Student or educator? Sign up with your school email and get 6 months of Pro free.
CoChat

The DOI matches its Crossref record; 30 of the 30 references it cites were found in a registry or an index.

A framework to automatically detect near-falls using a wearable inertial measurement cluster

Open access

Abstract

Accurate and automatic assessments of body segment kinematics via wearable sensors are essential to provide new insights into the complex interactions between active lifestyle and fall risk in various populations. To remotely assess near-falls due to balance disturbances in daily life, current approaches primarily rely on biased questionnaires, while contemporary data-driven research focuses on preliminary fall-related scenarios. Here, we worked on an automated framework based on accurate trunk kinematics, enabling the detection of near-fall scenarios during locomotion. Using a wearable inertial measurement cluster in conjunction with evaluation algorithms focusing on trunk angular acceleration, the proposed sensor-framework approach revealed accurate distinguishment of balance disturbances related to trips and slips, thereby minimising false detections during activities of daily living. An important factor contributing to the framework’s high sensitivity and specificity for automatic detection of near-falls was the consideration of the individual’s gait characteristics. Therefore, the sensor-framework presents an opportunity to substantially impact remote fall risk assessment in healthy and pathological conditions outside the laboratory.

Overview

The claims and statistics of this paper haven’t been checked yet.

Free: it reads the open-access text and checks the first 5 claims that cite a source we can check. Create a free account to check every claim.

A few free checks a day · papers already verified open instantly · results are public

