ZebraStereoID

DOI

ZebraStereoID dataset contains video recordings of 2 groups of zebras. Both groups were recorded by 2 drones simultaneously (from opposing viewpoints), resulting in 4 videos in total. The groups consisted of 2 and 7 zebras, and were observed on different days.

The dataset consists of: 1) 4 videos, 30 FPS (1_videos.zip) 2) extracted frames for all videos (2_frames.zip) 3) detection JSON files for all frames, for all videos (3_annotations.zip) 4) cropped annotations for all frames, for all videos (4_cropped_bounding_boxes.zip)

Acquisition Equipment: All videos in the dataset were recorded using two off-the-shelf drones: 2 Parrot Anafi 4K. Videos were recorded at 4K resolution (4096x2160 pixels at 23.98 fps).

Acquisition Location: A region in Kenya characterized by arid and semi-arid savannahs and woodlands.

Acquisition Date and Time: All videos were recorded during daytime (between 0800 & 1800 local time), in the months of July 2022.

Acquisition Procedure: The research team first searched for the presence of zebra herds through manual observation from an off-road vehicle. After spotting the zebras, which typically happened during the first hour of the search, the vehicle was stopped approximately 200 m from the herd. One or two drones were then manually deployed such that the take-off sound did not startle the herd; the drones were placed behind the vehicle, on the opposite side of the zebra herd. After take-off, the drones quickly ascended to an altitude of approx. 100 m. From there, they were cruised, at a relatively constant altitude, towards the center of the herd, and then slowly descended up to approx. 10-20 m above the herd. This procedure was performed by the drone pilot by manually observing the drone and the zebra herd, simultaneously. Once the drones reached close to and above the zebras, they were flown following the zebras, keeping as many zebras as possible in their cameras' field of view. To this end, the pitch angle of the camera gimbal and the yaw orientation of the drone were manually controlled by the pilot. While the drone cameras did not have any optical zoom, digital zoom was also never performed. Even during the period of approach (ascend and initial cruise), the pilot attempted to keep the animals in the drone camera's field of view, however this was not always successful. The drones were kept following the herd until the battery levels approached a pre-determined threshold to return safely. At that point, the drones were manually flown back to the start position, near the vehicle. A complete flight was approx. sim 25 minutes, out of which the useful video recordings range from 5 to 20 minutes.

Metadata is provided in ZebraStereoID_metadata.xlsx. The coulmns in the xlsx are the following: - animal_id: ID label of the individual - date: date of the recording (220728 stands for 28th of July, 2022) - droneID: ID of the drones, as 2 were flying simultaneously - video_id: ID for the video out of the 4 - viewpoint: the body part that can be seen on the frames (no such info in this dataset) - frame_id: ID of the given video frame - image_width: width of the frame - image_height: height of the frame - bbox_x1: x coordinate to the top left corner of the bounding box - bbox_y1: y coordinate to the top left corner of the bounding box - bbox_x2: x coordinate to the bottom right corner of the bounding box - bbox_y2: y coordinate to the bottom right corner of the bounding box - bbox_width: width of the bounding box - bbox_height: height of the bounding box

Identifier
DOI https://doi.org/10.18419/DARUS-5957
Metadata Access https://darus.uni-stuttgart.de/oai?verb=GetRecord&metadataPrefix=oai_datacite&identifier=doi:10.18419/DARUS-5957
Provenance
Creator Ahmad, Aamir ORCID logo; Rubenstein, Daniel I. ORCID logo; Zábó, András ORCID logo
Publisher DaRUS
Contributor Ahmad, Aamir
Publication Year 2026
Funding Reference Cyber Valley CyVy-RF-2020-13
Rights CC BY-NC-SA 4.0; info:eu-repo/semantics/openAccess; http://creativecommons.org/licenses/by-nc-sa/4.0
OpenAccess true
Contact Ahmad, Aamir (University of Stuttgart)
Representation
Resource Type Dataset
Format application/zip; text/plain; text/tab-separated-values; application/vnd.openxmlformats-officedocument.spreadsheetml.sheet
Size 10170245210; 8498039994; 2485342; 294491287; 1442; 2587768; 1649094
Version 1.0
Discipline Biospheric Sciences; Computer Science; Computer Science, Electrical and System Engineering; Construction Engineering and Architecture; Earth and Environmental Science; Ecology; Engineering; Engineering Sciences; Environmental Research; Geosciences; Life Sciences; Natural Sciences