DE102017103607A1 - Method for determining the stride length of a pedestrian - Google Patents

Abstract

The invention relates to a method for determining the stride length of a pedestrian based on the detection of a step, for. As a function of the size of the resulting during walking or running Beinschwenkwinkels his legs, wherein the method a) a step of the pedestrian, z. B) a mathematical model describing the context of a pedestrian's step, e.g. Provided or developed based on the size of the leg swing angle and the resulting stride length depending on the size and / or movement of the body when walking or running, c) the mathematical model and / or the main moment of inertia measurement unit by a pedestrian at times d) measurement data of the main inertia measurement unit are input to the mathematical model ande) the step length by means of the calibrated model or the calibrated main inertial measurement unit based on the measurement data provided by the latter Walking or running the pedestrian is appreciated.

Description

The invention relates to a method for determining the stride length of a pedestrian based on the detection of a step. In particular, the invention relates to a method for improved and more accurate estimation of the stride length of a pedestrian due to step detection by a pedestrian-carried "wearable device" with integrated inertial measurement unit.

Wearable Devices find many different applications, including pedestrian navigation. In this case, the inertial measurement unit (IMU) is embedded in the wearable device. In inertial navigation systems (INS) based on IMUs that are not attached to the foot, the approach of the step length and movement direction estimation is used. This approach requires calibrating the empirical model to estimate the stride length. The present invention proposes a method for automatically calibrating empirical models for step length estimation. Two IMUs are used. One of these IMUs is at a user's body part remote from the feet or ankles, e.g. the thigh, and is used for the purpose of inertial navigation. The second IMU is embedded in the shoe and used to calibrate the empirical model of the first IMU. Two main results are achieved. First, calibration is only required during the first two minutes. Secondly, the automatic calibration reduces the mean error distance error as well as the standard deviation.

Introduction: One of the factors that has triggered the popularity of inertial sensors is the miniaturization of these sensors. Inertia sensors are found not only in smartphones, but also in wearable devices such as smartphones. in smart watches. In fact, it can be predicted that these sensors will also be integrated into clothing in the future. This increases the motivation for the further development of applications based on inertial sensors. In the present invention, the focus is on inertial navigation.

In-motion navigation using non-foot-mounted IMUs uses the step-length and moving-direction estimation approach to iteratively track the position of the pedestrian. According to this approach, estimating the stride length of the user is done in two main process steps. The first step is the direction estimate, which is usually done using the turn rate measurements. The second step is the step length estimation, which is done using an empirical model. The latter requires a calibration that is time consuming and error prone if the user is unfamiliar with the procedure.

This disadvantage, i. the need for calibration is handled in the prior art primarily in three ways. First, the model parameters can be set to predetermined values [1]. Second, a manual calibration can be done, e.g. by walking over a predetermined distance [2]. Third, cards can be used [3]. Nevertheless, these disadvantages can be reduced thanks to the wearable devices. Since IMUs will be integrated into footwear and shoes in the future, inertial data from various parts of the body will be available without incurring additional costs. Thus, IMUs can be used to simplify or automate the calibration process.

• • Bei einigen der Kalibrierungs-Ansätze ist ein a-priori-Wissen über die Länge des Wegs erforderlich.
• • Bei den Kalibrierungs-Ansätzen wird die mittlere Schrittlänge über eine bestimmte Strecke verwendet, statt die tatsächliche Länge des Schritts zu verwenden.
• • Der Ansatz, bei dem eine Landkarte verwendet wird, erfordert, dass die Landkarte verfügbar ist. Diese ist jedoch nicht stets verfügbar oder könnte Fehler enthalten.

The disadvantages of the prior art in this connection are the following:

• • Some of the calibration approaches require a priori knowledge of the length of the path.
• • The calibration approaches use the average stride length over a certain distance instead of using the actual length of the step.
• • The approach using a map requires the map to be available. However, this is not always available or could contain errors.

The object of the invention is to improve the estimation of the stride length using a step length calculation model.

