ACM Transactions on Programming Languages and Systems (TOPLAS)
A Model of Saliency-Based Visual Attention for Rapid Scene Analysis
IEEE Transactions on Pattern Analysis and Machine Intelligence
Comparing images using joint histograms
Multimedia Systems - Special issue on video content based retrieval
Saliency, Scale and Image Description
International Journal of Computer Vision
Parallel Computer Architecture: A Hardware/Software Approach
Parallel Computer Architecture: A Hardware/Software Approach
Attentional Selection for Object Recognition A Gentle Way
BMCV '02 Proceedings of the Second International Workshop on Biologically Motivated Computer Vision
Fast and Lock-Free Concurrent Priority Queues for Multi-Thread Systems
IPDPS '03 Proceedings of the 17th International Symposium on Parallel and Distributed Processing
Human-Robot dialogue for joint construction tasks
Proceedings of the 8th international conference on Multimodal interfaces
Detection of visual attention regions in images using robust subspace analysis
Journal of Visual Communication and Image Representation
Integrating multimodal cues using grammar based models
UAHCI'07 Proceedings of the 4th international conference on Universal access in human-computer interaction: ambient interaction
Integrating language, vision and action for human robot dialog systems
UAHCI'07 Proceedings of the 4th international conference on Universal access in human-computer interaction: ambient interaction
| Hi-index | 0.00 |
In this paper we propose an attention-based vision system for the JAST interactive dialog robot. The robotic vision system incorporates three submodules: object recognition, gesture recognition and self recognition. The performance boost of our biologically inspired vision system is based on two assumptions: first, generally attention is attracted by regions of high intensity or hue gardients as well as scene dynamics (bottom-up attention attraction), and second, attentioninal focus can be directed by higher level modules, whether volitional or not, in an inhibitory or reinforcing way (top-down attention control). The system proposed in this paper is able to utilize these assumptions and organize its computational efforts accordingly. Integrated into an efficient data management architecture, the vision system is capable of continuously publishing results to the cognitive layer of the robot and thus enables operations in realtime. Furthermore, the modular system structure and the asynchronous communication paradigm allows for efficient integration of additional modules, be it visual or any other sensory input data. The main contribution of this work is the application of neuroscience findings and biologically plausible theories of attention based visual processing to a real-world robotic setup. Here, our experimental results show tremendous speed-ups using either the bottom-up attention attractors or the principle of top-down attention control as input data filters for further visual analysis, reaching the peak in a combination of the two.