Out-of-Distribution Detection of Human Activity Recognition with Smartwatch Inertial Sensors.
Ontology highlight
ABSTRACT:
SUBMITTER: Boyer P
PROVIDER: S-EPMC7957807 | biostudies-literature |
REPOSITORIES: biostudies-literature
Similar Datasets
Project description:Mobile sensing has allowed the emergence of a variety of solutions related to the monitoring and recognition of human activities (HAR). Such solutions have been implemented in smartphones for the purpose of better understanding human behavior. However, such solutions still suffer from the limitations of the computing resources found on smartphones. In this sense, the HAR area has focused on the development of solutions of low computational cost. In general, the strategies used in the solutions are based on shallow and deep learning algorithms. The problem is that not all of these strategies are feasible for implementation in smartphones due to the high computational cost required, mainly, by the steps of data preparation and the training of classification models. In this context, this article evaluates a new set of alternative strategies based on Symbolic Aggregate Approximation (SAX) and Symbolic Fourier Approximation (SFA) algorithms with the purpose of developing solutions with low computational cost in terms of memory and processing. In addition, this article also evaluates some classification algorithms adapted to manipulate symbolic data, such as SAX-VSM, BOSS, BOSS-VS and WEASEL. Experiments were performed on the UCI-HAR, SHOAIB and WISDM databases commonly used in the literature to validate HAR solutions based on smartphones. The results show that the symbolic representation algorithms are faster in the feature extraction phase, on average, by 84.81%, and reduce the consumption of memory space, on average, by 94.48%, and they have accuracy rates equivalent to conventional algorithms.
| S-EPMC6263747 | biostudies-other
Project description:This paper defines two major data sets 1) from wearable inertial measurement sensors and 2) wearable ECG SHIMMER™ sensors. The first dataset is devised to benchmark techniques dealing with human behavior analysis based on multimodal inertial measurement wearable SHIMMER™ sensors unit during research studies "Fall Detection System for the Elderly Based on the Classification of Shimmer Sensor Prototype Data" [2] and "A novel fall detection algorithm for elderly using SHIMMER wearable sensors" [3]. The SHIMMER inertial sensor is a lightweight sensing device, incorporated with tri-axial accelerometer, a tri-axial gyroscope and tri-axial magnetometer, mounted on the waist of the subjects. The second dataset is developed to assess the feasibility of using SHIMMER™ wearable third generation ECG sensors for identification of basic heart anomalies by remote ECG analysis. The experimental protocol was carried out according to the Timed Up and Go (TUG) test [1], which is mainly used in fall detection and fall risk assessment systems specially designed for elderly. Three daily life activities such as standing still, walking and sitting on chair and getup were performed along with fall activity in controlled environment. This dataset is available on Data in Brief Dataverse [4] and a data repository [5].
| S-EPMC6920471 | biostudies-literature
Project description:This paper makes the VISTA database, composed of inertial and visual data, publicly available for gesture and activity recognition. The inertial data were acquired with the SensHand, which can capture the movement of wrist, thumb, index and middle fingers, while the RGB-D visual data were acquired simultaneously from two different points of view, front and side. The VISTA database was acquired in two experimental phases: in the former, the participants have been asked to perform 10 different actions; in the latter, they had to execute five scenes of daily living, which corresponded to a combination of the actions of the selected actions. In both phase, Pepper interacted with participants. The two camera point of views mimic the different point of view of pepper. Overall, the dataset includes 7682 action instances for the training phase and 3361 action instances for the testing phase. It can be seen as a framework for future studies on artificial intelligence techniques for activity recognition, including inertial-only data, visual-only data, or a sensor fusion approach.
| S-EPMC9117293 | biostudies-literature
Project description:Surfing is currently one of the most popular water sports in the world, both for recreational and competitive level surfers. Surf session analysis is often performed with commercially available devices. However, most of them seem insufficient considering the surfers’ needs, by displaying a low number of features, being inaccurate, invasive or not adequate for all surfer levels. Despite the fact that performing manoeuvres is the ultimate goal of surfing, there are no available solutions that enable the identification and characterization of such events. In this work, we propose a novel method to detect manoeuvre events during wave riding periods resorting solely to the inertial sensors embedded in smartphones. The proposed method was able to correctly identify over 95% of all the manoeuvres in the dataset (172 annotated manoeuvres), while achieving a precision of up to 80%, using a session-independent validation approach. These findings demonstrate the suitability and validity of the proposed solution for identification of manoeuvre events in real-world conditions, evidencing a high market potential.