1. a) ein Schritt des Fußgängers, z. B. die Größe des Beinschwenkwinkels, mittels einer Haupt-Trägheitsmesseinheit ermittelt wird, die von dem Fußgänger an einer Stelle seines Körpers oberhalb der Fußgelenke getragen wird,
2. b) ein mathematisches Modell zur Beschreibung des Zusammenhangs eines Schritts eines Fußgängers, z. B. anhand der Größe des Beinschwenkwinkels und der sich ergebenden Schrittlänge in Abhängigkeit von der Größe und/oder der Bewegung des Körpers beim Gehen oder Laufen bereitgestellt oder entwickelt wird,
3. c) das mathematische Modell und/oder die Haupt-Trägheitsmesseinheit mittels einer zeitweise vom Fußgänger an zumindest einem seiner Füße oder Fußgelenke zu tragenden Hilfs-Trägheitsmesseinheit kalibriert wird,
4. d) Messdaten der Haupt-Trägheitsmesseinheit in das mathematische Modell eingegeben werden und
5. e) die Schrittlänge mittels des kalibrierten Modells bzw. der kalibrierten Haupt-Trägheitsmesseinheit anhand der von dieser gelieferten Messdaten beim Gehen oder Laufen des Fußgängers geschätzt wird.

To solve this problem, a method for determining the stride length of a pedestrian based on the detection of a step, for. B. depending on the size of the resulting when walking or running Beinschwenkwinkels his legs, said in the process

1. a) a step of the pedestrian, z. The size of the leg swing angle is determined by means of a main inertial measurement unit carried by the pedestrian at a location of his body above the ankles,
2. b) a mathematical model describing the context of a pedestrian's step, e.g. On the basis of the size of the leg swing angle and the resulting stride length as a function of the size and / or or the movement of the body is provided or developed while walking or running,
3. c) the mathematical model and / or the main inertial measurement unit is calibrated by means of an auxiliary inertial measurement unit to be temporarily worn by the pedestrian on at least one of his feet or ankles,
4. d) measurement data of the main inertial measurement unit are entered into the mathematical model and
5. e) the step length is estimated by means of the calibrated model or the calibrated main inertial measuring unit on the basis of the measured data supplied by the latter when the pedestrian walks or runs.

An advantageous development of the invention is specified in claim 2.

According to the invention, it is proposed to use an IMU embedded in the shoe to calibrate empirical models for step size estimation. To illustrate the concept, a model based on an IMU inserted into the pocket is calibrated. It could be e.g. to trade a smartphone.

• 1 die Verteilung der Schrittlänge über dem Beinschwenkwinkel für verschiedene Fußgänger, empirisch ermittelt,
• 2 die rekursive Schätzung des Offset b (Schnittpunkt der Ausgleichslinien der 1 mit der Y-Achse); jede Linie repräsentiert einen anderen Testgang,
• 3 eine schematische Darstellung der Schätzung der Schrittlänge auf Basis des Beinschwenkwinkels und
• 4 eine Kurve zur Fehlerbetrachtung.

An embodiment of the invention will be explained in more detail with reference to the drawing. In detail, they show:

• 1 the distribution of the stride length over the leg pivot angle for different pedestrians, determined empirically,
• 2 the recursive estimation of the offset b (intersection of the compensation lines of the 1 with the Y-axis); each line represents a different trial,
• 3 a schematic representation of the estimation of the stride length based on the Beinschwenkwinkels and
• 4 a curve for error consideration.

wobei Δθ_k die Amplitude der Oberschenkel-Winkelwellenform zum k-ten Zeitpunkt ist und (a, b) die Parameter des Modells der Regressionsgeraden erster Ordnung sind. Bei dem Term e_k handelt es sich um eine nicht beobachtbare Zufallsvariable, die den Fehler in dem Modell repräsentiert [5].Problem: Pedestrian dead reckoning using the step length and motion estimation approach requires estimating the stride length of the pedestrian (see eg DE 10 2015 209 607 A1 ). Assuming that the pivoting (θ) of the thigh of the pedestrian can be estimated, ie with the help of a thigh-mounted IMU as in 3 , the step length (S _k ) of the pedestrian can be modeled at the k-th time as: $$S_k=a\cdot \mathrm{\Delta }{\theta }_k+b+e_k.$$

where Δθ _{k is} the amplitude of the femoral angle waveform at the k th time and (a, b) are the parameters of the model of the first order regression line. The term e _k is an unobservable random variable that represents the error in the model [5].

The model is in 1 illustrated, wherein the data of three pedestrians A, B, C are reproduced. Each step is marked by a cross [2]. The main disadvantage of the model is that it requires the use of universal parameters. These are estimated to match the regression lines of a set of pedestrians. This means that the parameters do not optimally model the regression line of a single pedestrian.

1 shows that the model slope, a, between individuals is similar to the differences in offset, b, [4]. Based on this, the author proposes in [2] the model ( 1 ) in two stages. First, the slope must be determined for an individual pedestrian. Second, the offset b must be estimated by going over a certain distance [2]. Although this approach is effective, it makes it necessary to know the route design of a particular route. In addition, the approach requires that the pedestrian, before using the INS, undertake a gait intended solely for calibration.

