APPARATUS AND METHOD FOR MEASURING THE ACCURATE POSITION OF MOVING OBJECTS IN AN INDOOR ENVIRONMENT

Abstract

An apparatus and method is disclosed for efficiently sensing and tracking objects in an indoor environment by simultaneously measuring the object movement with an inertial navigation system and a reference location positioning system. By combining the measurements obtained with accelerometers, gyroscopes, angle estimators and the reference system using an extended Kalman filter based approach, a position estimate is obtained with high reliability and precision accuracy. Improvement in performance is obtained by the incorporation of dynamic mode switching and forward-backward smoothing in the location position estimator.

Description

BRIEF DESCRIPTIONS OF THE DRAWINGS

The present invention is described with the help of accompanying drawings:

FIG. 1 shows the present invention on location tracking system.

FIG. 2 shows the process and measurement model for moving object in presence of reference signal.

FIG. 3 shows the location position estimator.

FIG. 4 shows the Kalman-filter designed for moving object in presence of reference signal.

DETAILED DESCRIPTION OF THE INVENTION

The present invention provides an apparatus and method for measuring the position of moving objects with an inertial navigation system, a reference positioning system, and a Kalman-filter based fusion mechanism, which leads to a solution that offers higher accuracy and improved reliability than the application of any of the two basic type-specific location positioning solutions. FIG. 1 shows the preferred embodiment of the location tracking system under consideration. It comprises a reference positioning system 101, an inertial measurement unit 102 with a gyroscope 121 and accelerometers 122, a co-ordinate transformer 103 with an angle estimator 131 and rotation unit 132, and a location position estimator 104. The outputs of the reference positioning system 101 and co-ordinate transformer 103 are connected to the location position estimator 104. The reference positioning system 101 provides estimates r^r_X(t) and r^r_Y(t) on the current position of the object in a target co-ordinate system (X,Y). The estimates are updated not necessarily continuously in time, but whenever new position measurements are available. Each estimate r^r(t) randomly deviates from the position r(t) of the object according to following relation

r ^r(t)=r(t)+_r(t),

where υ_r(t) is additive white Gaussian noise (AWGN) with a variance σ²_υr. The inertial measurement unit 102 as disclosed above measures acceleration and angular velocity in a right-handed Cartesion co-ordinate system (x,y,z) that is body-fixed to the inertial measurement unit 102. In present embodiment, the inertial measurement unit 102 is mounted on the object in such a way that the z-axis is pointing in the opposite direction of the gravity vector; therefore, any movement of the object on a plain floor can be tracked with one gyroscope 121 and two accelerometers 122. Accelerometers 122 provide real-time measurements of acceleration components aⁱ_x(t) and aⁱ_y(t) of the moving object in x- and y-direction. As these measurements are rather noisy due to the vibrations of the object and often biased by a non-negligible, time-varying offset, each measurement aⁱ(t) is generated from the acceleration a(t) and bias b(t) according to the equation

a ⁱ(t)=a(t)+b(t)+υ_a(t),

where υ_a(t) is additive white Gaussian noise (AWGN) with a variance σ²_υa.

The gyroscope 121 measures the angular velocity {dot over (ψ)}_zⁱ(t) around the z-axis. Based on this noisy measurement, the angle estimator 131 continuously estimates the rotation angle {circumflex over (ψ)}_zⁱ(t). In a low-noise environment, this operation can be performed with an integrator. The output value of angle estimator 131 has to be initialized so that the x- and y-axis are aligned to the target co-ordinate system (X,Y). After this initial calibration step, the rotation unit 132 continuously transforms the measurements aⁱ_x(t) and aⁱ_y(t) to the target co-ordinate system by

$\left[\begin{array}{c}{a}_X^i\left(t\right)\\ a_Y^i\left(t\right)\end{array}\right]=\left[\begin{array}{cc}\cos \left({\hat{\psi }}_x^i\right)& -\sin \left({\hat{\psi }}_x^i\right)\\ \sin \left({\hat{\psi }}_x^i\right)& \cos \left({\hat{\psi }}_x^i\right)\end{array}\right]\left[\begin{array}{c}{a}_x^i\left(t\right)\\ a_y^i\left(t\right)\end{array}\right]$

The dynamic and random movements of the object along the X- and Y-axis of the target co-ordinate system and the resulting signals provided by the inertial measurement unit 102 with co-ordinate transformer 103 and the reference positioning system 101 can be modelled as shown in FIG. 2. The acceleration a(t) of the object along one axis of the co-ordinate system is modelled by a first-order Markov process defined by

{dot over (a)}(t)=−αa(t)+w_a(t), α≧0

where w_a(t) represents white Gaussian noise with variance σ²_wa. α defines the correlation between successive acceleration values and is inversely proportional to the time maneuver constant of the object.