Integrity signals

Registry

Crossref · 10.1038/s44172-024-00325-x

Cites

30 of 30 references verified0 citation errors · 0 not checkable

Statistics

Figures

Fig. 1 Matrix scheme of the automated distinguishment and near-fall type classifications during locomotion by the framework. The results of the framework in combination with the inertial measurement cluster (IMC) measurements for each of the 627 trials (y-axis) are represented by black crosses and plotted against the corresponding original exercise (x-axis). The framework’s results using a single inertial measurement unit (IMU) measurements for each of the 627 trials (y-axis) are represented by blue crosses. Red crosses indicate that more perturbed locomotion events were identified than were actually applied in the respective IMU measurement trials. The corresponding classification of the framework is shown on the y-axis ordered in anterior loss of balance; posterior loss of balance; trip; slip; ADLs (from bottom to top). ADLs refer to the summary of all simulated daily activities. The corresponding original exercise (x-axis) is indicated by stick figures (from left to right): anterior loss of balance; posterior loss of balance; trip; slip; pick-and-drop; staircase walking; sit-to-stand; walking; running. The main diagonal of the matrix (shaded in grey) represents correct classifications.
Fig. 1 Matrix scheme of the automated distinguishment and near-fall type classifications during locomotion by the framework. The results of the framework in combination with the inertial measurement cluster (IMC) measurements for each of the 627 trials (y-axis) are represented by black crosses and plotted against the corresponding original exercise (x-axis). The framework’s results using a single inertial measurement unit (IMU) measurements for each of the 627 trials (y-axis) are represented by blue crosses. Red crosses indicate that more perturbed locomotion events were identified than were actually applied in the respective IMU measurement trials. The corresponding classification of the framework is shown on the y-axis ordered in anterior loss of balance; posterior loss of balance; trip; slip; ADLs (from bottom to top). ADLs refer to the summary of all simulated daily activities. The corresponding original exercise (x-axis) is indicated by stick figures (from left to right): anterior loss of balance; posterior loss of balance; trip; slip; pick-and-drop; staircase walking; sit-to-stand; walking; running. The main diagonal of the matrix (shaded in grey) represents correct classifications.CC BY · PubMed Central
Fig. 2 Matrix scheme of the automated classification of detected perturbations during locomotion by the framework. The results are divided into the classifications relying on the inertial measurement cluster (IMC; black crosses) and inertial measurement unit (IMU; blue crosses) kinematics. The applied types of perturbations are presented on the bottom axis with stick figures (from left to right): anterior loss of balance; posterior loss of balance; trip; slip. The y-axis (left) indicates the classification results, i.e. whether the initial trunk movement was directed anteriorly or posteriorly. The y-axis (right) specifies the identified preceding locomotion with respect to the perturbation onset. The main diagonal of the matrix (shaded in grey) represents correct classifications.
Fig. 2 Matrix scheme of the automated classification of detected perturbations during locomotion by the framework. The results are divided into the classifications relying on the inertial measurement cluster (IMC; black crosses) and inertial measurement unit (IMU; blue crosses) kinematics. The applied types of perturbations are presented on the bottom axis with stick figures (from left to right): anterior loss of balance; posterior loss of balance; trip; slip. The y-axis (left) indicates the classification results, i.e. whether the initial trunk movement was directed anteriorly or posteriorly. The y-axis (right) specifies the identified preceding locomotion with respect to the perturbation onset. The main diagonal of the matrix (shaded in grey) represents correct classifications.CC BY · PubMed Central
Fig. 3 Schematic illustration of the inertial measurement cluster (IMC) and the applied perturbation paradigms. The model in (a) shows the IMC's attachment to the participant’s trunk using adjustable straps. The Xsens MTi-20 VRU IMUs were mounted on a rigid base and are represented by the orange boxes (a). Loss of balance in anterior or posterior directions during standing was initiated via a sudden release from a static, inclined position (lean-and-release task (b)). The inclination was maintained by a supporting cable horizontally attached to a belt around the participant’s pelvis, with the other end connected to a pneumatically driven brake-and-release system. During walking, we provided unpredictable trip- and slip-like perturbations (c). These perturbations were introduced by a pneumatically driven brake-and-release device with Teflon cables attached to the ankles, or by electronically triggered tripping and slipping elements (represented by the black elements (c)). The tripping element was activated upon touchdown of the trailing leg, while the anterior shift (slipping) was initiated upon touchdown of the perturbed leg onto the respective frame. For all tasks, the perturbation magnitudes varied from moderate to strong by adjusting the gait velocity (trips and slips) or changing the body’s inclination (lean-and-release task). Safety harnesses were worn during all tasks to prevent any part of the body, except the feet, from contacting the ground.