It is assumed that inertial data is available from the foot of the pedestrian, for example because an IMU is integrated in the pedestrian's shoe. In this case, information about the stride length of the pedestrian can be obtained by means of the foot IMU [6]. This information can be used to calibrate the model ( 1 ) be used. In addition, since calibration is required only at the beginning of the aisle, the foot IMU can be turned on once calibration is complete.

Thus, the problem underlying the present invention is how to reduce the offset b in the model (FIG. 1 ) with a pedestrian who carries two IMUs at the same time.

Automatic calibration: The calibration according to the invention estimates the offset b for each individual by taking advantage of the knowledge of the stride length provided by the IMU embedded in the shoe [6].

2 Fig. 10 shows an example of the offset b estimated by the following proposed calibration method.

wobei N die Gesamtzahl der Schritte ist. Für einen jeden Test-Gang kann die Energie des Fehlers gegenüber den plausiblen Werten des Versatzes b abgebildet werden, z.B. wie in 4. Mathematisch ausgedrückt kann der Versatz b jeden realen Wert annehmen. Unter dem physikalischen Aspekt jedoch ist sein Wert auf einen bestimmten Bereich beschränkt. Der Wert von b hängt von dem Wert der Neigung, a, und der Amplitude der Oberschenkel-Winkelwellenform, Δθ, während des Gehens ab. Der erste Wert, a, ist an den universellen Wert fixiert, der in der Größenordnung von 0,05m/° liegt. Der zweite Wert, Δθ, liegt im Bereich von 15°to 60°, je nach der Gehgeschwindigkeit [2]. In Anbetracht dessen, dass die menschliche Schrittlänge in Abhängigkeit von der Körpergröße und der Gehgeschwindigkeit im Bereich von 0,5 m bis 1,5 m liegt, sind die Werte von normalerweise innerhalb maximal ±1 m eingeschränkt. Diese Begrenzung der Werte von b ist erforderlich, um zu verstehen, warum in 4 der Versatz b nur im Bereich von ±10 m wiedergegeben ist. Ferner ist diese Beschränkung erforderlich, um die Durchführbarkeit des nächsten Schritts, d.h. der Minimierung von (2), zu gewährleisten. The proposed calibration method is based on the least squares method. Let S ' _k be the k-th stride length known from the foot-mounted IMU. The mean square error in the step length estimation, x (b), is given by: $$\xi \left(b\right)={\displaystyle \underset{k=1}{\overset{N}{\Sigma }}{\left(S_k^{\text{'}}-a\cdot \mathrm{\Delta }{\theta }_k-b\right)}^2.}$$

where N is the total number of steps. For each test cycle, the energy of the error can be mapped against the plausible values of offset b, eg as in 4 , In mathematical terms, the offset b can take any real value. From a physical point of view, however, its value is limited to a specific range. The value of b depends on the value of the slope, a, and the amplitude of the femoral angle waveform, Δθ, during walking. The first value, a, is fixed at the universal value, which is on the order of 0.05m / °. The second value, Δθ, is in the range of 15 ° to 60 °, depending on the walking speed [2]. Considering that the human step length is in the range of 0.5 m to 1.5 m, depending on height and walking speed, the values are normally restricted within a maximum of ± 1 m. This limitation of the values of b is required to understand why in 4 the offset b is reproduced only in the range of ± 10 m. Furthermore, this limitation is necessary to ensure the feasibility of the next step, ie the minimization of ( 2 ), to ensure.

4 shows that in the area where b for the model ( 1 ), a single value of b exists which minimizes the energy of the error. This value satisfies: $$\frac{\partial \xi \left(b\right)}{\partial b}=\mathrm{0th}$$

Thus, solving the above equation for offset b results in: $$b_k=\frac{1}{N}{\displaystyle \underset{k=1}{\overset{N}{\Sigma }}\left(S_k^{\text{'}}-a\cdot \mathrm{\Delta }{\theta }_k\right).}$$

The latter can be seen as a low-pass filtered estimate of the offset b over all steps. In addition to minimizing the quadratic error ( 4 ), this is simple and can be implemented recursively: $$b_k=b_{k-1}\cdot \frac{k-1}{k}+\frac{S_k^{\text{'}}-a\cdot \mathrm{\Delta }{\theta }_k}{k},$$

This recursive equation avoids the need to store information, e.g. The set of values of S '.

