EP-4740853-A1 - METHOD, SYSTEM, AND DEVICES FOR DETECTING AND QUANTIFYING BODY MOTION

Abstract

The present disclosure relates to a method for motion detection and quantification. The method comprises the steps of obtaining a time-domain data signal based on raw motion data, determining a time-frequency representation, TFR, of the data signal, determining a ridge in the TFR, wherein the ridge corresponds to locations of maximum energy or maximum amplitude in the TFR and comprises a motion-related frequency, fa , per each time step of the data signal, splitting the time-domain data signal into one or more stationary time step segments based on the determined ridge, reconstructing motion data based on the one or more stationary time step segments of the data signal, and determining and quantifying the motion data based on temporal characteristics of the reconstructed motion data.

Inventors

• ENA BERNAD, Alejandro
• MAZZÀ, Claudia
• BELACHEW, Shibeshih Mitiku
• REYES PUPO, Óscar Gabriel

Assignees

• Indivi AG

Dates

Publication Date

20260513

Application Date

20241108

Claims (15)

1. Method for motion detection and quantification, the method comprising: a) obtaining a time-domain data signal based on raw motion data; b) determining a time-frequency representation, TFR, of the data signal; c) determining a ridge in the TFR, wherein the ridge corresponds to locations of maximum energy or maximum amplitude in the TFR and comprises a motion-related frequency, fa, per each time step of the time-domain data signal; d) splitting the time-domain data signal into one or more stationary time step segments based on the determined ridge; e) reconstructing motion data based on the one or more stationary time step segments of the data signal; and f) determining and quantifying the motion data based on temporal characteristics of the reconstructed motion data.
2. The method according to claim 1, wherein, in a), the raw motion data is obtained from one or more inertial measurement unit, IMU, sensors which preferably comprise one or more of: an accelerometer; and/or a gyroscope.
3. The method according to any of claims 1 to 2, wherein, in c), determining the ridge comprises: splitting the time-domain data signal and the TFR into a plurality of overlapping time step windows, and obtaining a dominant frequency, ff, for each time step window.
4. The method according to claim 3, wherein obtaining ff for each time step window comprises: determining a main frequency, f0 , using autocorrelation on the respective data signal time step window, and correcting f0 by determining the closest dominant frequency in the frequency spectrum of the corresponding TFR time step window.
5. The method according to claim 4, wherein in c), fa is obtained based on the obtained ff values of the time step windows.
6. The method according to claim 5, wherein in c), for each respective time step, fa is obtained by averaging the ff values of all the time step windows that contain the respective time step.
7. The method according to any one of claims 3 to 6, wherein each time step window overlaps adjacent time step windows by more than 70%, preferably by more than 80%, more preferably by more than 90%, even more preferably by more than 95% of the window time step size.
8. The method according to any one of claims 1 to 7, wherein in d), splitting the time-domain data signal into stationary time step segments comprises: determining stationary time step segments based on one or more jumps in the ridge that correspond to changes of the ridge that exceed a predetermined frequency threshold, and splitting the motion data into stationary time step segments separated by respective time steps where the one or more jumps in the ridge are determined.
9. The method according to claim 8, wherein the predetermined frequency threshold is in a range from 0.05 Hz to 0.2 Hz, preferably from 0.07 to 0.13 Hz, most preferably about 0.1 Hz.
10. The method according to any one of claims 1 to 9, wherein in e), reconstructing the motion data comprises, for each stationary time step segment: determining a mean main frequency, fs , of a respective stationary time step segment by averaging fa values of the respective segment of the ridge that corresponds to the time steps of the respective stationary time step segment; determining a frequency spectrum of the respective segment of the TFR; determining upper and lower bounds for the band frequency around fs ; and filtering each respective stationary time step segment using a bandpass filter.
11. The method according to any one of claims 9 to 10, wherein in e), reconstructing the motion data further comprises reconstructing a final filtered signal based on the respective filtered stationary time step segments.
12. The method according to the preceding claim, wherein in e), determining and quantifying the motion data comprises determining motion events based on one or more of: local extrema in the reconstructed motion data, zero-crossing in the reconstructed motion data, an amplitude of the reconstructed motion data, and/or a distance between local extrema based on frequencies comprised by the ridge.
13. An apparatus comprising at least one processor configured to perform a method according to any one of the preceding claims.
14. Computer program product comprising instructions that, when executed by an apparatus comprising at least one processor, cause the apparatus to perform a method according to any one of claims 1 to 12.
15. Computer-readable storage medium comprising the computer program product according to the preceding claim.

