Статьи

1. Deep Auto-Set: A Deep Auto-Encoder-Set Network for Activity Recognition Using Wearables

Используют сырые данных с акселерометров и гироскопов для построения карты активности с автоматическим именованием активностей разного рода. Применяют свёрточную сеть

 Datasets

• WISDM Actitracker dataset [21]: This dataset contains 1,098,207 triaxial accelerometer readings gathered from 36 users which reflect activity patterns of walking, jogging, sitting, standing, and climbing stairs. The acceleration measurements are collected with Android mobile phones at a constant sampling rate of 20Hz. We randomly select recordings from 8 users as the testing set and use the remaining data as our training and validation sets.

• Opportunity dataset [22]: This dataset comprises annotated recordings from a wide variety of on-body sensors configured on four subjects while carrying out morning activities. The annotations include several modes of locomotion along with a Null activity (referring to nonrelevant activities) which makes the recognition problem much more challenging. For data collection, subjects were instructed to perform five Activities of Daily Living (ADL) runs as well as a drill session with 20 repetitions of a predefined sequence of activities. Each sample in the resulting dataset corresponds to 113 real valued signal measurements recorded with a sampling rate of 30 Hz. We employ the same subset of data as in the Opportunity challenge [22] for training and testing purposes: ADL runs 4 and 5 collected from subjects 2 and 3 compose our testing set, and the remainder of the recordings from subjects 1,2 and 3 form our training and validation sets.


2. PerceptionNet: A Deep Convolutional Neural Network for Late Sensor Fusion

Статья об эффективном объединении данных различных датчиков или данных с разных осей одного датчика для последующей классификации активностей людей. Проверяют три способа: CNN, LSTM и PerceptionNet (их собственных метод). Также в статье присутствует обзор со ссылками на работы различных способов классификации активностей: A. Autoencoders, B. Convolutional Neural Networks, C. Deep learning on spectrogram, D. Convolutional Recurrent Neural Networks

Datasets:

1) UCL The UCL HAR dataset consists of tri-axial accelerometer and of tri-axial gyroscope sensor data, collected by a waistmounted smartphone (Samsung Galaxy S II smartphone). A group of 30 volunteers, with ages ranging from 19 to 48 years, executed six daily activities (standing, sitting, laying down, walking, walking downstairs and upstairs). The mobile sensors produced 3-axial linear acceleration and 3-axial angular velocity data with a sampling rate of 50 Hz and were segmented into time windows of 128 values (2.56 sec), having a 50% overlap. Furthermore, the dataset is separated into train data.

2) PAMAP2 The PAMAP2 HAR dataset contains 12 lifestyle activities (such as walking, cycling, ironing, etc.) from 9 participants wearing 3 Colibri wireless inertial measurement units (IMU) and a heart rate monitor. The 3 IMUs had a sampling frequency of 100Hz, were placed on the dominant arm, on the chest and on the dominant side's ankle, and produced tri-axial accelerometer, gyroscope and magnetometer data. In order to obtain the same sampling rate with the UCL dataset and the same sensor signals, we downsampled the PAMAP2 dataset to 50Hz and selected only the accelerometer and gyroscope data. The resulting dataset had 18 dimensions, with the same time window (2.56 sec) and overlap (50%) as the UCL dataset. 


3. Understanding and Improving Recurrent Networks for Human Activity Recognition by Continuous Attention

Применяют sensor и temporal attention mechanism для LSTM (RNN). Сравнивают точность распознавания различных LSTM моделей на трёх датасетах. Один из датасетов содержит действия людей с болезнью Паркинсона.

Datasets:

1) PAMAP2 dataset (Physical Activity Monitoring for Aging People 2)

2) Daphnet Gait (DG) dataset [2]. It contains recordings of 10 Parkinson’ disease (PD)patients instructed to perform activities that are likely to induce freezing of gait. Freezing of gait (FOG) is common in advanced PD, where affected patients struggle to initiate movements such as walking. The goal is to detect these freezing incidents.

3) Skoda Mini Checkpoint (Skoda) dataset [27], which describes the activities of assembly-line workers in a car production scenario. The dataset contains a worker wearing 19 accelerometers on both arms while performing 46 activities in the factory at one of the quality control checkpoints.


4. Estimation of Spatial-Temporal Gait Parameters based on the Fusion of Inertial and Film-Pressure Signals

Анализ особенностей походки по датчикам в ботинках


5. Wearable, Epidermal and Implantable Sensors for Medical Applications

Статья про различные носимые датчики и про их медицинские применения.

Примеры:

  • Девайс распознаёт в организме желание курить и delivers medication, чтобы он этого не делал.
  • Умные линзы от Гугл. Измеряют уровень глюкозы, что необходимо для людей с диабетом
  • Ткань одежды, которая распознаёт позу, дыхание, сердцебиение и снимает электрокардиограмму
  • Графеновые сенсоры

Вообще статья больше про оптимизацию расхода энергии сенсоров, их питание и про передачу данных с них.


6. RapidHARe: A computationally inexpensive method for real-time human activity recognition from wearable sensors

Очень похожая статья на первые, но она подробная. В ней сравниваются множество методов (RapidHARe, RapidHAReDF — предложенные методы, ANN, RNN и HMM)

Датасет

1) Использует свой датасет со множеством датчиков на ногах

To perform our experiments, we have recorded a total of 5 hours of data from 18 participants performing 8 different activities. These participants were healthy young adults: 4 females and 14 males with an average age of 23.67 years (standard deviation [STD]: 3.69), an average height of 179.06cm (STD: 9.85), and an average weight of 73.44kg (STD: 16.67). The participants performed a combination of activities at normal speed in a casual way, and there were no obstacles placed in their way. For instance, starting in the sitting position, the participant was instructed to perform the following activities: sitting, standing up, walking, going up the stairs, walking, sitting down.


