How can AI reduce fall injuries in the workplace?

Antonios E. Gkikakis; Darwin G. Caldwell; Fabio Pera; Jes\'us Ortiz; Maria Teresa Settino; Nicholas Cartocci; Roberto F. Pitzalis

REVIEW 3 major objections 6 minor 31 references

How can AI reduce fall injuries in the workplace?

T0 review · 3 major / 6 minor · reviewed 2026-08-07 · deepseek-v4-flash

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗

Pith's one-line read A two-stage neural pipeline detects workplace falls and estimates impact time within about 160 ms.

desk verdict The fall-detection evaluation is the weak link: a random sequence split with demographics as inputs means the reported accuracy may be identity memorization rather than generalization to unseen users; the KAN time-to-impact part is more plausible but still needs a cleaner evaluation. read the letter →

arxiv 2505.24507 v1 pith:D2IRLIZB submitted 2025-05-30 eess.SP

Nicholas Cartocci , Antonios E. Gkikakis , Roberto F. Pitzalis , Fabio Pera , Maria Teresa Settino , Darwin G. Caldwell , Jesús Ortiz This is my paper

classification eess.SP

keywords falldetectionworkplacesafetyLSTMKolmogorov-Arnoldnetworktime-of-impactestimationwearablesensorsSisdatasetIMU

verification ladder T0 review T1 audit T2 compute T3 formal

The pith

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The reading

This paper is trying to establish that a wearable AI system can both recognize a fall as it begins and estimate when the person will hit the ground, before impact. If true, office and small-business workers could wear a device that gives a real-time warning or triggers protection, reducing the long absences falls currently cause. The proposed evidence is a two-stage pipeline tested on the public SisFall dataset: an LSTM-based recurrent network detects fall sequences with 82.6% average sensitivity and 98.4% specificity, while a Kolmogorov-Arnold network estimates time of impact with roughly 160 ms root-mean-square error. The work is aimed at an agile, working-age population, where fast reactions make false alarms a major usability issue.

What carries the argument

The mechanism that carries the argument is a two-stage wearable sensing pipeline. The first stage is the Fall Detection Neural Network, an LSTM-based recurrent network that consumes 18 inputs -- raw accelerations, gyroscope rates, quaternion orientation, body angle $\theta$ relative to gravity, and subject demographics -- and emits the probability of falling. The second is a Kolmogorov-Arnold Network (KAN), a function representation of the time-to-impact map as sums of univariate piecewise linear basis functions fit with the Newton-Kaczmarz method, using five features chosen by correlation and minimum-redundancy maximum-relevance selection. The SisFall recordings and the Musci et al. fall labels supply the shared ground truth that connects detection accuracy and impact-time error.

What would settle it

A subject-independent rerun of the same experiment -- train on some subjects' falls and ADLs, test on complete held-out subjects -- would settle the generalization question: if TPR, TNR, or RMSE degrades sharply, the model was memorizing subjects. Complement it with a fresh deployment recording falls from workers not in SisFall and checking whether the 160 ms impact-time estimate still appears before ground contact.

Watch

Extended reading notes

Core claim

The central claim is that a fall can be recognized and timed from waist-worn accelerometer and gyroscope data in real time using a two-stage neural architecture. The first stage, the Fall Detection Neural Network, consumes 18 channels -- raw accelerations from two accelerometers, gyroscope rates, quaternion orientation, body angle relative to gravity, plus the subject's age, height, weight, and gender -- and outputs a falling probability from stacked LSTM layers. The second stage is a Kolmogorov-Arnold Network that estimates remaining time to impact from five selected features using piecewise linear basis functions fit by the Newton-Kaczmarz method. On SisFall the detector reports an average true positive rate of 82.6%, a true negative rate of 98.4% on fall sequences, 94.4% on activities of daily living, and the impact-time estimator reaches roughly 160 ms RMSE. The paper also notes that the time-of-impact estimate saturates at about 140 ms very close to the ground, which defines how much warning the system can actually give.

Load-bearing premise

The load-bearing assumption is that the reported accuracy reflects learning what a fall looks like, not recognizing the specific people in the dataset, because training and test sequences come from the same subjects and the model is given each subject's age, height, weight, and gender.

