ANMS: Asynchronous Non-Maximum Suppression in Event Stream

Qianang Zhou,JunLin Xiong,Youfu Li

2023-03-19

Abstract:The non-maximum suppression (NMS) is widely used in frame-based tasks as an essential post-processing algorithm. However, event-based NMS either has high computational complexity or leads to frequent discontinuities. As a result, the performance of event-based corner detectors is limited. This paper proposes a general-purpose asynchronous non-maximum suppression pipeline (ANMS), and applies it to corner event detection. The proposed pipeline extract fine feature stream from the output of original detectors and adapts to the speed of motion. The ANMS runs directly on the asynchronous event stream with extremely low latency, which hardly affects the speed of original detectors. Additionally, we evaluate the DAVIS-based ground-truth labeling method to fill the gap between frame and event. Evaluation on public dataset indicates that the proposed ANMS pipeline significantly improves the performance of three classical asynchronous detectors with negligible latency. More importantly, the proposed ANMS framework is a natural extension of NMS, which is applicable to other asynchronous scoring tasks for event cameras.

Computer Vision and Pattern Recognition

What problem does this paper attempt to address?

The paper attempts to address the problem of implementing the Non-Maximum Suppression (NMS) algorithm in event cameras to improve the accuracy and efficiency of feature detection. Specifically:

1. **Background and Challenges**:
- In traditional frame-based vision systems, NMS is a widely used post-processing algorithm for tasks such as feature detection, edge detection, and object detection.
- Event cameras, due to their asynchronous nature, can provide extremely high temporal resolution when dealing with changes in brightness, making them suitable for fast motion and high dynamic range scenes.
- However, existing NMS methods in event cameras either have high computational complexity or lead to frequent discontinuities, thereby limiting the performance of event-based corner detectors.

2. **Research Objectives**:
- Propose a general asynchronous non-maximum suppression framework (ANMS) that can run directly on asynchronous event streams with very low latency.
- Refine the coarse feature stream output by the original detector through the ANMS framework to improve detection accuracy.
- Validate the effectiveness of ANMS on public datasets and demonstrate its significant performance improvement on three classic asynchronous detectors in different scenarios.

3. **Main Contributions**:
- Designed a general and adaptive asynchronous non-maximum suppression pipeline (ANMS) that significantly enhances the performance of all available detectors on public datasets.
- Proposed a simple and effective scoring method for the event-based FAST algorithm, which reuses intermediate results.
- Introduced a method for evaluating ground truth labels and improved ground truth accuracy by adjusting parameters.

In summary, the paper aims to address the limitations of NMS in event cameras through the ANMS framework, thereby enhancing the overall performance of event-based feature detectors.