Similarly, a time-dependent bias b(t) introduced by sensor imperfections can be modelled by

{dot over (b)}(t)=−βb(t)+w_b(t), β≧0

where w_b(t) is additive white Gaussian noise (AWGN) with variance σ_wb². The correlation coefficient β, however, takes on a larger value than a because the bias b(t) changes with a much slower rate than the acceleration a(t). The position r(t) of the object, which is modeled at the output of the integrator 209, relates to its acceleration a(t), which is modeled at the output of the integrator 202, according to

{umlaut over (r)}(t)={dot over (υ)}(t)=a(t)

where υ(t) denotes the velocity of the object.

The above three linear differential equations represent the continuous-time process model of the investigated system, which can be written in the state-space form

{dot over (x)}(t)=Fx(t)+Gw(t)

where the state vector x(t), the process noise w(t), and the corresponding matrices F and G are given by

$x\left(t\right)=\left[\begin{array}{c}r\left(t\right)\\ \upsilon \left(t\right)\\ a\left(t\right)\\ b\left(t\right)\end{array}\right],F=\left[\begin{array}{cccc}0& 1& 0& 0\\ 0& 0& 1& 0\\ 0& 0& -\alpha & 0\\ 0& 0& 0& -\beta \end{array}\right],w\left(t\right)=\left[\begin{array}{c}{w}_a\left(t\right)\\ w_b\left(t\right)\end{array}\right],G=\left[\begin{array}{cc}0& 0\\ 0& 0\\ 0& 1\\ 1& 0\end{array}\right]$

The measurement model is chosen in accordance with equations r^r(t)=r(t)+v_r(t) and aⁱ(t)=a(t)+b(t)+υ_a(t) as

z(t)=Hx(t)+v(t),

where the output vector z(t), measurement noise v(t), and the matrix H are given by

$z\left(t\right)=\left[\begin{array}{c}{a}^i\left(t\right)\\ r^r\left(t\right)\end{array}\right],H=\left[\begin{array}{cccc}0& 0& 1& 1\\ 1& 0& 0& 0\end{array}\right],v\left(t\right)=\left[\begin{array}{c}{\upsilon }_a\left(t\right)\\ {\upsilon }_r\left(t\right)\end{array}\right]$

The process and measurement models given by above two equations reflects the behavior of the system if the object is moving and location estimates are provided by the reference positioning system. If the object is stationary, a different model for the dynamic behaviour of the object can be obtained by incorporating the constraint a(t)=υ(t)=0 into the system matrix F of the process model as explained above. Similarly, if reference location position estimates are not available, the matrix H in the measurement model can be modified so that the signal r^r(t) in the output vector z(t) is set to 0.

FIG. 3 elaborates the location position estimator 104 in FIG. 1, where the output signals from reference positioning system 101 and co-ordinate transformer 103 are connected to the forward Kalman-filter 303 for providing location-position estimates {circumflex over (r)}(t) in real-time, the backward Kalman-filter 305 with a recording 302 and smoothing 306 unit for providing optimal smoothed estimates {tilde over (r)}(t), and the motion and reference detector 301 with the control unit 304 for controlling the operation of the estimator 104. Here in this section the proposed location-positioning method and corresponding approach are only given for one dimension of the target co-ordinate system (X,Y).

The forward Kalman-filter 303 continuously monitors the noisy measured signals acceleration aⁱ(t) and reference position r^r(t) and computes in real-time a location-position estimate {circumflex over (r)}(t) of the object based on a chosen process and measurement model. This is achieved by, firstly, replicating the process and measurement model without noise sources in the Kalman-filter 303 to generate a state vector estimate {circumflex over (x)}(t) and a corresponding output vector {circumflex over (z)}(t)=H{circumflex over (x)}(t) and, secondly, continuously updating the state vector estimate according to

{circumflex over ({dot over (x)}(t)=F{circumflex over (x)}(t)+K(t)[z(t)−{circumflex over (z)}(t)]

{circumflex over (x)}(t) is an optimal estimate of the state vector x(t) with respect to a mean-squared error criterion. The Kalman-filter 303 is driven by the error between the measured signal vector z(t) and the reconstructed output vector {circumflex over (z)}(t) weighted by the Kalman gain matrix K(t). For obtaining the optimum Kalman gain settings, the covariance matrix of the state vector estimation error is computed.

