Notes on the database (aka documentation)

Jump to downloads (on this page).
Jump to citation (on this page).

File name pattern

This section provides an overview of the file naming convention for database motion files and annotation files.

Database motion files

Each filename contains information about

For uncontrolled transport (experimental set-up 1) the filenames are set up as follows:
<objID>_ex<exNo>-rad-<trialID>-grasp-su<subjID>.txt

Example:

01_ex01-rad-002-grasp-su01.txt

For controlled transport (experimental setup 3) the filenames are set-up as follows:
<objID>_<graspID>_ex<exNo>-rad-<trialID>-grasp-su<subjID>.txt

Example:

01_11_ex03-1-rad-001-grasp-su04.txt

In both of them

objID
represents the ID of the object grasped,
graspID
stands for the ID of the grasp used to grasp the object,
exNo
abbreviates the number of the experiment,
trialID
labels the ID of the trial (basically a counter) and
subjID
is short for the ID of the subject who executed the motion

Annotation files

The annotation of a database file is stored in an additional text file. The name of this file consists of the name of the database file extended by "_annot", i.e. the annotation of the database file
    01_11_ex03-1-rad-004-grasp-su04.txt
is stored in the file
    01_11_ex03-1-rad-004-grasp-su04_annot.txt.

File content

In this section we will give details on the content structure of the database motion files and annotation files.

Database motion files - data ordering

A motion file consists of several rows of data. Each row represents one frame (also called a sample). Entries in a row of data are separated by a blank. Each data row consists of the entries visualised in figure 2.1 and are listed more extensively in table 2.1.

Visualisation of data ordering in a data row of a motion file. The sensor IDs 0, ..., 22 are consistent with table 2.1. Values are grouped according to meaning (date, thumb, fingers, and wrist/palm). Ungrouped entries do not explicetly belong th one of the groups (such as abduction/adduction sensors) and are left standing solitary.

In table 2.1, the sensor names are predeceded by a letter indicating the finger the joint belongs to. Here, T abbreviates thumb, I index finger, etc. The abduction/adduction sensors are between two fingers, hence carry both adjacent fingers, e.g. TI-AA stands for thumb-index finger adduction/abduction. Figure 2.1 shows a more compact representation of the data ordering in motion files also grouping sensors by fingers. For a schematic kinematic chain used as basic hand model see figure 2.2.

Examplary data row (raw 8bit data, extract)

2014 09 12 08 49 21.71200 1 101 114 128 107 136 94 0 0 119 94 0 ... YYYY MM DD HH mm ss.sssss S 0 1 2 3 4 5 6 7 8 9 10 ...
List of data row entries.
* value is 0 in an 18-sensor dataglove (for 8bit raw data) or value is inferred (for angle data)
** not supported ($\Rightarrow$ value 0)
data type (format) sensor ID key
year (YYYY) 1
month (MM) 2
day (DD) 3
hour (HH) 4
minute (mm) 5
seconds (ss.sssss) 6
Switch (bool) 7
T-TMCJ (float) 0 8
T-MCPJ (float) 1 9
T-IPJ (float) 2 10
TI-AA (float) 3 11
I-MCPJ (float) 4 12
I-PIJ (float) 5 13
I-DIJ (float) 6 * 14
I-AA (float) 7 ** 15
data type (format) sensor ID key
M-MCPJ (float) 8 16
M-PIJ (float) 9 17
M-DIJ (float) 10 18
MI-AA (float) 11 19
R-MCPJ (float) 12 20
R-PIJ (float) 13 21
R-DIJ (float) 14 * 22
RM-AA (float) 15 23
L-MCPJ (float) 16 24
L-PIJ (float) 17 25
L-DIJ (float) 18 * 26
LR-AA (float) 19 27
P-Arch (float) 20 28
W-FE (float) 21 29
W-AA (float) 22 30
Schematic kinematic chain of a model hand together positions and IDs of Cyberglove sensors. The abbreviations used for joints in the kinematic chain stand for (proximal, distal) interphalangeal joints (PIPJ, DIPJ, IPJ), metacarpophalangeal joint (MCPJ), carpometacarpal joint (CMCJ) and trapeziometacarpal joint (TMCJ or TMJ). Filled circles represent joints with two degrees of freedom (DoF) in that joint, unfilled circles represent joints with one DoF.
Cyberglove sensors are labeled with their ID (see
table 2.1) and marked by violet rectangles (joint sensors) or curves (ab-/adduction sensors, abbreviated AA).

Annotation files

Annotation files document the grasp phases (segments) in the database motion file. Each database file annotated was split into five phases: rest, reaching, grasp, retraction, rest. This information is captured in the annotation files: The first line is a comment starting with "#". The rest of the file contains one line per segment. Each line can be devided into three parts: first frame of the current segment, last frame of the current segment and name of the action. First and last frame together form the segment interval. The name of the action is preceeded by a dot (.) if the respective phase is considered to be more or less constant (fingers stay approximately in the same position).

The phases reaching and retraction are (mainly) named according to what is happening in that phase. Therefore the textual interval annotation may vary (i.e. reaching is represented by e.g. reach-overopen-closing, transition, closing, etc. and retraction may be called e.g. retract-opening, transition, opening etc.).

Exemplary annotation file:

# interval annotation 1 25 .rest 24 82 reach-overopen-closing 81 165 .grasp 164 234 retract-opening 233 282 .rest

Appendix

Here, a short overview on additional available material and information is given.

Information on the following matters is provided:

Picture of objects used for grasping

Objects used for grasping: classic notebook, oval jar, tennis ball, mug, cube, bottle crate, small cylinder, pen, bottle, glass, business card, bowl, cylinder large, carton (7.6 x 26 x 37.5[cm3]). Objects in red were used only in controlled transport.

Downloads

Download database including annotation files (.7z)
Download visualisation code (.7z)
Download segmentation code (.7z)

Citations

_____[segmentation]___________________________________

For citation of the database or the segmentation code please use

Katharina Stollenwerk, Anna Vögele, Björn Krüger, André Hinkenjann, and Reinhard Klein. Automatic temporal segmentation of articulated hand motion. In Lecture Notes in Computer Science, Computational Science and Its Applications (ICCSA). Vol. 9787, pp. 433 -- 449. July 2016.

bib (database) bib (segmentation code)  | pdf | web | Springer

_____[full database GraspDB14]________________________

As per January 2018, we are happy to also provide the full recorded database (GraspDB14) for download: Browse GraspDB14. It accompanies the technical report "GraspDB14 -- Documentation on a database of grasp motions and its creation". Should you find it usefull, please cite the following technical report

Katharina Stollenwerk, André Hinkenjann, and Reinhard Klein. GraspDB14 -- Documentation on a database of grasp motions and its creation. Technical Report No 01-2018, Hochschule Bonn-Rhein-Sieg - University of Applied Sciences, Department of Computer Science. issn 1869-5272. January 2018.

bib (database GraspDB) bib (technical report)  | web | publication server H-BRS