Connectionist Decision Systems for a Visual Search Problem
Uwe R. Zimmer · 1994
ion Levels Stop! Feedback Raw data Edges Scan! Efficiency Feedback Edge/Surface Map Ready! Look! Feedback } figure 1 : Focus of attention system The statistical informations are spacial densities of edges, free-space and unknown regions. In order to collect and represent these densities we use a special, dynamic self-organizing map with 3-dimensional simplexes (tetrahedrons) as the basic structure. The simplexes dividing the space in Voronoi regions, attributed with informations about the amount of sensor data collected in this region as well as densities (edges/ free-space). The network is adapted by replacing single cells by new simplexes or by fusing neighbouring simplexes, according to several rules (see [7] for all the insertion/fusion criteria) . This unsupervised process gives the needed input for the area-of-interest selection. Up to now this decision is done by simple heuristics (search strategies), without using the efficiency feedback from the higher abstraction levels. In...