FIG. 4 shows one of the embodiments for realizing the Kalman-filter 303 obtained for the process and measurement model given in corresponding equations as discussed above. The filter 303 comprises eight time-varying Kalman gains K₁₁(t), K₁₂(t), K₂₁(t) . . . K₄₂(t), two adders 401 and 402 at the input of the Kalman gains, four adders 404 . . . 407 at the output of the Kalman gains, and the process and measurement model 408. The estimates for the position {circumflex over (r)}(t), velocity {circumflex over (υ)}(t), acceleration â(t) of the object, and accelerometer bias {circumflex over (b)}(t) are provided at the output of four integrators in the process and measurement model 408. Since three of the four integrators are cascaded, the filter 303 is especially sensitive to DC-offsets present at the input of the integrators leading to severe drift effects in the integrator output signals. To avoid this inherent deficiency of the filter solution, additional bias and drift compensation techniques are applied as discussed below.

The forward Kalman-filter 303 provides estimates {circumflex over (x)}(t) of the state vector x(t) in real-time by taking into account all information of the measurement signals obtained up to time t. An even better estimate {tilde over (x)}(t) in terms of estimation accuracy can be obtained by performing forward-backward smoothing. This approach exploits additional information on the state vector which is contained in the measurement signals received after time t. The optimal smoothed estimate {tilde over (x)}(t) can thus not be computed in real time, but only after some delay required for collecting future measurements and afterwards post-processing all available data. However, this inherent disadvantage of optimal smoothing is of no importance if the state vector does not change its value when the approach is implemented. Therefore, we propose to apply forward-backward smoothing when the object is not moving in order to enhance the precision of state-vector estimates at past time instants and thus significantly improve the accuracy of the estimate of the current position of the object.

The forward-backward smoothing approach can be realized with a recording unit 302, a forward Kalman-filter 303, a backward Kalman-filter 305, and a smoothing unit 306 as shown in FIG. 3. When the object starts moving at any arbitrary time t₀, the recording unit 302 starts storing the signals acceleration aⁱ(t) and reference position r^r(t) until the object stops at time t₁. At this time t₁, the location position estimator 104 starts performing forward-backward smoothing over the fixed time-interval [t₀, t₁] based on the stored data to obtain optimal smoothed estimates {tilde over (x)}(t) of the state-vector x(t) for any t [t₀, t₁]. The forward Kalman-filter 303 is implemented as discussed above. The filter 303 uses at time t₀ the initial conditions {circumflex over (r)}(t₀)=r₀, {circumflex over (b)}(t₀)=b₀, â(t₀)={circumflex over (υ)}(t₀)=0, runs forward in time from t₀ to t, and computes the forward state vector estimate {circumflex over (x)}(t). The backward Kalman-filter 305 has the same structure as the forward Kalman-filter 303, but the algorithm runs backward in time from t₁ to t. It provides the estimate {circumflex over (x)}^b(t). As initial conditions for the backward filter, the acceleration â(t₁) and the velocity {circumflex over (υ)}(t₁) are set to zero because the object is not moving, while the initial position {circumflex over (r)}(t₁) and bias {circumflex over (b)}(t₁) are modelled as Gaussian random variables. The smoothing unit 306 optimally combines the estimates {circumflex over (x)}^f(t) provided by the forward Kalman-filter 303 and {circumflex over (x)}^b(t) provided by the backward filter 305 to generate the optimal smoothed estimate {tilde over (x)}(t). The optimal combining of the two estimates is accomplished by relatively weighting the estimates in accordance to the covariance matrices of the corresponding state vector estimation error, which are also provided by the forward Kalman-filter 303 and backward Kalman-filter 305.

The motion and reference detector 301 continuously monitors the acceleration signal aⁱ(t) and reference position signal r^r(t), firstly, to detect whether the object is in-motion or stationary and, secondly, to decide for the presence-of-reference or absence-of-reference position signal. The second event can also be directly signaled from the reference positioning system 101 to the location position estimator 104. Motion detector 301 detects the object's motion by continuously tracing the magnitude of the acceleration measurement signal aⁱ(t), which strongly oscillates when the object is moving. The detection can thus be performed by sampling this signal at the rate of 1/T, averaging N consecutive samples, and then deciding for the event object stationary if the obtained mean value is below a pre-defined threshold μ, as defined by

$\stackrel{\_}{a}\left(\mathrm{kT}\right)=\frac{1}{N}\sum _{l=0}^{N-1}a^i\left(\mathrm{kT}-\mathrm{lT}\right)\{\begin{array}{cc}\le \mu & \mathrm{no}\text{-}\mathrm{movement}\\ >\mu & \mathrm{in}\text{-motion}\end{array}$

The parameters N and μ are chosen so that even short stops of the object can be reliably detected and false decisions for not moving are avoided.

