Very low resolution depth images of 200,000 poses - Open Research Data - Bridge of Knowledge

Your account

login

Search

Very low resolution depth images of 200,000 poses

Description

A dataset represents simulated images of depth sensor seeing a single human pose, performing 200,000 random gestures. The depth images as vectors of pixels are stored with ground truth positions of every relevant joint.

3D graphics software (Blender 3D) was used to model, pose and render human figure, creating depth image and appropriate ground truth for each pose. It was assumed, that the sensor has a horizontal field of view 57° (same as Kinect), and the depth image was rendered to 60×50 pixels in 8-bit greyscale (3-bit less, i.e. 8 times lower depth resolution than Kinect).

The human figure was positioned in an empty space, without a background and other foreground objects. In real applications to obtain the same conditions a separation of foreground object based on depth information should be performed, e.g. by assuming the body to be the object closest to a camera, and discarding the ones positioned further by comparing their depth values.

The dataset represents a wide variation of the upper body poses with biologically correct random joints positions, with wrists positions uniformly covering the available space in front of the figure. The body absolute position was changed randomly to introduce shifts in x, y and z directions relative to the fixed sensor. The movement ranges were limited to allow only far reaching hands to leave the frame, whilst the rest of the figure remained visible. 200,000 depth images were created.

The applications of the database include training of object detection and tracking algorithm for body pose recognition.

Illustration of the publication

Dataset file

fullBody-exclFeet-correctRanges-SymFullRot200k.zip
388.0 MB, S3 ETag 86b55c56237dbbd8d06e4cbdf81b46c1-1, downloads: 20
The file hash is calculated from the formula
hexmd5(md5(part1)+md5(part2)+...)-{parts_count} where a single part of the file is 512 MB in size.

Example script for calculation:
https://github.com/antespi/s3md5
download file fullBody-exclFeet-correctRanges-SymFullRot200k.zip

File details

License:
Creative Commons: by 4.0 open in new tab
CC BY
Attribution

Details

Year of publication:
2020
Verification date:
2020-12-17
Dataset language:
English
Fields of science:
  • information and communication technology (Engineering and Technology)
  • biomedical engineering (Engineering and Technology)
DOI:
DOI ID 10.34808/84xp-vz47 open in new tab
Verified by:
Gdańsk University of Technology

Keywords

References

Cite as

Authors

seen 182 times