FA-Harris: A Fast and Asynchronous Corner Detector for Event Cameras (1906.10925v4)

Abstract: Recently, the emerging bio-inspired event cameras have demonstrated potentials for a wide range of robotic applications in dynamic environments. In this paper, we propose a novel fast and asynchronous event-based corner detection method which is called FA-Harris. FA-Harris consists of several components, including an event filter, a Global Surface of Active Events (G-SAE) maintaining unit, a corner candidate selecting unit, and a corner candidate refining unit. The proposed G-SAE maintenance algorithm and corner candidate selection algorithm greatly enhance the real-time performance for corner detection, while the corner candidate refinement algorithm maintains the accuracy of performance by using an improved event-based Harris detector. Additionally, FA-Harris does not require artificially synthesized event-frames and can operate on asynchronous events directly. We implement the proposed method in C++ and evaluate it on public Event Camera Datasets. The results show that our method achieves approximately 8x speed-up when compared with previously reported event-based Harris detector, and with no compromise on the accuracy of performance.

• The paper introduces a novel FA-Harris algorithm that leverages a global Surface of Active Events to streamline asynchronous event processing with 8× speed improvements.
• The method employs a multi-stage pipeline with event filtering, efficient candidate selection, and Harris-inspired refinement to enhance corner detection accuracy.
• The approach demonstrates promising real-time performance for applications in robotics and SLAM, setting the stage for future integration with adaptive machine learning models.

FA-Harris: A Fast and Asynchronous Corner Detector for Event Cameras

The development of neuromorphic sensors, such as event cameras, has ushered in new possibilities for dynamic scene analysis in fields such as robotics and computer vision. Unlike conventional cameras, event cameras offer high temporal resolution and robustness to rapid motion and high dynamic range scenes. The paper "FA-Harris: A Fast and Asynchronous Corner Detector for Event Cameras" addresses the gap in feature detection methodologies tailored specifically for the asynchronous nature of event cameras, which play a critical role in applications like Simultaneous Localization and Mapping (SLAM).

Methodological Overview

The core contribution of this paper is the FA-Harris, a novel corner detection method optimized for event cameras. The pipeline of FA-Harris encompasses several components: an event filter, a Global Surface of Active Events (G-SAE) managing unit, a corner candidate selecting unit, and a corner candidate refining unit.

1. Event Filter: The process begins with the filtering of events to reduce redundancy, ensuring that only significant changes trigger processing, thus optimizing performance.
2. G-SAE Maintenance: Diverging from traditional local SAEs that require maintenance for each pixel's locality, the FA-Harris employs a G-SAE that maintains the most recent timestamps for all pixels globally. This structure significantly enhances processing speed by reducing data management complexity.
3. Corner Candidate Selection: This component efficiently narrows down potential corner candidates using the G-SAE, thereby minimizing computational load by discarding irrelevant data at an early stage.
4. Corner Candidate Refinement: Inspired by the Harris detector, the refinement process determines the cornerness based on a calculated score from asynchronous events. This stage ensures high accuracy in corner detection, leveraging the spatial-temporal coherence inherent in event data.

Performance Evaluation

The FA-Harris method is implemented in C++ and benchmarked against existing methods on public Event Camera Datasets. It delivers an approximate 8× improvement in processing speed over previously established event-based Harris detectors without compromising detection accuracy. This performance gain is chiefly attributed to the G-SAE's effective event management and the streamlined candidate refinement process, which together facilitate real-time operation on high-frequency event streams.

Implications and Future Work

The introduction of FA-Harris sets a precedent for efficient and asynchronous corner detection in the field of event-based vision. As the utility of event cameras expands, especially in high-speed and complex environments, methods like FA-Harris that leverage the asynchronous nature of events will become increasingly critical.

Future avenues of research could include integrating FA-Harris with machine learning models to enhance its adaptability and robustness across a broader range of environmental conditions. Additionally, the development of feature descriptors with scale and rotation invariance would further augment the capabilities of event-based vision systems, enabling more robust feature matching and loop closure detection in event-based SLAM systems.

Conclusion

This paper significantly contributes to the field of neuromorphic vision by introducing a corner detection algorithm specifically designed to harness the unique advantages of event cameras. The improvements in processing speed and accuracy highlighted in the FA-Harris method not only confirm the viability of event cameras for real-time applications but also provide a robust foundation for future advancements in feature-based SLAM and other vision-based applications.