| S-EPMC7256581 | biostudies-literature
Project description:This article describes a dataset for human activity recognition with inertial measurements, i.e., accelerometer and gyroscope, from a smartphone and a smartwatch placed in the left pocket and on the left wrist, respectively. Twenty-three heterogeneous subjects (μ = 44.3, σ = 14.3, 56% male) participated in the data collection, which consisted of performing five activities (seated, standing up, walking, turning, and sitting down) arranged in a specific sequence (corresponding with the TUG test). Subjects performed the sequence of activities multiple times while the devices collected inertial data at 100 Hz and were video-recorded by a researcher for data labelling purposes. The goal of this dataset is to provide smartphone- and smartwatch-based inertial data for human activity recognition collected from a heterogeneous (i.e., age-diverse, gender-balanced) set of subjects. Along with the dataset, the repository includes demographic information (age, gender), information about each sequence of activities (smartphone's orientation in the pocket, direction of turns), and a Python package with utility functions (data loading, visualization, etc). The dataset can be reused for different purposes in the field of human activity recognition, from cross-subject evaluation to comparison of recognition performance using data from smartphones and smartwatches.
| S-EPMC10700515 | biostudies-literature
Project description:Due to significant advances in sensor technology, studies towards activity recognition have gained interest and maturity in the last few years. Existing machine learning algorithms have demonstrated promising results by classifying activities whose instances have been already seen during training. Activity recognition methods based on real-life settings should cover a growing number of activities in various domains, whereby a significant part of instances will not be present in the training data set. However, to cover all possible activities in advance is a complex and expensive task. Concretely, we need a method that can extend the learning model to detect unseen activities without prior knowledge regarding sensor readings about those previously unseen activities. In this paper, we introduce an approach to leverage sensor data in discovering new unseen activities which were not present in the training set. We show that sensor readings can lead to promising results for zero-shot learning, whereby the necessary knowledge can be transferred from seen to unseen activities by using semantic similarity. The evaluation conducted on two data sets extracted from the well-known CASAS datasets show that the proposed zero-shot learning approach achieves a high performance in recognizing unseen (i.e., not present in the training dataset) new activities.
| S-EPMC7038698 | biostudies-literature
Project description:By developing awareness of smartphone activities that the user is performing on their smartphone, such as scrolling feeds, typing and watching videos, we can develop application features that are beneficial to the users, such as personalization. It is currently not possible to access real-time smartphone activities directly, due to standard smartphone privileges and if internal movement sensors can detect them, there may be implications for access policies. Our research seeks to understand whether the sensor data from existing smartphone inertial measurement unit (IMU) sensors (triaxial accelerometers, gyroscopes and magnetometers) can be used to classify typical human smartphone activities. We designed and conducted a study with human participants which uses an Android app to collect motion data during scrolling, typing and watching videos, while walking or seated and the baseline of smartphone non-use, while sitting and walking. We then trained a machine learning (ML) model to perform real-time activity recognition of those eight states. We investigated various algorithms and parameters for the best accuracy. Our optimal solution achieved an accuracy of 78.6% with the Extremely Randomized Trees algorithm, data sampled at 50 Hz and 5-s windows. We conclude by discussing the viability of using IMU sensors to recognize common smartphone activities.
| S-EPMC7038357 | biostudies-literature
Project description:BACKGROUND:Wearable accelerometers have greatly improved measurement of physical activity, and the increasing popularity of smartwatches with inherent acceleration data collection suggest their potential use in the physical activity research domain; however, their use needs to be validated. OBJECTIVE:This study aimed to assess the validity of accelerometer data collected from a Samsung Gear S smartwatch (SGS) compared with an ActiGraph GT3X+ (GT3X+) activity monitor. The study aims were to (1) assess SGS validity using a mechanical shaker; (2) assess SGS validity using a treadmill running test; and (3) compare individual activity recognition, location of major body movement detection, activity intensity detection, locomotion recognition, and metabolic equivalent scores (METs) estimation between the SGS and GT3X+. METHODS:To validate and compare the SGS accelerometer data with GT3X+ data, we collected data simultaneously from both devices during highly controlled, mechanically simulated, and less-controlled natural wear conditions. First, SGS and GT3X+ data were simultaneously collected from a mechanical shaker and an individual ambulating on a treadmill. Pearson correlation was calculated for mechanical shaker and treadmill experiments. Finally, SGS and GT3X+ data were simultaneously collected during 15 common daily activities performed by 40 participants (n=12 males, mean age 55.15 [SD 17.8] years). A total of 15 frequency- and time-domain features were extracted from SGS and GT3X+ data. We used these features for training machine learning models on 6 tasks: (1) individual activity recognition, (2) activity intensity detection, (3) locomotion recognition, (4) sedentary activity detection, (5) major body movement location detection, and (6) METs estimation. The classification models included random forest, support vector machines, neural networks, and decision trees. The results were compared between devices. We evaluated the effect of different feature extraction window lengths on model accuracy as defined by the percentage of correct classifications. In addition to these classification tasks, we also used the extracted features for METs estimation. RESULTS:The results were compared between devices. Accelerometer data from SGS were highly correlated with the accelerometer data from GT3X+ for all 3 axes, with a correlation ?.89 for both the shaker test and treadmill test and ?.70 for all daily activities, except for computer work. Our results for the classification of activity intensity levels, locomotion, sedentary, major body movement location, and individual activity recognition showed overall accuracies of 0.87, 1.00, 0.98, 0.85, and 0.64, respectively. The results were not significantly different between the SGS and GT3X+. Random forest model was the best model for METs estimation (root mean squared error of .71 and r-squared value of .50). CONCLUSIONS:Our results suggest that a commercial brand smartwatch can be used in lieu of validated research grade activity monitors for individual activity recognition, major body movement location detection, activity intensity detection, and locomotion detection tasks.