Description

Field of disclosure The disclosure relates to methods, systems, and devices for detecting and quantifying motion of body parts, body segments, and/or the whole body. The disclosure relates, in particular, to methods, systems, and devices for human motion detection and quantification using a frequency-adaptive segmentation of motion data. Technical background Human motion quantification is highly relevant in the assessment and monitoring of e.g. neurological or musculo-skeletal diseases that impair an individual's movement. Here, obtaining objective and measurable data about an individual's motor abilities is of paramount importance in different domains, such as gait, balance, coordination and dexterity. As an example of human motion, gait is typically described as a cyclic movement, with the duration of individual gait cycles typically quantified by detecting (i) the interval between two subsequent initial contacts of the same foot with the ground (stride) and (ii) the interval between initial contact of one foot and next initial contact of the contralateral foot (step). Changes in these essential parameters of gait are typically first indicators of impaired motor abilities. Other examples of human motion are sit-stand transitions where an individual repeats cycles of standing and sitting, and alternate pronation-supination movements such as, e.g., rotating the forearm back and forth between a palm down orientation and a palm up orientation. Gait impairments are commonly quantified through observational gait analysis and quantitative gait analysis. Observational gait analysis techniques are subject to bias and have limited accuracy [1]. On the other hand, quantitative gait analysis techniques seek to eliminate possible human interpretation bias [2, 3] by employing technologies such, e.g., as pressure-sensitive walkways and motion capture systems. However, these technologies are expensive and require a laboratory environment, which limits their use to very infrequent and supervised assessments of an individual's gait. The increased prevalence and technological advancement of modern mobile and wearable devices in recent years has led to new techniques and technologies that aim to leverage sensors typically provided in these modern devices such as, e.g., digital cameras, microphones, global positioning systems (GPS), and inertial measurement units (IMUs - accelerometers, gyroscopes, magnetometers). These new techniques and technologies aim to provide affordable, and accurate monitoring tools [4, 5] for both clinical use and home-based monitoring. In conjunction, algorithms for quantification of sensor-based measurements from the IMU data [6] have been proposed. However, when applied to IMU data from mobile devices, such as smartphones, and unsupervised data collection conditions, the accuracy of measurements is heavily affected by lower quality of the electronics, movement artifacts, and by the varying orientation and positioning of the devices relative to an individual's body. The varying severity of neurological impairments, causing irregular or jerky movements, further exacerbate these issues. Overall, even basic gait analysis remains highly challenging because of suboptimal accuracy of step detection and segmentation in relation to the aforementioned technical and physiological sources of variability. Simple methods such as thresholding [7] or zero-crossing detection [8] have been proposed that are easy to implement, but face challenges with individuals with abnormal gait patterns, and advanced methods utilizing deep learning [9] may offer higher accuracy but are difficult to implement as they require substantial computational resources and availability of large, labeled datasets. Overall, main drawbacks of the state-of-the-art can be summarized as: High dependence on device location: For example, algorithms designed for data located near the body's Center of Mass, e.g., L5 - 5th lumbar vertebrae, cannot be readily translated to a device worn in another body location like a clothing pocket such as, e.g., a pants pocket, as is commonly the case for an approach utilizing smartphones.Required assumption of a steady walking cadence: This leads to significant challenges when applied to individuals with neurological diseases who exhibit high variability in their gait patterns that becomes even more pronounced in daily-life environments.Limited accuracy in the presence of noise and artifact: Robust signal pre-processing algorithms are lacking. For example, filters at fixed frequencies are often used for the entire data signal, meaning that not all the background noise is removed. This can lead to an insufficient signal-to-noise ratio. There is therefore a need to address the above issues of the state-of-the art. Summary of the disclosure Some or all issues of the state-of-the-art are addressed by the disclosure as defined by the independent claims. Preferred embodiments are defined by the sub-features of the depe