learns object categories and applies the resulting classifiers to images. In this work, we present a computational approach that follows the idea of Pylyshyn and detects proto-objects without prior knowledge about categories or properties of objects. The detection is based on a visual attention system that detects salient blobs which are improved by iterative segmentation steps. As input device, we use a depth camera that provides color as well as depth information and is used to create a 3D representation of the scene. Detected proto-objects are projected into this 3D scene map and incrementally updated when data from new viewpoints is available. The system is able to find unknown objects and to create 3D object models without prior knowledge in real-world scenarios. |
| Up Home |
|---|