Depending on the detected events object in-motion/stationary and presence/absence-of-reference position signal, the control unit 304 dynamically switches the Kalman-filters 303 and 305 between the following modes of operation:

Mode 1: Object In-Motion and Presence-of-Reference Position Signal

In this configuration, the control unit 304 enables the operation of the forward Kalman-filter as shown in FIG. 4. The filter 303 provides estimates of the position, acceleration, and velocity of the object as well as estimates of the accelerometer bias in real-time. The acceleration signal aⁱ(t), reference position signal r^r(t), and the mode parameters are recorded with recoding unit 302 in a storage device in order to perform forward-backward smoothing in Mode 3.

Mode 2: Object In-Motion and Absence-of-Reference Position Signal

Since no reference signal is available, the location position estimator 104 has to derive an estimate which is entirely based on the accelerometer measurements aⁱ(t). This change can be modelled by setting all elements in the second row of the measurement matrix H to zero. The control unit 304 reconfigures the Kalman-filter 303 so that this change in the measurement model is also reflected in the filter 303.

Moreover, the control unit 304 freezes the bias estimate {circumflex over (b)}(t) in the filter 303 to its current value when the mode of operation if switched from presence-of-reference to absence-of-reference position signal. This freeze operation is recommendable to avoid observability problems of simultaneous changes of acceleration and bias in the system model at the absence of a reference signal. The acceleration signal aⁱ(t), reference position signal r^r(t), and the mode parameters are recorded in the storage device with recoding unit 302 to perform forward-backward smoothing in Mode 3.

Mode 3: No-Movement of the Object

Whenever the motion detector 301 detects no movement of the object, the control unit 304 resets the acceleration estimate â(t) and the velocity estimate {circumflex over (υ)}(t) in the Kalman-filter 303 to zero. This operation calibrates the integrator output signals in the process and measurement model 408 to the known reference values of zero, and thus prevents offset accumulation by the integrators. Since the acceleration and velocity of the object is zero, an enhanced process model for the dynamic behavior of the object is obtained by incorporating the constraint a(t)=υ(t)=0 into the process equations. The control unit 304 reconfigures the Kalman-filter 303 so that this change in the process model is also reflected in the filter 303. Since the location position estimator 104 has recorded the signals aⁱ(t) and r^r(t) while the object was moving as explained in Mode 1, the recorded data stored in the last time interval before the object reached its stop can be post-processed by performing forward-backward smoothing with smoothing unit 306. The control unit 304 initializes the execution of the smoothing procedure after the event no-movement has been detected. After post-processing the recorded data, the location position estimator 104 provides the optimal smoothed estimate of the objects location.

It is believed that the present invention and many of its attendant advantages will be understood by the foregoing description. It is also believed that it will be apparent that various changes may be made in the form, construction and arrangement of the components thereof without departing from the scope and spirit of the invention or without sacrificing all of its material advantages. The form herein before described being merely an exemplary embodiment thereof, and it is the intention of the following claims to encompass and include such changes.

Claims

1. An apparatus for providing accurate position of a moving object in an indoor environment, comprising: a Reference Positioning System (RPS)an Inertial Measurement Unit (IMU) installed on said moving object.a Coordinate Transformer (COT) coupled to the outputs of said IMU, anda Location Position Estimator (LPE) coupled to the outputs of said RPS and said COT.

2. An apparatus for providing accurate position of a moving object in an indoor environment, as claimed in claim 1, wherein said IMU comprising: one or more gyroscopic units, andone or more accelerometer units.

3. An apparatus for providing accurate position of a moving object in an indoor environment, as claimed in claim 1, wherein said COT comprising: an angle estimator coupled to the outputs from the IMU, anda rotation unit coupled to the outputs from said IMU and the output of said angle estimator.

4. An apparatus for providing accurate position of a moving object in an indoor environment, as claimed in claim 1, wherein said LPE comprising, for each dimension of the position coordinates, a forward Kalman Filter having one input coupled to the output from the RPS for receiving the reference position signal and a second input coupled to the output from said COT for obtaining the rotated accelerometer measurements.

