Event-driven proto-object based saliency in 3D space to attract a robot's attention.

Area of Science:

• Robotics
• Computer Vision
• Neuroscience

Background:

• Robots require object organization from visual input for environmental interaction.
• Depth information is crucial for directing attention to objects.
• Existing depth-based attention models are limited by synchronous RGB-D cameras.

Purpose of the Study:

• To develop a bio-inspired, bottom-up attention model for robots utilizing event-driven sensing.
• To generate depth-based saliency maps for robot interaction with complex visual input.
• To leverage the advantages of event cameras for enhanced robotic perception.

Main Methods:

• Utilized event cameras on the iCub humanoid robot for asynchronous visual data capture.
• Developed a model exploiting event-driven sensing to extract edge, disparity, and motion information.
• Generated depth-based saliency maps to guide robot attention.

Main Results:

• The system successfully identified salient objects in cluttered and dynamic scenes.
• Demonstrated robust object selection in real-world experiments.
• Showcased the model's effectiveness in selecting objects near the robot.

Conclusions:

• The proposed event-driven attention model enables robots to effectively interact with complex visual environments.
• This approach offers advantages over traditional methods by using high-resolution, low-latency event cameras.
• The system benefits downstream applications such as object segmentation, tracking, and robot interaction.