UMAFall dataset (Universidad de Málaga-FALL
& ADL repository) Eduardo Casilari (Coordinator), Jose-Antonio Santoyo-Ramón, Jose
Manuel Cano-García Departamento de Tecnología Electrónica ecasilari@uma.es April 2017 |
|
|
DESCRIPTION OF UMAFall DATASET
Eduardo Casilari, Jose A. Santoyo-Ramón, Jose M. Cano-García
Universidad de Málaga (Departamento de Tecnología Electrónica)
Departamento de Tecnología Electrónica, Universidad de Málaga, 29071 Malaga (Spain)
Telephone No. 34 952132755, Fax No. 34 952131447
June 2018
Abstract. This document describes the files
containing the mobility traces generated by a group of 19 experimental subjects
that emulated a set of predetermined ADL (Activities of Daily Life) and falls.
The traces are aimed at evaluating fall detection algorithms and can be freely
downloaded from:
http://webpersonal.uma.es/de/ECASILARI/Fall_ADL_Traces/UMA_FALL_ADL_dataset.html
1. Description of the
testbed
The initial experimental testbed (with 17
volunteers) was developed by Jose Antonio Santoyo-Román for his Msc. Thesis (presented in June 2016), which was supervised
by Eduardo Casilari, associate professor in the University of Malaga (Spain).
The second version of the previous dataset included two extra experimental
subjects and new types of ADL (in particular, several hand activities, such as
clapping hands, raising the hands, opening a door or making a phone call).These
new samples were generated in April 2017.
During the execution of the movements, the
subject transports a network consisting of five wireless nodes: an Android smartphone (which is located in a trouser pocket)
and four motes attached to different parts of the body (ankle, wrist, chest and
waist) through elastic bands, as it is illustrated in Figure 1.
Fig.1. Location of the
sensors (red arrows) and the smartphone (green arrow)
The motes were implemented in SimpleLink Multi-Standard CC2650 SensorTag
units of Texas Instruments, which are provided with a Bluetooth Low Energy
(BLE) interface and a multi-chip MPU-9250 module by InvenSense,
housing a tri-axis accelerometer, a triaxial gyroscope and a magnetometer.
During the tests, two different smartphone
models were employed. The models and characteristics of the built-in
accelerometers are presented in Table 1.
Smartphone Model |
Integrated
Accelerometer |
Range |
Resolution
|
Samsung
S5 |
MPU6500
(Invensense) |
±2 𝑔
|
6.103515·10−5
𝑔 |
LG
G4 |
LGE
Accelerometer (BOSCH) |
±16 𝑔
|
1.213651
·10−4 𝑔 |
Table 1. Characteristics of the employed
smartphones
The orientation of the sensors (SensorTag and Smartphones), which is sketched in Figure 2,
was the same for all the experiments. The picture indicates the orientation of
the sensors for the five considered locations when the subject's body is
standing up with the hands down. In the case of the wrist, x-axis is parallel
to the arm length.
Please remark that in the measurements captured
by SensorTags x-axis represents the direction that is
perpendicular to the floor while for the smartphone the equivalent axis is
y-axis.
Fig.2. Orientation of
the onbody sensors during the experiments
All the experiments were executed in a domestic
environment (see pictures in Figure 3), including a bedroom (A), a living room
(B) and scales in an apartment block. Falls were mimicked on a mattress on a
terrace.
Fig.3. Domestic environment
where the experiments took place
In the deployed architecture, as soon as every
experiment is initiated, the four Sensortags and the
Smartphone activate their embedded mobility sensors (accelerometer, gyroscope
and magnetometer), which periodically capture the corresponding magnitude. The
sampling rate in the smartphone was fixed to 200 Hz, while the employed rate in
the SensorTags was 20 Hz.
The program running in the SensorTags
sends via BLE the captured samples to a specific app running in the smartphone.
The app, which was specifically designed for the testbed, is in charge of
storing the measurements transmitted from the four SensorTags
as well as those captured by the smartphones. For every received sample, the
smartphone associates a timestamp and the Bluetooth MAC address of the mote
that transmitted it. For each experiment, all the samples from the five
mobility sensors (the four SensorTags and the
smartphone) are stored in a CSV (Comma Separated Value) file.
Thus, each CSV file includes the measurements
of the 5 sensors for a single movement (ADL or fall) executed by a particular
subject. All the movements are monitored during 15 seconds.
The whole dataset includes 746 files, which are
compressed in the file: UMAFall_Dataset.zip.
The typology of the executed movements can be
visualized (except for four types of ADLs, numbered 9 to 12 in the next
section) in a set of videos which can be downloaded from the same Web page of
the traces.
