Technology Category
Computer Vision: Motion & TrackingMachine Learning: Multi-class/Multi-label Learning & Extreme ClassificationPlanning, Routing, and Scheduling: Activity and Plan Recognition
Event-Based Crossing Dataset (EBCD)
🤖 AI Summary
Traditional event datasets rely on fixed-threshold event generation, compromising the trade-off between detail preservation and noise suppression, and exhibiting poor generalization in dynamic outdoor scenes. To address this, we introduce MT-Event—the first multi-threshold event detection dataset tailored for traffic scenarios—capturing synchronized event streams of pedestrians and vehicles across ten adjustable thresholds (4–75), enabling joint modeling of sparsity and noise suppression. We propose the first multi-threshold acquisition framework, systematically revealing the nonlinear impact of threshold selection on detection performance and empirically validating the critical trade-off: low thresholds amplify noise, while high thresholds cause target loss (inducing up to 18.7% mAP variation). We benchmark MT-Event on YOLOv4/v7, EfficientDet-b0, MobileNet-v1, and HOG, advancing event representation from static-threshold to environment-adaptive paradigms. The dataset is publicly released, providing a large-scale, fine-grained, and reproducible evaluation benchmark for event-driven detection research.
Technology Category
Application Category
📝 Abstract
Event-based vision revolutionizes traditional image sensing by capturing asynchronous intensity variations rather than static frames, enabling ultrafast temporal resolution, sparse data encoding, and enhanced motion perception. While this paradigm offers significant advantages, conventional event-based datasets impose a fixed thresholding constraint to determine pixel activations, severely limiting adaptability to real-world environmental fluctuations. Lower thresholds retain finer details but introduce pervasive noise, whereas higher thresholds suppress extraneous activations at the expense of crucial object information. To mitigate these constraints, we introduce the Event-Based Crossing Dataset (EBCD), a comprehensive dataset tailored for pedestrian and vehicle detection in dynamic outdoor environments, incorporating a multi-thresholding framework to refine event representations. By capturing event-based images at ten distinct threshold levels (4, 8, 12, 16, 20, 30, 40, 50, 60, and 75), this dataset facilitates an extensive assessment of object detection performance under varying conditions of sparsity and noise suppression. We benchmark state-of-the-art detection architectures-including YOLOv4, YOLOv7, EfficientDet-b0, MobileNet-v1, and Histogram of Oriented Gradients (HOG)-to experiment upon the nuanced impact of threshold selection on detection performance. By offering a systematic approach to threshold variation, we foresee that EBCD fosters a more adaptive evaluation of event-based object detection, aligning diverse neuromorphic vision with real-world scene dynamics. We present the dataset as publicly available to propel further advancements in low-latency, high-fidelity neuromorphic imaging: https://ieee-dataport.org/documents/event-based-crossing-dataset-ebcd
Problem
Research questions and friction points this paper is trying to address.
Addresses fixed threshold limitations in event-based vision datasets
Introduces multi-threshold dataset for dynamic pedestrian/vehicle detection
Evaluates object detection performance under varying noise/sparsity conditions
Innovation
Methods, ideas, or system contributions that make the work stand out.
Multi-thresholding framework for event representation
Ten distinct threshold levels for varied conditions
Benchmarks state-of-the-art detection architectures
🔎 Similar Papers
No similar papers found.
