A Dataset for Visual Navigation with Neuromorphic Methods.

Area of Science:

• Computer Vision
• Neuromorphic Engineering
• Robotics

Background:

• Standardized benchmarks have driven progress in computer vision.
• Event-driven vision and neuromorphic methods require dedicated benchmarks for comparison with conventional approaches.

Purpose of the Study:

• To introduce novel datasets for evaluating frame-free and frame-based visual navigation methods.
• To facilitate comparisons between neuromorphic and conventional computer vision techniques.

Main Methods:

• Development of synthetic and real-world datasets featuring static scenes and rigid camera motion.
• Inclusion of diverse data modalities: images, events, optical flow, 3D camera motion, and depth.
• Generation of simulated event data from existing frame-based optical flow datasets.

Main Results:

• Datasets provide comprehensive data for evaluating visual navigation algorithms.
• Enables direct comparison of neuromorphic and traditional computer vision methods.
• Facilitates accuracy assessment for both frame-based and frame-free approaches.

Conclusions:

• The presented datasets are crucial for advancing event-driven vision research.
• These benchmarks will enhance the visibility and impact of neuromorphic methods in computer vision.
• Standardized evaluation is key to progress in visual navigation and related fields.