Editorial extensions

If this is right

The detector can run on the 200 Hz waist-worn sensor data of SisFall, so a plug-and-play wearable device is a plausible deployment target.
Because the KAN needs only five features, three already in the dataset and two derived from accelerometer and gyroscope, the impact-time stage can work with a simpler sensor configuration than the full 18-channel detector.
The time-of-impact estimate saturates at about 140 ms close to the ground, so any protective actuation has a small, finite warning window to exploit.
Some activities and fall types are systematically harder: fall F10 (forward while trying to sit down) lowers detection, and high-intensity ADLs such as D04 cause the false-positive rate to rise, indicating where the approach needs more data.
The age, height, weight, and gender inputs let the model condition on body type, which matters for a workforce that is more agile than the elderly populations most fall studies target.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

A subject-independent test -- holding out whole people rather than random sequences -- is the natural next check; because same-subject sequences appear in both training and test and demographics are inputs, the reported numbers may partly reflect person recognition rather than fall dynamics.
The observed ~140 ms floor in the impact-time estimate sets a physical constraint: an airbag, brace, or posture assist must deploy within about a tenth of a second of the alarm to change the outcome.
A joint policy that uses the detector's confidence and the KAN's lead time together could tune false alarms against warning time; the paper explicitly leaves this combination for future work.

Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, and a circularity audit.

Desk Editor's Note

read the letter

The short version: this is a plausible system paper – an LSTM fall detector plus a KAN that estimates time to impact, tested on SisFall – but the headline fall-detection accuracy (TPR 82.6%, TNR 98.4%) is not evidence of generalization to new users because of how the train/test split was done. The concern is right: the 1798 fall sequences come from 24 subjects, each with five repetitions, and the 60/20/20 split is random per sequence. With age, height, weight, and gender as inputs, the model can essentially read the subject's identity off the demographics and the per-sensor signature. So the reported numbers likely measure interpolation to new sequences of known people, not performance on an unseen worker. The stated aim is a plug-and-play device for diverse workers, so this matters.

What the paper does well: it is clearly written, uses the public SisFall dataset, reports per-subject and per-activity results, and it is honest about two known confounds – the subjectivity of the Musci annotation and the fact that the 160 ms RMSE is distorted by a ~140 ms floor near impact. The KAN time-to-impact piece is a reasonable continuation of the authors' Fall-KAN work, and the integration with an LSTM detector is a legitimate engineering step.

The other soft spots are secondary but real. Feature selection for the KAN (correlation/mRMR) appears to be run on the whole dataset before the train/validate/test split, which is a mild leakage path even if the selected features look generic. There are no error bars or confidence intervals, no baseline comparison in the results (despite a mention in the methods), and no measurement of actual inference latency on the wearable target. None of these are fatal by themselves; combined, they push the paper into 'promising but unproven' territory.

Who is it for: people working on pre-impact fall detection and wearable edge systems. The paper deserves a serious referee, but it should come back with a hard requirement: rerun the FDNN evaluation with a subject-independent split, report per-subject performance separately, and add error bars and a baseline. If the numbers survive that, the paper is meaningful. As it stands, I would not cite the fall-detection accuracy as evidence of generalization.

My recommendation: send to peer review, but expect major revision. It is not desk-reject material; the integration is sensible and the flaw is fixable.

Referee Report

3 major / 6 minor

Summary. The manuscript proposes a two-stage wearable system for mitigating workplace falls: a recurrent neural network (FDNN) with LSTM layers that detects an ongoing fall, and a Kolmogorov-Arnold Network (KAN) that estimates the time to impact. The FDNN is trained on 1798 fall sequences from the SisFall dataset and the KAN is trained on five selected inertial features. The authors report an average true positive rate of 82.6% and true negative rate of 98.4% on fall sequences, a true negative rate of 94.4% on activities of daily living, and a root mean square error of approximately 160 ms for time-to-impact estimation. The paper also presents per-subject and per-activity breakdowns and a qualitative example of the KAN output.

