Video Dataset for COVID-19 Social Distancing and Human Detection Validation

Published: 12-11-2020| Version 1 | DOI: 10.17632/xh6m6gxhvj.1
Contributors:
Wasiq Khan,
Faisal Nawaz,
Abir Hussain

Description

The automated detection of the policy violators in busy places (e.g. public palces, cities, universities etc.) using deep learning and video surveillance such as street camera, is one of the hot research topic during the current COVID-19 pandemic situation. However, existing works lacks the reliable and solid validation of the estimated distance between two subjects in real-time video frame/s. We present a video/images dataset containing multiple participants standing at fixed positions while maintaining a static distance between them (i.e. 2 meter, 1 meter as ground truth distance). The capturing device (i.e. a video camera) is revolved around the participants to capture the distance from varying aspects such as distance from source, translations, rotations, zoom-in/out, and view angles (aerial, bird’s-eye view etc.). Each video frame contains the annotations (i.e. rectangular coordinates and anonymous subject ID) for a pair of subjects (i.e. any two persons) standing at a fixed distance from each other, which can be used as ground truth to validate the experimental results produced by the deep learning and computer vision based social distance measurements. Dataset (Files and Folders): 1) Two Meter Distance (Folder): This folder contains 2 video recordings with corresponding annotations for at-least 2 subjects in each video frame who are standing at fixed distance of exactly 2 meters Video 1: Top View (aerial perspective) Ground truth frames: This folder contains 1917 images (.png). Each image shows a rectangular box indicating the selected persons within the image who are standing 2-meter apart from each other. Annotations: This folder contains the 1917 annotations (.xml) for the images and can be used to validate the experimental outcomes. Ground_truth_2_Meter_video1.avi: This file contains the processed ground truth video frames Video2: Perspective/Horizontal View Ground truth frames: This folder contains 379 images (.png). Each image shows a rectangular box indicating the selected persons within the image who are standing 2-meter apart from each other. Annotations: This folder contains the 379 annotations (.xml) for the images and can be used to validate the experimental outcomes. Ground_truth_2_Meter_video2.avi: This file contains the processed ground truth video frames 2) One Meter Distance (Folder): This folder contains one video recordings of perspective view with corresponding annotations for at-least 2 subjects in each video frame who are standing at fixed distance of exactly 1 meter Ground truth frames: This folder contains 1689 images (.png). Each image shows a rectangular box indicating the selected persons within the image who are standing 1-meter apart from each other. Annotations: This folder contains the 1689 annotations (.xml) for the images and can be used to validate the experimental outcomes. Ground_truth_1_Meter_video.avi: This file contains the processed ground truth video frames

Files