| S-EPMC6386649 | biostudies-literature
Project description:Although several studies have used wearable sensors to analyze human lifting, this has generally only been done in a limited manner. In this proof-of-concept study, we investigate multiple aspects of offline lift characterization using wearable inertial measurement sensors: detecting the start and end of the lift and classifying the vertical movement of the object, the posture used, the weight of the object, and the asymmetry involved. In addition, the lift duration, horizontal distance from the lifter to the object, the vertical displacement of the object, and the asymmetric angle are computed as lift parameters. Twenty-four healthy participants performed two repetitions of 30 different main lifts each while wearing a commercial inertial measurement system. The data from these trials were used to develop, train, and evaluate the lift characterization algorithms presented. The lift detection algorithm had a start time error of 0.10 s ± 0.21 s and an end time error of 0.36 s ± 0.27 s across all 1489 lift trials with no missed lifts. For posture, asymmetry, vertical movement, and weight, our classifiers achieved accuracies of 96.8%, 98.3%, 97.3%, and 64.2%, respectively, for automatically detected lifts. The vertical height and displacement estimates were, on average, within 25 cm of the reference values. The horizontal distances measured for some lifts were quite different than expected (up to 14.5 cm), but were very consistent. Estimated asymmetry angles were similarly precise. In the future, these proof-of-concept offline algorithms can be expanded and improved to work in real-time. This would enable their use in applications such as real-time health monitoring and feedback for assistive devices.
| S-EPMC7219665 | biostudies-literature
Project description:BACKGROUND:Time-resolved quantification of physical activity can contribute to both personalized medicine and epidemiological research studies, for example, managing and identifying triggers of asthma exacerbations. A growing number of reportedly accurate machine learning algorithms for human activity recognition (HAR) have been developed using data from wearable devices (eg, smartwatch and smartphone). However, many HAR algorithms depend on fixed-size sampling windows that may poorly adapt to real-world conditions in which activity bouts are of unequal duration. A small sliding window can produce noisy predictions under stable conditions, whereas a large sliding window may miss brief bursts of intense activity. OBJECTIVE:We aimed to create an HAR framework adapted to variable duration activity bouts by (1) detecting the change points of activity bouts in a multivariate time series and (2) predicting activity for each homogeneous window defined by these change points. METHODS:We applied standard fixed-width sliding windows (4-6 different sizes) or greedy Gaussian segmentation (GGS) to identify break points in filtered triaxial accelerometer and gyroscope data. After standard feature engineering, we applied an Xgboost model to predict physical activity within each window and then converted windowed predictions to instantaneous predictions to facilitate comparison across segmentation methods. We applied these methods in 2 datasets: the human activity recognition using smartphones (HARuS) dataset where a total of 30 adults performed activities of approximately equal duration (approximately 20 seconds each) while wearing a waist-worn smartphone, and the Biomedical REAl-Time Health Evaluation for Pediatric Asthma (BREATHE) dataset where a total of 14 children performed 6 activities for approximately 10 min each while wearing a smartwatch. To mimic a real-world scenario, we generated artificial unequal activity bout durations in the BREATHE data by randomly subdividing each activity bout into 10 segments and randomly concatenating the 60 activity bouts. Each dataset was divided into ~90% training and ~10% holdout testing. RESULTS:In the HARuS data, GGS produced the least noisy predictions of 6 physical activities and had the second highest accuracy rate of 91.06% (the highest accuracy rate was 91.79% for the sliding window of size 0.8 second). In the BREATHE data, GGS again produced the least noisy predictions and had the highest accuracy rate of 79.4% of predictions for 6 physical activities. CONCLUSIONS:In a scenario with variable duration activity bouts, GGS multivariate segmentation produced smart-sized windows with more stable predictions and a higher accuracy rate than traditional fixed-size sliding window approaches. Overall, accuracy was good in both datasets but, as expected, it was slightly lower in the more real-world study using wrist-worn smartwatches in children (BREATHE) than in the more tightly controlled study using waist-worn smartphones in adults (HARuS). We implemented GGS in an offline setting, but it could be adapted for real-time prediction with streaming data.
| S-EPMC6386646 | biostudies-literature