Attribute-Weighted Time Surface-Based Denoising Method for Space Object Event Streams

Event streams generated by Dynamic Vision Sensors (DVS) for space object detection often contain substantial background noise. Existing denoising methods face challenges in balancing the preservation of useful events, noise removal, and real-time processing. In this paper, we present a simple but effective method for denoising space object event streams, leveraging attribute-weighted time surface. By incorporating attribute weights into the time surface, we enable efficient, asynchronous denoising based on the spatiotemporal correlations of events. Additionally, to address the limitations of traditional evaluation metrics when dealing with imbalanced datasets and to provide a comprehensive assessment of a method’s ability to suppress noise while preserving signal events, we introduce the Precision-Recall curve as a qualitative evaluation tool and propose the harmonic mean of the area under the curve and the maximum F1 Score (HMAF) as a new quantitative evaluation metric for event stream denoising. Experimental results show that the proposed denoising method effectively retains valid events, reduces noise, and surpasses seven state-of-the-art methods in terms of accuracy and robustness on three datasets, and maintains excellent real-time performance. Furthermore, our suggested evaluation metric can quantify the denoising accuracy and conduct an objective and comprehensive benchmark evaluation of different denoising methods.