In 2 the offset b is reproduced in accordance with ( 5 ) is appreciated for 3 pedestrians during 6 courses. It can be seen that the offset estimate converges after about 2 minutes. The convergence of the offset shows that the calibration requires only a limited time at the beginning of a gear. After this time, called the convergence time, no further data from the foot IMU is required. Thus, the foot IMU can be shut down, eg to save energy.

An interesting result in 2 is that the displacement b of the pedestrian 1 gives two different values. This is the case because one of the aisles was taken on a different day and with a different location of the IMU on the thigh. This behavior shows an advantage of the proposed algorithm. Automatic calibration estimates the offset best suited to the model ( 1 ), although the pocket IMU is located at a different part of the thigh.

As can be seen from the above description, the method according to the invention deals with the necessity of calibrating the models for step length estimation. The method according to the invention is used as a special step length estimation method. This is explained in [2].

The step size estimation model is used to estimate the length of a pedestrian's step as it goes. There are numerous types of such models, some of which are listed in [2], [7], [8] and in references within these publications. The main feature of these models is that they have to be adapted to each pedestrian.

wobei S_L die Schrittlänge ist und p₁,p₂,...,p_n die Parameter repräsentieren, die einige physische Merkmale der menschlichen Gangart, z.B. die Geh-Frequenz, auf die Schrittlänge beziehen.The step length estimation model can be written as: $$S_L=f\left({\mathrm{<figure-callout\; id="1"\; label="p"\; filenames="DE102017103607A1\_0007.png"\; state="\{\{state\}\}">p</figure-callout>}}_1.{\mathrm{<figure-callout\; id="2"\; label="p"\; filenames="DE102017103607A1\_0007.png"\; state="\{\{state\}\}">p</figure-callout>}}_2.....p_n\right)$$

where S _L is the stride length, and p _1, p _2, ..., p _n represent the parameters relating some physical characteristics of the human gait, such as the Go-frequency, to the step length.

• • Setzen der Parameter des Modells auf vorbestimmte Werte [1]. In diesem Fall wird das Modell zur Schrittlängenschätzung die tatsächliche Schrittlänge des Fußgängers überschätzen oder unterschätzen.
• • Manuelle Kalibrierung [2], z.B. durch Gehen über eine vorbestimmte bekannte Strecke und Zählen der Anzahl von Schritten. Diese beiden Werte erlauben das Schätzen der mittleren Schrittlänge. Dann werden die Parameter des Modells geschätzt, um die mittlere Schrittlänge anzupassen.
• • Kalibrierung mit Hilfe von Landkarten [3]. In diesem Fall geht der Fußgänger über eine vorbestimmte Strecke, deren Länge aufgrund der Landkarte bekannt ist. Dann wird ein Ansatz ähnlich dem zuvor beschriebenen Ansatz verfolgt. Anschließend muss die Anzahl der auf dieser Strecke gegangenen Schritte gezählt werden. Daraufhin wird die mittlere Schrittlänge geschätzt, und anschließend werden die Parameter des Modells geschätzt, um diese mittlere Schrittlänge anzupassen.

The current state of the art follows two main approaches:

• • Set the parameters of the model to predetermined values [1]. In this case, the step size estimation model will overestimate or underestimate the pedestrian's actual stride length.
• • Manual calibration [2], eg by walking over a predetermined known distance and counting the number of steps. These two values allow estimating the average stride length. Then the parameters of the model are estimated to adjust the mean stride length.
• • Calibration using maps [3]. In this case, the pedestrian goes over a predetermined distance, the length of which is known due to the map. Then, an approach similar to the approach described above is followed. Then count the number of steps taken on this route. The average stride length is then estimated, and then the parameters of the model are estimated to adjust this median stride length.

In the method according to the invention, no user profiles are required. As previously mentioned, calibrating the model requires ( 6 ) that the pedestrian attaches an IMU to his foot.

• • Gesundheitssektor
• • umgebungsunterstütztes Wohnen
• • Trägheitsnavigation für Massenmarkt-Anwendungen oder Anwendungen für professionelle Benutzer.

This approach can be used in all fields where it is necessary to estimate the length of a pedestrian's step by means of such a model, for example:

• • Health sector
• • Ambient assisted living
• • Inertia navigation for mass market applications or professional user applications.