Fig. 3 Schematic illustration of the inertial measurement cluster (IMC) and the applied perturbation paradigms. The model in (a) shows the IMC's attachment to the participant’s trunk using adjustable straps. The Xsens MTi-20 VRU IMUs were mounted on a rigid base and are represented by the orange boxes (a). Loss of balance in anterior or posterior directions during standing was initiated via a sudden release from a static, inclined position (lean-and-release task (b)). The inclination was maintained by a supporting cable horizontally attached to a belt around the participant’s pelvis, with the other end connected to a pneumatically driven brake-and-release system. During walking, we provided unpredictable trip- and slip-like perturbations (c). These perturbations were introduced by a pneumatically driven brake-and-release device with Teflon cables attached to the ankles, or by electronically triggered tripping and slipping elements (represented by the black elements (c)). The tripping element was activated upon touchdown of the trailing leg, while the anterior shift (slipping) was initiated upon touchdown of the perturbed leg onto the respective frame. For all tasks, the perturbation magnitudes varied from moderate to strong by adjusting the gait velocity (trips and slips) or changing the body’s inclination (lean-and-release task). Safety harnesses were worn during all tasks to prevent any part of the body, except the feet, from contacting the ground.CC BY · PubMed Central
Fig. 4 Schematic overview of the automated framework for detecting and distinguishing balance recovery responses. This schematic outlines the step-by-step process of the proposed framework designed to detect anomalies in trunk kinematics, with a focus on identifying balance recovery responses to perturbations and classifying the type of perturbation. Step (a) shows the remote monitoring of the trunk kinematics using the wearable inertial measurement cluster (IMC) placed on the participant’s trunk during locomotion tasks that include both normal activities of daily living (ADLs) and externally applied perturbations (e.g. trips, slips). Step (b) is the automated determination of personalised thresholds to take into account individual gait characteristics. In Step (c), the observation windows were detected based on the personalised thresholds and a simultaneous analysis of the trunk angular velocity and acceleration vector coordinate along the transverse axis, identifying time periods containing potential anomalies (necessary condition). In Step (d), the observation windows were analysed for frequency content, periodicity and outliers to assess whether the detected anomaly indicates a sufficient condition for a balance perturbation. Finally, in Step (e), the framework classifies the locomotor task and the type of perturbation, identifying the specific perturbation types.
Fig. 4 Schematic overview of the automated framework for detecting and distinguishing balance recovery responses. This schematic outlines the step-by-step process of the proposed framework designed to detect anomalies in trunk kinematics, with a focus on identifying balance recovery responses to perturbations and classifying the type of perturbation. Step (a) shows the remote monitoring of the trunk kinematics using the wearable inertial measurement cluster (IMC) placed on the participant’s trunk during locomotion tasks that include both normal activities of daily living (ADLs) and externally applied perturbations (e.g. trips, slips). Step (b) is the automated determination of personalised thresholds to take into account individual gait characteristics. In Step (c), the observation windows were detected based on the personalised thresholds and a simultaneous analysis of the trunk angular velocity and acceleration vector coordinate along the transverse axis, identifying time periods containing potential anomalies (necessary condition). In Step (d), the observation windows were analysed for frequency content, periodicity and outliers to assess whether the detected anomaly indicates a sufficient condition for a balance perturbation. Finally, in Step (e), the framework classifies the locomotor task and the type of perturbation, identifying the specific perturbation types.CC BY · PubMed Central
Fig. 5 Framework’s results of the automatically detected balance recovery response in a typical locomotion sequence of daily life. In (a), the raw data of the trunk’s angular velocity ωy and in (b), the angular acceleration component αy around the transversal axis (y-axis) are shown for perturbed walking, pick-and-drop task and unperturbed walking. The corresponding locomotion task is indicated with the stick figures on top of the raw data section. The black vertical lines highlight the automatically detected balance recovery response. Anomalies appeared in ωy for both perturbed locomotion and typical activities like object manipulation. However, by jointly analysing the trunk’s ωy and αy components, it became clear that only perturbed locomotion exhibits anomalies in both kinematic signals. Note that during time frame ≈ 110–115 s, the gait velocity was lower in relation to the time frame ≈ 140–145 s leading to different signal characteristics.
Fig. 5 Framework’s results of the automatically detected balance recovery response in a typical locomotion sequence of daily life. In (a), the raw data of the trunk’s angular velocity ωy and in (b), the angular acceleration component αy around the transversal axis (y-axis) are shown for perturbed walking, pick-and-drop task and unperturbed walking. The corresponding locomotion task is indicated with the stick figures on top of the raw data section. The black vertical lines highlight the automatically detected balance recovery response. Anomalies appeared in ωy for both perturbed locomotion and typical activities like object manipulation. However, by jointly analysing the trunk’s ωy and αy components, it became clear that only perturbed locomotion exhibits anomalies in both kinematic signals. Note that during time frame ≈ 110–115 s, the gait velocity was lower in relation to the time frame ≈ 140–145 s leading to different signal characteristics.CC BY · PubMed Central
Fig. 6 Schematic illustration of trunk kinematics and the subroutines. In (a) and (b), the perturbed and unperturbed trunk angular acceleration about the transverse axis αy are represented, respectively. In (c–f), the subroutines for the data processing used to automatically detect balance recovery responses are visualised. In (a), one result of an automatically detected balance recovery response for a trip is illustrated. In (b), the global extrema and specific local maxima of an unperturbed baseline walking are shown (blue and red circles, respectively; where the local maxima are a consequence of a touchdown). In c, the method to calculate the autocorrelation of the same measurement trial and in d, an exemplary segmentation of the kinematics curve path for the local extrema identifier algorithm is represented. In (e), method 2 for the outlier identification is shown for a given subset T, and f represents a histogram of the peak distributions analysis.