1. Information of the CSV file names
The name of the CSV file with the traces
indicates:
-The numerical
ID of the subject (from 1 to 19) that executed the movement. The personal
features (gender, age, height and weight) of each experimental subject are
presented in Table 2.
-The type
of the movement (ADL or FALL).
-The subtype
of movement (typology of the ADL or fall). The samples include 12 different
typologies of ADLs and 3 different types of falls.
·
Types of Executed
ADLs: 1) normal walking, 2) light jogging, 3) body bending,
4) hopping, 5) climbing stairs (up), 6) climbing stairs (down), 7) lying down
and getting up from a bed, 8) sitting down (and up) on (from) a chair, 9)
clapping hands (applauding), 10) raising the hands, 11) making a phone call,
12) opening a door.
·
Types of Emulated
Falls (on a mattress): 1) lateral, 2) frontal 3) backwards)
-The number of the trial of the same type and subtype executed by that user (as long
as subjects may repeat every movement up to 18 times).
-The date
(year, month, day) and time (hour,min,
sec.) in which the experiment was conducted.
Subject ID |
Gender |
Age |
Height (cm) |
Weight (kg) |
Subject 1 |
Female |
67 |
156 |
76 |
Subject 2 |
Female |
22 |
167 |
63 |
Subject 3 |
Male |
68 |
168 |
97 |
Subject 4 |
Male |
27 |
173 |
90 |
Subject 5 |
Male |
24 |
179 |
68 |
Subject 6 |
Male |
24 |
175 |
79 |
Subject 7 |
Male |
28 |
195 |
81 |
Subject 8 |
Female |
22 |
167 |
57 |
Subject 9 |
Male |
55 |
170 |
83 |
Subject 10 |
Male |
19 |
178 |
68 |
Subject 11 |
Male |
26 |
176 |
73 |
Subject 12 |
Female |
51 |
155 |
55 |
Subject 13 |
Female |
18 |
159 |
50 |
Subject 14 |
Female |
22 |
164 |
52 |
Subject 15 |
Male |
26 |
179 |
67 |
Subject 16 |
Male |
21 |
173 |
77 |
Subject 17 |
Female |
27 |
166 |
66 |
Subject 18 |
Male |
24 |
177 |
66 |
Subject 19 |
Female |
23 |
163 |
93 |
3. Content of the
files
Header. Every CSV file begins with a header
describing the characteristics of the experiment: the features of the Subject,
the type of movement (ADL, fall), a Boolean value indicating if the experiment
corresponds to a fall, the movement subtype, the number of the trial, the
number of employed sensors (5), the characteristics of the employed
accelerometers and the Bluetooth MAC addresses, ID and location of the five
nodes that integrate the network (the smartphone and the four SensorTags).
All the lines in the header begins with the
character ‘%’.
Traces
After the header, every line in the files
corresponds to a measurement captured by a particular mobility sensor of a
determined node (mote or SensorTag).
The format of the lines, which is also
explained in the file header, includes 7 numerical values separated by a
semicolon:
-The time (in ms)
since the experiment began.
-The number of the sample (for the same sensor
and node).
-The three real numbers describing the
measurements of the triaxial sensor (x-axis, y-axis and z-axis). The units are g,
°/s or µT depending on whether the
measurement was performed by an accelerometer, a gyroscope or a magnetometer,
respectively.
-An integer (0, 1 or 2) describing the type of
the sensor that originated the measurement (Accelerometer = 0
, Gyroscope = 1, Magnetometer = 2)
- An integer (from 0 to 4) informing about the
sensing node (the correspondence between this numerical code and the Bluetooth
MAC address and position of the motes is described in the file header).
The lines are ordered following the source
(sensor/mote) of the samples. So, the lines corresponding to the samples for
the same sensor and mote are presented at a stretch. So, two samples from
different sensors and/or motes that were measured at the same time are
separated in the files.
References
The source and authors of this publicly
available dataset should be acknowledged in all publications in which it is
utilized as by referencing any of the following papers as well as this
web-site:
·
Santoyo-Ramón,
José Antonio, Eduardo Casilari, and José Manuel Cano-García. "Analysis of
a smartphone-based architecture with multiple mobility sensors for fall
detection with supervised learning." Sensors 18.4 (2018): 1155.
·
Casilari,
Eduardo, Jose A. Santoyo-Ramón, and Jose M. Cano-García. "UMAFall: A Multisensor Dataset
for the Research on Automatic Fall Detection." Procedia Computer Science
110 (2017): 32-39.