DVS-UP-Fall Dataset

DVS-UP-Fall: Event-Based Dataset for Fall Detection The DVS-UP-Fall dataset is an event-based version of the publicly available UP-Fall Detection dataset, tailored for advancing fall detection research using neuromorphic vision sensors and spiking neural networks. This dataset has been generated by converting RGB video data to event streams using two leading event conversion toolkits: v2e (video-to-events) and v2ce (video-to-continuous-events). The dataset is structured into two main ZIP archives: 1. dvs-up-fall-v2e-dataset-160.zip: This archive contains event data generated using the v2e toolkit. 2. dvs-up-fall-v2ce-dataset-160.zip: This archive contains event data generated using the v2ce toolkit. Label Files To support supervised learning and benchmarking, we provide the following label files. CompleteDataSet.csv: Original UP-Fall labels including IMU, EEG, and IR sensor data. labels_multiclass_w1.0.csv: Multiclass activity labels for 1.0-second non-overlapping windows. labels_binary_w1.0.csv: Binary (fall vs. no-fall) labels for 1.0-second non-overlapping windows. labels_multiclass_w0.5.csv: Multiclass activity labels for 0.5-second windows. labels_binary_w0.5.csv: Binary labels for 0.5-second windows. This dataset supports both binary and multiclass activity classification tasks using event-based data and is suitable for training and evaluating spiking neural networks (SNNs). All data is organized per subject, activity, trial, and camera, consistent with the original UP-Fall dataset structure.