Fig. 6 Schematic illustration of trunk kinematics and the subroutines. In (a) and (b), the perturbed and unperturbed trunk angular acceleration about the transverse axis αy are represented, respectively. In (c–f), the subroutines for the data processing used to automatically detect balance recovery responses are visualised. In (a), one result of an automatically detected balance recovery response for a trip is illustrated. In (b), the global extrema and specific local maxima of an unperturbed baseline walking are shown (blue and red circles, respectively; where the local maxima are a consequence of a touchdown). In c, the method to calculate the autocorrelation of the same measurement trial and in d, an exemplary segmentation of the kinematics curve path for the local extrema identifier algorithm is represented. In (e), method 2 for the outlier identification is shown for a given subset T, and f represents a histogram of the peak distributions analysis.CC BY · PubMed Central
Fig. 7 Schematic illustration of the procedure to automatically determine observation windows. The procedure is divided into three steps (a–c), based on joint analysis to identify anomalies in the trunk kinematics. In (a), the detected anomalies in the trunk’s angular velocity component around the transverse axis ωy are clustered in temporary observation windows. An anomaly was defined as personalised threshold exceedance. In b, the combined analysis of ωy and the trunk’s angular acceleration component around the transverse axis αy is shown. In this step, we check whether there are anomalies in αy present in the extended observation windows (red vertical lines). In (c), the perturbation onset and the offset of the balance recovery response are determined for all extended observation windows containing anomalies in both kinematics, indicated by grey vertical lines. The offset determination takes into account the displayed threshold bandwidths.
Fig. 7 Schematic illustration of the procedure to automatically determine observation windows. The procedure is divided into three steps (a–c), based on joint analysis to identify anomalies in the trunk kinematics. In (a), the detected anomalies in the trunk’s angular velocity component around the transverse axis ωy are clustered in temporary observation windows. An anomaly was defined as personalised threshold exceedance. In b, the combined analysis of ωy and the trunk’s angular acceleration component around the transverse axis αy is shown. In this step, we check whether there are anomalies in αy present in the extended observation windows (red vertical lines). In (c), the perturbation onset and the offset of the balance recovery response are determined for all extended observation windows containing anomalies in both kinematics, indicated by grey vertical lines. The offset determination takes into account the displayed threshold bandwidths.CC BY · PubMed Central
Fig. 8 Schematic illustration of the subsequent data analysis. The flowchart of the subsequent data analysis is represented in (a) and the underlying steps to classify an observation window as a balance recovery response or daily activity in (b–g). The subsequent data analysis was divided in three steps. In each step, we analysed the detected observation window (necessary condition) and determined whether it contained additional predefined criteria (i.e. sufficient condition). The curve sections displayed in (b, c) represent a sample observation window identified for a running trail. Further, b, c show the analyses regarding periodic pattern or transient irregular frequency content including anomalies with outlier characteristics (step 1) for the coordinates around the transverse axis of the trunk’s angular acceleration αy and velocity ωy. Considering step 2, d shows a schematic of irregular frequency content evaluated by the autocorrelation coefficient peaks rk,αy of a balance recovery response, where e shows an expected (regular) frequency content with recognisable periodic patterns of unperturbed walking. In f, g, the trunk kinematics αy and ωy for the same balance recovery response are shown. The red lines indicate the identified observation window, the blue dotted lines indicate the time window (±0.15 s) centred around the local maxima peak with outlier characteristic. In g, the same time window is highlighted (blue dotted lines), i.e. the analysis of whether there also exists an outlier (step 3).
Fig. 8 Schematic illustration of the subsequent data analysis. The flowchart of the subsequent data analysis is represented in (a) and the underlying steps to classify an observation window as a balance recovery response or daily activity in (b–g). The subsequent data analysis was divided in three steps. In each step, we analysed the detected observation window (necessary condition) and determined whether it contained additional predefined criteria (i.e. sufficient condition). The curve sections displayed in (b, c) represent a sample observation window identified for a running trail. Further, b, c show the analyses regarding periodic pattern or transient irregular frequency content including anomalies with outlier characteristics (step 1) for the coordinates around the transverse axis of the trunk’s angular acceleration αy and velocity ωy. Considering step 2, d shows a schematic of irregular frequency content evaluated by the autocorrelation coefficient peaks rk,αy of a balance recovery response, where e shows an expected (regular) frequency content with recognisable periodic patterns of unperturbed walking. In f, g, the trunk kinematics αy and ωy for the same balance recovery response are shown. The red lines indicate the identified observation window, the blue dotted lines indicate the time window (±0.15 s) centred around the local maxima peak with outlier characteristic. In g, the same time window is highlighted (blue dotted lines), i.e. the analysis of whether there also exists an outlier (step 3).CC BY · PubMed Central