5. An apparatus for providing accurate position of a moving object in an indoor environment, as claimed in claim 4, wherein said LPE comprising for each dimension of the position coordinates: a recording unit having one input coupled to the output from the RPS and a second unit coupled to the output from said COT,a motion and reference detector having one input coupled to the output from the RPS and a second input coupled to the output from said COT,a backward Kalman Filter having its inputs connected to the outputs of said recording unit,a smoothing unit having one input coupled to an output of said forward Kalman Filter, a second input connected to the output of said backward Kalman Filter, anda control unit coupled to the output of said motion and reference detector, said control unit controlling the operation of said forward Kalman Filter, said backward Kalman Filter, and said smoothing unit.

6. An apparatus for providing accurate position of a moving object in an indoor environment as claimed in claim 4 or claim 5, wherein said forward Kalman Filter comprising: a process and measurement model that models the movement of the object along an axis of the target co-ordinate system, the relation between acceleration and position, and accelerometer imperfections; anda set of Kalman Filter gains whose setting is derived from the covariance matrix of the state-vector estimation error.

7. An apparatus for providing accurate position of a moving object in an indoor environment as claimed in claim 6, wherein: said process and measurement model comprises four integrators providing estimates for the position, velocity, and acceleration of the object, and for the accelerometer bias; andsaid set of Kalman Filter gains comprises eight gains.

8. An apparatus for providing accurate position of a moving object in an indoor environment as claimed in claim 5, wherein said backward Kalman Filter comprising a filter structure as defined for said forward Kalman Filter in claim 6 and claim 7; andmeans for backward processing in time the recorded said rotated accelerometer measurements and said reference position signal, starting with the acceleration and velocity estimates set to zero in said backward Kalman Filter.

9. An apparatus for providing accurate position of a moving object as claimed in claim 5, wherein said smoothing unit combines the estimates of said forward and backward Kalman Filters by weighting the estimates in accordance to the covariance matrices of the corresponding state vector estimation error.

10. An apparatus for providing accurate position of a moving object as claimed in claim 5, wherein said motion and reference detector comprising a sampling device for sampling the absolute value of said rotated accelerometer measurements;an averaging device for averaging a given number of consecutive samples; anda motion detecting device for identifying motion based on said average value being above a defined threshold.

11. A method for accurately determining the position of a moving object in an indoor environment using a combination of an Inertial Measurement Unit (IMU) and a Reference Positioning System (RPS), comprising the steps of: obtaining reference position data from said RPS,obtaining additional data on moving object from said IMU, andestimating the position of the moving object from the data obtained from the RPS and the IMU.

12. A method for accurately determining the position of a moving object in an indoor environment as claimed in claim 11, wherein said additional data is determined by the steps of: estimating angular position of said IMU in body-fixed co-ordinate system based on angular velocity measurements obtained from said IMU,obtaining accelerometer measurements from said IMU along the x- and y-axis of body-fixed co-ordinate system, andtransforming acceleration measurements into target co-ordinate system using the angular position estimate.

13. A method for accurately determining the position of a moving object in an indoor environment as claimed in claim 11, wherein said position estimation is performed by applying a set of forward Kalman Filters to determine the coordinates of the position.

14. A method for accurately determining the position of a moving object in an indoor environment as claimed in claim 11, wherein the position estimation is performed by the steps of: detecting first event “object motion status” based on said rotated accelerometer measurements, and detecting second event “presence or absence of said reference position signal”;switching between modes of operation depending on said first event and said second event according to: if object is in motion and said reference position signal is present, then said forward Kalman Filter provides estimates for position, acceleration, and velocity of object, and an estimate for accelerometer bias based on said rotated accelerometer measurements and said reference position signal;if object is in motion and said reference position signal is absent, then said forward Kalman Filter is reconfigured to reflect absence of reference position measurements in the filter and to freeze estimate of accelerometer bias to present value, and then provides estimates for determining position, acceleration, and velocity of object;if object is in motion, then said rotated acceleration measurements, said reference positioning signal, and said first and second events are recorded for performing forward-backward smoothing at the next stop of the object;if object is stationary, then the acceleration and velocity estimate in said forward Kalman Filter are set to zero, and said forward Kalman Filter is reconfigured to reflect the stationary object status in the filter; and said forward Kalman Filter, said backward Kalman Filter, and said smoothing unit provide smoothed position estimates by performing forward-backward smoothing based on said recorded rotated acceleration measurements and reference position signal.