BIBLIOGRAPHY

1. [1] T. Moder, K. Wisiol, P. Hafner, and M. Wieser, "Smartphone-based indoor positioning and motion motion recognition", Indoor Positioning and Indoor Navigation (IPIN), 2015 International Conference on, Oct 2015, pp. 1-8 ,
2. [2] E. Munoz Diaz, "Inertial Pocket Navigation System: Unaided 3D Positioning", Sensors, vol. 15, pp. 9156-9178, 2015 ,
3. [3] FT Alaoui, D. Betaille, and V. Renaudin, "A multi-hypothesis particle filtering approach for pedestrian dead reckoning", 2016 International Conference on Indoor Positioning and Indoor Navigation (IPIN), Oct 2016, pp. 1-8 ,
4. [4] Indoor and Indoor Navigation (IPIN), 2014 International Conference on, October 2014, p. 105-110 ,
5. [5] S. Haykin, Adaptive Filter Theory (3rd Ed.). Upper Saddle River, NJ, USA: Prentice-Hall, Inc., 1996.
6. [6] D. Bousdar, E. Munoz Diaz, and S. Kaiser, "Performance Comparison of foot-and-pocket-mounted inertial navigation systems," in International Conference on Indoor Positioning and Indoor Navigation, October 2016
7. [7] V. Renaudin, M. Susi, and G. Lachapelle, "Step Length Estimation Using Handheld Inertial Sensors", Sensors (Basel, Switzerland), vol. 12, no. 7, pp. 8507 {25, 2012 ,
8. [8th] Y. Sun, H. Wu, and J. Schiller, "A step-length estimation model for position tracking", 2015 International Conference on Location and GNSS (ICL-GNSS), pp. 1 {6, June 2015 ,

QUOTES INCLUDE IN THE DESCRIPTION

This list of the documents listed by the applicant has been generated automatically and is included solely for the better information of the reader. The list is not part of the German patent or utility model application. The DPMA assumes no liability for any errors or omissions.

Cited patent literature

• DE 102015209607 A1 [0012]

Cited non-patent literature

• T. Moder, K. Wisiol, P. Hafner, and M. Wieser, "Smartphone-based indoor positioning and motion motion recognition", Indoor Positioning and Indoor Navigation (IPIN), 2015 International Conference on, Oct 2015, pp. 1-8 [0031]
• E. Munoz Diaz, "Inertial Pocket Navigation System: Unaided 3D Positioning", Sensors, vol. 15, pp. 9156-9178, 2015 [0031]
• F.T. Alaoui, D. Betaille, and V. Renaudin, "A multi-hypothesis particle filtering approach for pedestrian dead reckoning", 2016 International Conference on Indoor Positioning and Indoor Navigation (IPIN), Oct 2016, pp. 1-8 [0031]
• E. Munoz Diaz and A.L.M. Gonzalez, "Step Detector and Step Length Estimator for an Inertial Pocket Navigation System", Indoor Positioning and Indoor Navigation (IPIN), 2014 International Conference on, Oct 2014, pp. 105-110 [0031]
• V. Renaudin, M. Susi, and G. Lachapelle, "Step Length Estimation Using Handheld Inertial Sensors", Sensors (Basel, Switzerland), vol. 12, no. 7, pp. 8507 {25, 2012 [0031]
• Y. Sun, H. Wu, and J. Schiller, "A step-length estimation model for position tracking", 2015 International Conference on Location and GNSS (ICL-GNSS), pp. 1 {6, June 2015 [0031]

Claims (2)

Method for determining the stride length of a pedestrian based on the detection of a step, for. As a function of the size of the resulting when walking or running Beinschwenkwinkels his legs, wherein in the process a) a step of the pedestrian, z. The size of the leg swing angle is determined by means of a main inertial measurement unit carried by the pedestrian at a location of his body above the ankles, b) a mathematical model describing the context of a pedestrian's step, e.g. Provided or developed based on the size of the leg swing angle and the resulting stride length, depending on the size and / or movement of the body when walking or running, c) the mathematical model and / or the main inertial measurement unit is calibrated by means of an auxiliary inertial measurement unit to be temporarily worn by the pedestrian on at least one of his feet or ankles, d) measurement data of the main inertial measurement unit are entered into the mathematical model and e) the step length is estimated by means of the calibrated model or the calibrated main inertial measuring unit on the basis of the measured data supplied by the latter when the pedestrian walks or runs. Method according to Claim 1 , characterized in that a smartphone and / or a tracking watch and / or a tracking bracelet and / or the same other so-called wearable device with integrated inertia measuring unit is used as the main inertial measuring unit.

Images