Database of Human Action Recognition

Data Specification:

1. Eight subjects including five males and three females are recruited for the database, and their ages     range from 24 to 35 years old. 

2. Each wireless inertial sensor node consists of a tri-axial accelerometer, a dual-axis gyroscope and a     single-axis gyroscope. Sampling frequency is 50 Hz.

    The sensor node used for data acquisition is shown as follows:

3. Five inertial sensor nodes are attached to the body parts of one subject including right wrist, left arm,     waist, right ankle and left thigh respectively, as shown in the following figure.

4. As the following list, data of ten different activities are acquired.

    A1--"Walk at the speed of 3 km/h",

    A2--"Walk at the speed of 5 km/h",

    A3--"Run at the speed of 6 km/h",

    A4--"Run at the speed of 8 km/h",

    A5--"Run at the speed of 12 km/h",

    A6--"Go downhill",

    A7--"Climb the hill",

    A8--"Practice gymnastics",

    A9--"Rope skipping",

    A10--"Cycle".

    Note: A1-A5 are executed on the treadmill, A6 and A7 are collected  under simulated condition   on     campus, A9 and A10 are implemented in areal environment. Each action lasts 180 seconds.

5. The name for action data file is 'sitj.mat', where 'si' represents the ith serial number of sensor node,     and 'tj' represents the jth time.

Database of Swimming Research

Data Specification:

1. Four subjects including 3 males and 1 females are recruited for the dataset of swimming research,and     their ages range from 20 to 27 years old. Also the dataset contain the documents of the comparison     experiment. 

2. Dataset consist of raw data and matlab code which are used to display the results.

3. Each measurement node contains a 9-axis sensor. The highest sample frequency can reach 400 Hz.     The hardware consists of measurement nodes, a receiving node,and a personal computer (PC), shown     as follow:

4. Current research mainly focus on the accuracy of orientation estimation algorithm for capturing     swimming postures, and compare the individual differences in swimming posture. Hence, we attach    measurement node on the lumbar of swimming athletes, measurement node on this position of     swimmer body numbered ‘ node 2 ’, as shown in the following figure (a).

5. Currently, we have carried out the comparison experiment in lab environment, which the     performance of wearable motion capture system is compared with NDI motion tracking system. we     also tested our wearable motion capture system in swimming pool. In the future, we will capture the     posture of athletes’ whole-body motion in swimming training, hence, we are testing Vicon motion      tracking system for further validating our system, shown in above figure (b).

6. As the following list, data of four competitive swimming styles are acquired.

    bc—"swim back crawl at self-selected training speeds",

    bf—"swim butterfly at self-selected training speeds",

    bs—"swim breaststroke at self-selected training speeds",

    fs—"swim free style at self-selected training speeds",

7. The folder named “NDI comparison experiment” contain the data for our comparison experiment.     

8. Please download IEEE paper to see more details "Using Wearable Sensors to Capture Posture of the     Human Lumbar Spine in Competitive Swimming", DOI: 10.1109/THMS.2019.2892318     

Database of Affective Actions Recognition in Dyadic Interactions

Data Specification:

1. Eleven volunteers (8 male and 3 female volunteers) are recruited for the database, and their ages range from 25 to 35 years old.

2. The experiment platform used including several sensor nodes and a sink node. Each sensor node contains a triaxial accelerometer, a triaxial gyroscope, and a triaxial magnetometer. Sampling frequency is 100 Hz.

    The sensor node and sink node used are shown as follows:

3. Five inertial sensor nodes are attached to the body parts of one subject including back, left arm, right arm, left wrist and right wrist respectively, as shown in the following figure.

4. The dyadic interactions are divided into two predefined situations: positive situation and negative situation. All the collected actions that related to the emotions of two interaction situations are shown in the following Table.

5. The name for action data file is “subjectI.mat”, where 'I' represents number of the subject. Each element of the data contains the data of five sensor nodes. The first three columns of each sensor node are the acceleration data, fourth, fifth and sixth columns are the data collected from gyroscope, and the last three columns are data of magnetometer. Each action data file also contains the corresponding label “LABEL.mat”.

    Label Specification is as follows:

    A1--Hold both palms out,

    A2--Thumb up(right hand),

    A3--Forward leans,

    A4--Hands on one’s head,

    A5--Arms akimbo,

    A6--Crossed arms,

    A7--Neck touching,

    A8--Thumbs down(both hands),

CONTACE US

For download database, please contact qiu(at)mail(dot)dlut(dot)edu(dot)cn.