Significance. If the reported metrics held under a rigorous, subject-independent evaluation protocol, the proposed combination of an LSTM-based fall detector and a KAN time-to-impact estimator could be a worthwhile contribution to real-time fall prevention for industrial workers. The use of the public SisFall dataset, the per-activity and per-subject result tables, and the explicit treatment of time-to-impact estimation are strengths. However, the current evaluation protocol does not support the central claim of generalization to new users, and the feature selection procedure for the KAN appears to leak information from the test fold. The paper is a brief report rather than a full methods paper; no code or detailed methodology is provided, so independent reproduction would be difficult.

major comments (3)

[§3] In §3, after shuffling the 1798 fall sequences from 24 subjects, the dataset is split 60/20/20 into train/validation/test without subject partitioning, while the model inputs include age, height, weight, and gender (Table 3). Since each subject performed five repetitions of each activity, the test set almost certainly contains sequences from subjects also present in the training set. The reported average TPR (82.6%) and TNR (98.4%) are therefore consistent with the model recognizing the subject—through demographics or sensor-specific characteristics—rather than detecting fall dynamics on unseen users, which is the stated plug-and-play goal. Please re-evaluate with a subject-independent split (e.g., leave-one-subject-out) and, ideally, with an ablation that excludes the demographic features.
[§4.1–§4.2] The correlation and mRMR feature selection in §4.1 is described before the train/validation/test split in §4.2, suggesting that it is applied to the whole dataset. The test repetition (the fifth repetition for each subject) is then used to compute the RMSE that supports the central claim of approximately 160 ms. If feature selection used the test repetition to choose features, the reported RMSE is optimistically biased. Please move feature selection inside the cross-validation loop, or explicitly state and demonstrate that the test fold did not influence feature choice.
[§7 and Conclusions] The paper reports an average RMSE of about 160 ms for time-to-impact estimation, but the Conclusions state that for very small time-of-impact values the estimate is distorted by a constant error of about 140 ms in the presented example. This implies a systematic and large error in the most safety-critical period immediately before impact. Please report the RMSE and bias as a function of time-to-impact, especially for the last 200 ms before impact, and discuss how the system would trigger a mitigation mechanism given this error structure.

minor comments (6)

[Abstract vs. §6] The abstract cites 2706 ADL sequences in SisFall, while §6 states that the FDNN was tested on 2701 ADL sequences; please reconcile this discrepancy.
[Table 3 and §3] Table 3 lists 19 features (four demographic, nine inertial, four quaternion, and two derived), but the text in §3 says the input has 18 entries; please correct the count or the table.
[§3, Figure 1] The architecture description references Layers 1–9 but the paper does not include the actual figure; a textual description of the tensor shapes and connections would help the reader follow the design.
[§5–§6] No comparison with existing fall detection or time-to-impact methods is provided; a quantitative comparison on a common protocol would help position the contribution relative to prior work.
[§3 and §4.2] Details of the training procedure are incomplete: the optimizer, learning rate, and early-stopping criterion for the FDNN are not specified, and the k-fold scheme for the KAN parameter search (number of folds, fold construction) is not described.
[§4] The fall-interval definition relies on a threshold on the standard deviation of acceleration; please provide a sensitivity analysis or justify the threshold choice, since the time-of-impact labels depend on it.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: the fall-detection and time-of-impact results are empirical fits to an external dataset with held-out splits; self-citations are not load-bearing.

full rationale

The derivation chain is data-driven rather than definitionally self-referential. The fall-detection RNN is trained, validated, and tested on separate portions of the 1798 fall sequences, and the reported TPR/TNR values are empirical outcomes, not identities imposed by input features. The time-of-impact KAN is trained on three repetitions, validated on one, and tested on the last repetition per subject; the ~160 ms RMSE is a measured error against labels defined from Musci et al. annotations and the sampling period, and it is not equal to any network input by construction. The demographic inputs (age, height, weight, gender) combined with a random rather than subject-independent split raise a legitimate generalization concern about possible subject-identity leakage, and feature selection performed on all data before the KAN split is a possible leakage path, but these are evaluation-validity issues, not circularity reductions: no fitted parameter is being renamed as a prediction and no equation reduces to its own input. Self-citations such as [14] point to the authors' prior Fall-KAN work, but this paper independently specifies the basis functions, Newton-Kaczmarz solver, fine-tuning procedure, and feature selection; no load-bearing argument rests solely on an unverified self-citation. Therefore the evidentiary bar for flagging circularity is not met.

