The aim of KFall dataset is to contribute technology development for elderly fall detection and injury prevention. It was acquired by 32 young subjects with 21 types of activities of daily living (ADLs) and 15 types of falls from an inertial sensor attached on low back. In total, it contains 5075 motion files with 2729 ADL motions and 2346 fall motions. In addition, for each fall motion, ready-to-use fall labels (fall initialization and fall impact moment) based on synchronized video references were also included.
sensor location
Experiment setup
Organization of sensor data
Description
Under sensor_data.zip file, there are 32 sub folders named by subject ID (e.g., SA06). Under each sub folder, it includes all the motion files from the corresponding subject. The naming rule of the motion file is explained here. Take "SA06T01R01.csv" as an example, "SA06" means Subject ID is 06; "T01" means Task ID is 01; "R01" means Trial ID is 01. Except Task 01, 11, 12 and 17, they are required for 1 trial while other tasks normally have 5 trials. However, due to Bluetooth disconnection, signal sychronization and miscounting issue, some tasks could have 4 trials or 6 trials (Note: SA01~SA05 were the subjects for pilot test, we excluded them. Due to system error, SA34's data was missing.). For each motion file (.csv), it contains 11 columns which are TimeStamp(s), FrameCounter, acceleration (unit: g); gyroscope (unit: °/s) and euler angle(°) along three axes.
Organization of label data
Description
Under label_data.zip file, there are 32 label files named by subject ID. For each label file, there are 6 columns: Task Code (Task ID), Description, Trial ID, Fall_onset_frame and Fall_Impact_frame. The last two columns are based on sychronized video reference. Take the first record in SA06_label.xlsx (header is not considered) as an example, it means the fall onset frame and fall impact frame are 130 and 208 in the corresponding data file (SA06T20R01.csv), respectively.
If you want to download the KFall dataset, please fill in the form below. The KFall team will review your request and verify your identity and institutional affiliation. Once this authentication process has been done, we will contact you at the email address as you registed and authorize access to the KFall dataset. At that time the download button would be enabled. Readers are not allowed to transfer the KFall dataset to any third-parties.
If you use this dataset, please cite as follows:
Yu X, Jang J, and Xiong S*, 2021. A Large-scale Open Motion Dataset (KFall) and Benchmark Algorithms for Detecting Pre-impact Fall of the Elderly Using Wearable Inertial Sensors. Frontiers in Aging Neuroscience.