Evaluation of Event-Based Corner Detectors.

Area of Science:

• Computer Vision
• Robotics
• Sensor Technology

Background:

• Event-Based (EB) cameras offer superior performance in extreme lighting and motion compared to traditional cameras.
• Existing EB corner detection methods lack standardized evaluation, hindering direct comparison and development.
• Previous evaluations used limited criteria and operational conditions, potentially introducing bias.

Purpose of the Study:

• To establish a unified and rigorous evaluation procedure for bio-inspired Event-Based (EB) corner detectors.
• To compare the performance of five existing EB corner detection techniques under diverse and challenging conditions.
• To facilitate a better understanding of EB camera mechanisms and promote the development of more efficient detectors.

Main Methods:

• Evaluation of five EB corner detectors using a public dataset featuring extreme illumination.
• Application of a novel, unified procedure for consistent and unbiased comparison across detectors.
• Utilization of both intensity and trajectory information from the dataset for comprehensive analysis.

Main Results:

• Demonstrated the feasibility of rigorous EB corner detector comparison without manual labeling, even in challenging acquisition scenarios.
• Identified performance variations among existing EB corner detectors under extreme conditions.
• Established a benchmark dataset and methodology for future EB corner detector evaluations.

Conclusions:

• The proposed standard unified EB corner evaluation procedure enables objective assessment of detector performance.
• This methodology overcomes limitations of previous evaluations, offering insights into EB camera capabilities.
• The study paves the way for improved EB corner detection algorithms and enhanced EB camera technology.