Assumptions & free parameters 8 free parameters · 3 assumptions · 0 invented entities

The central evaluation rests on the public SisFall dataset, its subjective annotations, and the assumption that lab-simulated falls resemble workplace falls. The KAN and FDNN hyperparameters are tuned on data; the KAN introduces no new physical or conceptual entities beyond the learned basis functions of the prior Fall-KAN work.

free parameters (8)

KAN internal piecewise linear basis functions n = 4
Selected by k-fold cross-validation in Section 4.2; controls the internal complexity of the KAN representation.
KAN external piecewise linear basis functions q = 64
Selected by k-fold cross-validation in Section 4.2; controls the number of external basis functions.
Newton-Kaczmarz regularization parameter mu = 0.0625
Selected by k-fold cross-validation in Section 4.2; learning rate/regularization for the KAN solver.
Time window w = 50 ms
Selected by k-fold cross-validation in Section 4.2; the window of data used to predict time of impact.
FDNN LSTM inner dimension = 24
Chosen by hand in Section 3 ('chosen to be 24'); architecture hyperparameter.
FDNN dropout rate = 50%
Chosen in Section 3; dropout applied to LSTM layers during training.
FDNN batch size = 128
Set in Section 3 training procedure.
FDNN number of epochs = 64
Set in Section 3 training procedure.

assumptions (3)

domain assumption The SisFall dataset and the manual annotations by Musci et al. [19] correctly identify fall periods and impact times.
Section 2 notes that SisFall has no video references and labeling is subjective; the time-of-impact ground truth depends on these annotations.
domain assumption The six-axis Kalman filter provides accurate body orientation from the accelerometers and gyroscope.
Section 3 uses the Kalman filter output (quaternion, theta, theta-dot) as input features; any error propagates into the models.
domain assumption The simulation falls performed by volunteers in a lab approximate real workplace falls.
Section 2 describes SisFall as simulated falls by volunteers; the paper's applicability to real workplaces rests on this similarity.

how reviews work

0 comments

Cite this review

Pith. "Pith review of How can AI reduce fall injuries in the workplace?." pith.science (2026). https://pith.science/paper/D2IRLIZB

@misc{pith2026250524507,
  author       = {Pith},
  title        = {Pith review of: How can AI reduce fall injuries in the workplace?},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/D2IRLIZB}},
  note         = {Machine review of arXiv:2505.24507}
}

read the original abstract

Fall-caused injuries are common in all types of work environments, including offices. They are the main cause of absences longer than three days, especially for small and medium-sized businesses (SMEs). However, data, data amount, data heterogeneity, and stringent processing time constraints continue to pose challenges to real-time fall detection. This work proposes a new approach based on a recurrent neural network (RNN) for Fall Detection and a Kolmogorov-Arnold Network (KAN) to estimate the time of impact of the fall. The approach is tested on SisFall, a dataset consisting of 2706 Activities of Daily Living (ADLs) and 1798 falls recorded by three sensors. The results show that the proposed approach achieves an average TPR of 82.6% and TNR of 98.4% for fall sequences and 94.4% in ADL. Besides, the Root Mean Squared Error of the estimated time of impact is approximately 160ms.

Discussion (0). Sign in to comment.

Reference graph

Works this paper leans on

31 extracted references · 28 canonical work pages

[1]

Deaths from falls have increased much faster than any other type of injury in the past two decades

Introduction Every year, 684,000 people lose their lives, as reported by the World Health Organization (WHO). Deaths from falls have increased much faster than any other type of injury in the past two decades. In addition to the number of deaths, 172 million people are disabled by falls each year [1]. This increase is due to many factors, such as aging an...

work page
[2]

It contains 15 types of falls and 19 types of Activities of Daily Living (ADLs), listed in Tables 1 and 2

SisFall The publicly available SisFall [18] dataset is used to train the proposed approach because it is one of the most comprehensive datasets with many subjects and falls. It contains 15 types of falls and 19 types of Activities of Daily Living (ADLs), listed in Tables 1 and 2. Code Activity F01 Fall forward while walking caused by a slip F02 Fall backw...

work page
[3]

Fall Detection Data Processing A deep recurrent neural network (RNN) was chosen for data-based fall detection, where the data used to train the neural network are obtained from the SisFall dataset [18] labeled in

work page
[4]

Time of impact Data Processing Overall, 19 signals are considered at the beginning of this study. From each 15 -second-long experiment, only the part related to the fall was extracted, thus excluding the initial activity, e.g., walking, jogging, sitting, etc., and the final parts where t he subject remains motionless after hitting the ground. This interva...

