PEDRo: an Event-based Dataset for Person Detection in Robotics

Chiara Boretti, Philippe Bich, Fabio Pareschi, Luciano Prono, Riccardo Rovatti, Gianluca Setti

Research output: Chapter in Book/Report/Conference proceedingConference contribution

4 Scopus citations

Abstract

Event-based cameras are devices based on neuromorphic sensors that are gaining popularity in different fields, including robotics. They are suitable for tasks where high-speed, low-latency, low-power operations are required. Person detection is one of these, to allow mobile robots to monitor areas and navigate in crowded environments. Most of the available event-based datasets that contain annotated human figures and collected with a moving camera are designed for autonomous driving tasks. Yet, robotic tasks are certainly not limited to the recognition of pedestrians walking on sidewalks, which makes the above datasets of limited utility. To address this impasse, we introduce a new dataset called PEDRo, which is fully manually labeled. This dataset has been specifically developed for person detection and it counts a total number of 43259 bounding boxes included in 119 recordings. A moving DAVIS346 event-based camera has been used to collect events in a large variety of indoor and outdoor scenarios with various lighting and meteorological conditions (such as sunny, rainy and snowy). To the best of our knowledge, this is now the largest available dataset for event-based person detection, which has been recorded with a moving camera and manually labeled.
Original languageEnglish (US)
Title of host publication2023 IEEE/CVF Conference on Computer Vision and Pattern Recognition Workshops (CVPRW)
PublisherIEEE
DOIs
StatePublished - Jun 2023

Bibliographical note

KAUST Repository Item: Exported on 2023-08-21

Fingerprint

Dive into the research topics of 'PEDRo: an Event-based Dataset for Person Detection in Robotics'. Together they form a unique fingerprint.

Cite this