7. Classify, predict, detect, anticipate and synthesize: Hierarchical recurrent latent variable models for human activity modeling

Ещё одна похожая на остальные статья о моделировании человеческой активности. Модель классифицирует и предвидит человеческую активность. Интересна очень странными датасетами

1) CAD-120: The CAD-120 dataset [18] consists of 4 subjectsperforming 10high-leveltasks, such as cleaning a microwave or having a meal, in 3 trials each. These activities are further annotated with 10 sub-actions, such as moving and eating and 12 object affordances, such as movable and openable.

2) UTKinect-Action3D Dataset: The UTKinect-Action3D Dataset (UTK) [33] consists of 10 subjects each recorded twice performing 10 actions in a row. The sequences are recorded with a kinect device (30 fps) and the extracted skeletons consist of 20 joints.

3) SBU Kinect Interaction Dataset: The SBU dataset [37] contains around 300 recordings of seven actors (21 pairs of two actors) performing eight different interactive activities such as hugging, pushing and shaking hands. The data was collected with a kinect device at 15 fps. While the dataset contains color and depth image, we make use of the 3D coordinates of 15 joints of each subject


8. Human activity recognition based on time series analysis using U-Net

Аналогичная всем остальным статья с кучей методов и кучей датасетов

Датасеты:

1) UCI HAR dataset is one of the most famous open datasets in the field of HAR, provided by University of California Irvine. Wearing the smartphone at the waist and using the built-in accelerometer and gyroscope, 30 volunteers aged between 19 and 48 attend the data collection. The data collected include a 3-axis acceleration time series and a 3-axis angular velocity time series. The sampling frequency is 50 Hz. The dataset contains 6 types of activities, respectively: WALKING, WALKING_UPSTAIRS, WALKING_DOWNSTAIRS, SITTING, STANDING, LAYING

2) UCI HAPT dataset UCI HAPT dataset is an extension of the UCI HAR dataset. The data were also collected from 30 volunteers aged 19 to 48 using a smartphone worn around the waist. The datasets include six basic types of activity in the UCI HAR dataset as well as six postural transitioning activities. Specifically, it is stand-to-sit, sit-to-stand, sit-to-lie, lie-to-sit, stand-to-lie and lie-to-stand

3) Sanitation dataset. The self-collected Sanitation dataset is collected from the open environment. A triaxial accelerometer worn in a wrist smart watch is used to collect seven types of daily work activity data of sanitation workers. The sampling frequency is 25 Hz. These seven types of activity are: walk, run, sweep, bweep (sweep using big broom), clean, dump and daily activities. The size of the whole dataset is 266555 x 3, which contains 266555 samples. Each sample contains X, Y and Z three axis acceleration values


Статьи, которые я просмотрел очень плохо, но они касаются именно медицинского приложения. И там есть что почитать. Может, по названию, эбстракту и заключению придут какие-то идеи

9. Simultaneous 12-Lead Electrocardiogram Synthesis using a Single-Lead ECG Signal: Application to Handheld ECG Devices

10. AMBULATORY ATRIAL FIBRILLATION MONITORING USING WEARABLE PHOTOPLETHYSMOGRAPHY WITH DEEP LEARNING

11. A Wearable IoT Aldehyde Sensor for Pediatric Asthma Research and Management

12. SPECMAR: Fast Heart Rate Estimation from PPG Signal using a Modified Spectral Subtraction Scheme with Composite Motion Artifacts Reference Generation

13. Recurrent Neural Networks based Obesity Status Prediction Using Activity Data

14. Parkinson’s Disease Assessment from a Wrist-Worn Wearable Sensor in Free-Living Conditions: Deep Ensemble Learning and Visualization

15. DeepHeart:Semi-Supervised Sequence Learning for Cardiovascular Risk Prediction

16. Latest Progresses in Developing Wearable Monitoring and Therapy Systems for Managing Chronic Diseases

17. Fog Computing in Medical Internet-of-Things: Architecture, Implementation, and Applications

18. HealthAdvisor: Recommendation System for Wearable Technologies enabling Proactive Health Monitoring

Важная статья

Авторы пытаются сделать рекомендательную систему, определяющую возможные заболевания человека по различным вносимым факторам о его возрасте/месте проживания/вредных привычках и истории болезней. Система рекомендует использовать определённые носимые устройства для отслеживания соответствующих болезням параметров

Machine learning classifier обучается на наборе входных данных и определяет болезни, которым может быть подвержен человек. the demographics information such as age, gender, location of residence, ethnicity, etc., as well as the person’s Electronic Medical Records (EMR).

"We obtained the following attributes from a publicly available data source [4] and evaluated it using the Weka library [11] for different models such as Decision Tree, Logistic Regression, LibSVM and OneR [12]. We ran it for 50 target classes (disease risks) and trained it for 135,000 data points [1]. We found that the Decision Tree Model had the lowest Root Mean Square Error"

[13] База данных носимых устройств. На этом сайте есть очень много интересных вещей

[4] Причины смерти в разных странах и оценка продолжительности жизни

[11] Средство анализа информации на сайте выше

Related Work

Genetic studies have been used in the past to predict and identify the top risk diseases for a person [5, 6, and 7]. Recent work on using machine learning algorithms for predicting individual disease risk has been investigated [8, 14, 15, and 16]