work page
[5]

The results below relate to the data in SisFall and analyze the performance of the approach both with respect to different ac tivities and for different subjects

Results This section describes the results obtained considering our proposed RNN for Fall Detection and KAN for Time of Impact Estimation. The results below relate to the data in SisFall and analyze the performance of the approach both with respect to different ac tivities and for different subjects

work page
[6]

Figures 2 and 3 show the TNR and TPR as a function of the subjects and the activities, respectively

Fall Detection Once the FDNN is trained, it is tested on the Test dataset (~360 sequences). Figures 2 and 3 show the TNR and TPR as a function of the subjects and the activities, respectively. The tables show the average value between the repetitions, and the blank cells mean that there are no samples of those subjects and activities in the Test dataset. ...

work page
[7]

The performance obtained is generally positive, on average the RMSE on the 1798 falls is about 160 𝑚𝑠

Time of Impact Estimation Figure 5 shows the RMSE for different subjects and falls, considering the best parameters on the test data. The performance obtained is generally positive, on average the RMSE on the 1798 falls is about 160 𝑚𝑠. On some subjects, the algorithm obtains worse results, such as for SA10; likewise, some types of falls, such as F13. Par...

work page
[8]

Conclusions The Fall Detection Neural Network performed excellently on SisFall with an average TPR of 82.6% and TNR of 98.4% in fall sequences and 94.4% in ADL sequences. While with the KAN, only five features were found to be useful for time of impact estimation, of which 3 were already in the dataset, and two were calculated using an accelerometer and a...

work page

Show all 31 references

[9]

Latest research trends in fall detection and prevention using machine learning: A systematic review,

S. Usmani, A. Saboor, M. Haris, M. A. Khan, and H. Park, “Latest research trends in fall detection and prevention using machine learning: A systematic review,” Sensors, vol. 21, no. 15, p. 5134, Jul. 2021, doi: 10.3390/s21155134

2021 doi
[10]

J. D. Stanaway et al., “Global, regional, and national comparative risk assessment of 84 behavioural, environmental and occupational, and metabolic risks or clusters of risks for 195 countries and territories, 1990–2017: a systematic analysis for the Global Burden of Disease S...

1990 doi
[11]

Medical Costs of Fatal and Nonfatal Falls in Older Adults,

C. S. Florence, G. Bergen, A. Atherly, E. Burns, J. Stevens, and C. Drake, “Medical Costs of Fatal and Nonfatal Falls in Older Adults,” J Am Geriatr Soc, vol. 66, no. 4, pp. 693–698, Apr. 2018, doi: 10.1111/jgs.15304

2018 doi
[12]

Step safely: strategies for preventing and managing falls across the life-course. Geneva: World Health Organization,

“Step safely: strategies for preventing and managing falls across the life-course. Geneva: World Health Organization,” 2021

2021
[13]

La valutazione ergonomica del rischio caduta in piano

A. CURA et al., “La valutazione ergonomica del rischio caduta in piano”
[14]

Recognition of Physiological Patterns during Activities of Daily Living Using Wearable Biosignal Sensors,

N. Cartocci et al., “Recognition of Physiological Patterns during Activities of Daily Living Using Wearable Biosignal Sensors,” in 22nd Congress of the International Ergonomics Association (IEA 2024), Jeju: Springer, Aug. 2024

2024
[15]

Edge-AI in LoRa-based health monitoring: Fall detection system with fog computing and LSTM recurrent neural networks,

J. Pena Queralta, T. N. Gia, H. Tenhunen, and T. Westerlund, “Edge-AI in LoRa-based health monitoring: Fall detection system with fog computing and LSTM recurrent neural networks,” 2019 42nd International Conference on Telecommunications and Signal Processing, TSP 2019, pp. 60...