Verification

Claims the paper makes

The claims and statistics of this paper haven’t been checked yet.

Free: it reads the open-access text and checks the first 5 claims that cite a source we can check. Create a free account to check every claim.

A few free checks a day · papers already verified open instantly · results are public

Self-check

Percentages, sums and reported differences re-computed from the paper’s own numbers.

The self-check needs the paper’s full text.

Citations

Cites

Show all 30 references

Cited by

See all 9

Metadata

Authors
Gießler, M., Werth, J., Waltersberger, B., & Karamanidis, K.
Affiliations
  • London South Bank University (GB)
  • Offenburg University of Applied Sciences (DE)
  • University of Koblenz and Landau (DE)
Corresponding author
Gießler, Maximilian
Venue
Communications Engineering
Published
2024-12-16
Volume
3
Issue
1
Pages
181-181
ISSN
2731-3395 (ISSN-L)
OpenAlex
W4405421621
Cited by
6
References
30
Citations by year
2026: 6
FWCI
5.24
Type
journal article
Language
en
Open access
Licence
Locations
  • Communications Engineering (publishedVersion, cc-by)
  • PubMed Central (publishedVersion, cc-by)
  • PubMed (publishedVersion)
  • Opus-HSO (Offenburg University of Applied Sciences) (submittedVersion, cc-by)
  • DOAJ (DOAJ: Directory of Open Access Journals) (submittedVersion)
Topics
  • Health Sciences › Health Professions › Physical Therapy, Sports Therapy and Rehabilitation › Balance, Gait, and Falls Prevention
  • Physical Sciences › Computer Science › Computer Vision and Pattern Recognition › Context-Aware Activity Recognition Systems
  • Physical Sciences › Engineering › Biomedical Engineering › Gait Recognition and Analysis
Keywords
wearable inertial sensors, trunk kinematics, balance disorders, fall risk assessment, gait parameters, activities of daily living, remote monitoring

About these checks

Registry, retraction, Cites and statistics results are computed, not judged, with one exception: a printed reference that no index lists is read once by a model before it counts as a citation error. Claims are extracted by a model and kept only when their quote appears verbatim in the text; each cited source is then checked, with a model’s help, for whether it says what the claim relies on. Those verdicts can be wrong — read the claim’s quote and the cited source before relying on one.

Statistics and arithmetic are recomputed, not judged.

Confidence bands are shown only for fields calibrated on our annotated test set. None is calibrated yet, so no band is shown.

What counts as a claim ›How we measure confidence ›How we check this ›

Are you an author, or do you see a problem with this page? Request a review

How to cite

Cite CoChat for the checks on this page, and the paper for its own findings.

This page
CoChat. (2026). Citation check: A framework to automatically detect near-falls using a wearable inertial measurement cluster. https://app.cochat.ai/papers/10.1038/s44172-024-00325-x
The paper
Gießler, M., Werth, J., Waltersberger, B., & Karamanidis, K. (2024). A framework to automatically detect near-falls using a wearable inertial measurement cluster. Communications Engineering. https://doi.org/10.1038/s44172-024-00325-x

Data sources

Bibliographic data: Semantic Scholar, OpenAlex, Crossref.

Contains information from the Semantic Scholar Academic Graph Datasets, made available under the ODC Attribution License (ODC-BY 1.0).

Semantic Scholar’s data is described in: Kinney, R., et al. (2023). The Semantic Scholar Open Data Platform. arXiv:2301.10140.