A Coarse-to-Fine Taxonomy of Constellations for Fast Multi-class Object Detection

Using surfaces and surface relations in an Early Cognitive Vision system

“…In a number of works in computer vision, the potential of hierarchical structures were successfully exploited [12,17,18,20,27,30,61,[70][71][72]84]. Such approaches can be distinguished between designed [20,30] and learned [17,18,71,72] and hybrid models [12].…”

“…Such approaches can be distinguished between designed [20,30] and learned [17,18,71,72] and hybrid models [12]. Early examples of mainly designed hierarchical models are Fukushima's Neocognitron [20] and the model of Hummel and Biederman [30].…”

“…Some neurophysiological evidence draws a picture of a well balanced amount of prior structure in the human visual system that is required to bootstrap visual learning (for a more extensive discussion, see, e.g., [39,47]). One way of reducing the complexity of this learning problem is to learn different levels one after the other as, e.g., done in the work of Leonardis et al [17,18] on compositional architectures for 2D edge structures. Their system also allows for a certain degree of explicitness at the different hierarchical levels by attaching semantics to learned structures.…”

“…Their system also allows for a certain degree of explicitness at the different hierarchical levels by attaching semantics to learned structures. As said above, our system is completely designed but provides richer information than, e.g., [17,18] by covering not only 2D edge information but also 3D and surface information.…”

Real-time Learning and Detection of 3D Texture-less Objects: A Scalable Approach

“…Using a fast similarity measure, an object is detected within 300-900ms. Building a hierarchy of detectors was also achieved in [10]. Arrangements of edge fragments are learnt in an unsupervised statistical manner from training images, and are later combined into a hierarchy of detectors.…”