2019
[16]

A Machine Learning Approach for Fall Detection and Daily Living Activity Recognition,

A. Chelli and M. Patzold, “A Machine Learning Approach for Fall Detection and Daily Living Activity Recognition,” IEEE Access, vol. 7, pp. 38670–38687, 2019, doi: 10.1109/ACCESS.2019.2906693

2019
[17]

A Survey on Recent Advances in Wearable Fall Detection Systems,

A. Ramachandran and A. Karuppiah, “A Survey on Recent Advances in Wearable Fall Detection Systems,” Biomed Res Int, vol. 2020, 2020, doi: 10.1155/2020/2167160

2020 doi
[18]

SisFall: A fall and movement dataset,

A. Sucerquia, J. D. López, and J. F. Vargas-Bonilla, “SisFall: A fall and movement dataset,” Sensors (Switzerland), vol. 17, no. 1, p. 198, Jan. 2017, doi: 10.3390/s17010198

2017 doi
[19]

Using the two accelerometers and one gyroscope, the orientation of the subject was calculated through a six-axis Kalman filter [20] and expressed as a quaternion

as no-background. Using the two accelerometers and one gyroscope, the orientation of the subject was calculated through a six-axis Kalman filter [20] and expressed as a quaternion. Then, the angle 𝜃 and acceleration 𝜃̇ of the body relative to the ground were estimated. A full ...
[20]

Elderly fall detection with an accelerometer using lightweight neural networks,

G. Wang, Q. Li, L. Wang, Y. Zhang, and Z. Liu, “Elderly fall detection with an accelerometer using lightweight neural networks,” Electronics (Switzerland), vol. 8, no. 11, p. 1354, Nov. 2019, doi: 10.3390/electronics8111354

2019 doi
[21]

Elderly Fall Detection Using Wearable Sensors: A Low Cost Highly Accurate Algorithm,

M. Saleh and R. L. B. Jeannes, “Elderly Fall Detection Using Wearable Sensors: A Low Cost Highly Accurate Algorithm,” IEEE Sens J, vol. 19, no. 8, pp. 3156–3164, Apr. 2019, doi: 10.1109/JSEN.2019.2891128

2019
[22]

Fall Risk Reduction for the Elderly by Using Mobile Robots Based on Deep Reinforcement Learning,

T. Namba and Y. Yamada, “Fall Risk Reduction for the Elderly by Using Mobile Robots Based on Deep Reinforcement Learning,” Journal of Robotics, Networking and Artificial Life, vol. 4, no. 4, p. 265, 2018, doi: 10.2991/jrnal.2018.4.4.2

2018 doi
[23]

Falls in the elderly: Detection and assessment,

C. J. Lord and D. P. Colvin, “Falls in the elderly: Detection and assessment,” Proceedings of the Annual Conference on Engineering in Medicine and Biology, vol. 13, no. pt 4, pp. 1938–1939, 1991, doi: 10.1109/iembs.1991.684830

1938
[24]

Fall-KAN: Fall impact time estimation Kolmogorov-Arnold Network,

N. Cartocci, A. E. Gkikakis, F. Pera, M. T. Settino, D. G. Caldwell, and J. Ortiz, “Fall-KAN: Fall impact time estimation Kolmogorov-Arnold Network,” in 2024 4th International Conference on Electrical, Computer, Communications and Mechatronics Engineering (ICECCME), IEEE, Nov....

2024
[25]

Deep Learning-based wearable device to prevent fall from height injuries,

N. Cartocci, A. E. Gkikakis, D. G. Caldwell, and J. Ortiz, “Deep Learning-based wearable device to prevent fall from height injuries,” Zenodo, Mar. 2024. doi: 10.5281/zenodo.10722452

2024 doi
[26]

Artificial intelligence-based wearable solution to prevent fall from heights injuries for the next generation of workers,

N. Cartocci, A. Gkikakis, D. Caldwell, and J. Ortiz, “Artificial intelligence-based wearable solution to prevent fall from heights injuries for the next generation of workers,” in 2023 Slips, Trips and Falls (STF) International Conference, Toronto, Jun. 2023

2023
[27]

Real-time Fall Prevention system for the Next- generation of Workers,

N. Cartocci, A. Gkikakis, D. Caldwell, and J. Ortiz, “Real-time Fall Prevention system for the Next- generation of Workers,” in Workshop on Assistive Robotic Systems for Human Balancing and Walking: Emerging Trends and Perspectives @IROS 2022, Kyoto, Oct. 2022

2022
[28]

Online Fall Detection Using Recurrent Neural Networks on Smart Wearable Devices,

M. Musci, D. De Martini, N. Blago, T. Facchinetti, and M. Piastra, “Online Fall Detection Using Recurrent Neural Networks on Smart Wearable Devices,” IEEE Trans Emerg Top Comput, vol. 9, no. 3, pp. 1276–1289, Jul. 2021, doi: 10.1109/TETC.2020.3027454

2021
[29]

Compensation of magnetic disturbances improves inertial and magnetic sensing of human body segment orientation,

D. Roetenberg, H. J. Luinge, C. T. M. Baten, and P. H. Veltink, “Compensation of magnetic disturbances improves inertial and magnetic sensing of human body segment orientation,” IEEE Transactions on Neural Systems and Rehabilitation Engineering, vol. 13, no. 3, pp. 395–405, Se...

