AVRAM
Research Project Attentive Vision for Robotic Assistance Machines (AVRAM)
Subject
Early Clustering Approach towards Artificial Visual Attention for Increasing Performance of Active Vision Systems
Abstract
Despite its short history modeling visual attention has found many useful applications in different areas such as visual search, perceptual grouping, object and gesture recognition, object tracking, image and video compression and scene rendering in 3D graphics. We intended to broaden the scope to areas such as driver assistance systems, rescue robotics, aid systems for the elderly or handicapped, and other areas where machine vision could play a helping role for people. In this project, we demonstrated attention behaviors on real-time active vision systems integrated in mobile robots working in natural 3D environments with live sensor data so that this work became a milestone towards the long-term target of biologically inspired human assistance machines. For this purpose, we extended the modalities of our existing attention model by adding the necessary feature channels of motion and depth, and reformulating the rest of the attention model in accordance with the new changes. Moreover we introduced a spatial memory infrastructure into the model to store locations and features of the learned objects and to allow the vision system to operate autonomously for long time spans. In order to quantitatively measure the performance of this project in particular and evaluate the output of attention models in general, we built a sharable resource for performance assessment and benchmarking of visual attention models. The tool for attention evaluation was developed in close interaction with leading experts in the field and made available for the research community in this area.
Status
Completed
Partners
None
Financing
Deutsche Forschungsgemeinschaft