2005
[30]

Feature selection based on mutual information criteria of max-dependency, max-relevance, and min-redundancy,

Hanchuan Peng, Fuhui Long, and C. Ding, “Feature selection based on mutual information criteria of max-dependency, max-relevance, and min-redundancy,” IEEE Trans Pattern Anal Mach Intell, vol. 27, no. 8, pp. 1226–1238, Aug. 2005, doi: 10.1109/TPAMI.2005.159

2005 doi
[31]

Construction of the Kolmogorov-Arnold representation using the Newton-Kaczmarz method,

M. Poluektov and A. Polar, “Construction of the Kolmogorov-Arnold representation using the Newton-Kaczmarz method,” May 2023, [Online]. Available: http://arxiv.org/abs/2305.08194

2023 arXiv

Pith tools

Reviewed August 7, 2026 · model on record in the stance chip above.

[1] [1]

Deaths from falls have increased much faster than any other type of injury in the past two decades

Introduction Every year, 684,000 people lose their lives, as reported by the World Health Organization (WHO). Deaths from falls have increased much faster than any other type of injury in the past two decades. In addition to the number of deaths, 172 million people are disabled by falls each year [1]. This increase is due to many factors, such as aging an...

work page

[2] [2]

It contains 15 types of falls and 19 types of Activities of Daily Living (ADLs), listed in Tables 1 and 2

SisFall The publicly available SisFall [18] dataset is used to train the proposed approach because it is one of the most comprehensive datasets with many subjects and falls. It contains 15 types of falls and 19 types of Activities of Daily Living (ADLs), listed in Tables 1 and 2. Code Activity F01 Fall forward while walking caused by a slip F02 Fall backw...

work page

[3] [3]

Fall Detection Data Processing A deep recurrent neural network (RNN) was chosen for data-based fall detection, where the data used to train the neural network are obtained from the SisFall dataset [18] labeled in

work page

[4] [4]

Time of impact Data Processing Overall, 19 signals are considered at the beginning of this study. From each 15 -second-long experiment, only the part related to the fall was extracted, thus excluding the initial activity, e.g., walking, jogging, sitting, etc., and the final parts where t he subject remains motionless after hitting the ground. This interva...

work page

[5] [5]

The results below relate to the data in SisFall and analyze the performance of the approach both with respect to different ac tivities and for different subjects

Results This section describes the results obtained considering our proposed RNN for Fall Detection and KAN for Time of Impact Estimation. The results below relate to the data in SisFall and analyze the performance of the approach both with respect to different ac tivities and for different subjects

work page

[6] [6]

Figures 2 and 3 show the TNR and TPR as a function of the subjects and the activities, respectively

Fall Detection Once the FDNN is trained, it is tested on the Test dataset (~360 sequences). Figures 2 and 3 show the TNR and TPR as a function of the subjects and the activities, respectively. The tables show the average value between the repetitions, and the blank cells mean that there are no samples of those subjects and activities in the Test dataset. ...

work page

[7] [7]

The performance obtained is generally positive, on average the RMSE on the 1798 falls is about 160 𝑚𝑠

Time of Impact Estimation Figure 5 shows the RMSE for different subjects and falls, considering the best parameters on the test data. The performance obtained is generally positive, on average the RMSE on the 1798 falls is about 160 𝑚𝑠. On some subjects, the algorithm obtains worse results, such as for SA10; likewise, some types of falls, such as F13. Par...

work page

[8] [8]

Conclusions The Fall Detection Neural Network performed excellently on SisFall with an average TPR of 82.6% and TNR of 98.4% in fall sequences and 94.4% in ADL sequences. While with the KAN, only five features were found to be useful for time of impact estimation, of which 3 were already in the dataset, and two were calculated using an accelerometer and a...

work page