KR102124763B1 - Localization Algorithm Mixing Kalman and Particle Filters for Moving Object in Indoor Wi-Fi Environment - Google Patents

Abstract

In the Wi-Fi environment according to the present invention, the method for tracking the position of a moving object using a Kalman filter and a particle filter in a computer system using a Wi-Fi environment uses a Kalman filter and a particle filter in accordance with the characteristics of the moving pattern of the moving object. A moving object location tracking method comprising: collecting an RSSI signal value from an AP of a Wi-Fi (S100); Measuring the angular velocity when the moving object is moving along an arbitrary path (S200); When the moving object is moving along an arbitrary movement path, determining a straight line or a curved section by comparing with a preset threshold (S300) and analyzing a pattern for a moving direction of the moving object and applying a filter suitable for this. It is characterized by including (S400).

Description

Localization Algorithm Mixing Kalman and Particle Filters for Moving Object in Indoor Wi-Fi Environment using Kalman Filter and Particle Filter in Wi-Fi Environment

The present invention relates to a method for tracking a moving object in consideration of characteristics of a moving object and reducing errors due to a wireless environmental element by mixing a Kalman filter and a particle filter in an indoor Wi-Fi environment. will be.

Filter is used in many ways. In light, a color filter is used to limit the wavelength range, and an interference filter is used to narrow the band. An electric filter is used to limit the frequency band of radio waves or electric signals. Filters are also referred to as interfering with the passage of particles and passing only gases or liquids or particles of any size or less.

The filter in the present invention means that it serves to filter the necessary values in some data. Also, a filter that filters out the necessary values when new data is added to existing data is called a recursive filter. Recursive filters are good for computational efficiency and efficient use of memory in that they use the previous result values.

The Kalman Filter is a filter developed by Rudolf E. Kalman, which is a recursive filter that tracks the dynamic state of noise and is based on measurements made over time. Kalman filtering is an optimal state estimation process applied to a dynamic system that includes irregular external disturbances, and Kalman filter carries noise measured for each discrete real time interval. It is a recursive algorithm of linear, unbiased, and minimum error variance for optimal estimation of unknown state variables of dynamic systems from data. It has been widely used in satellite navigation, missile trajectory, radar, etc. in the industrial field. With the recent development of high-speed, high-performance microprocessors, Kalman filters are increasingly used in very complex real-time processing systems. Since the Kalman filtering procedure is designed to estimate state variables in a linear model, if the model is nonlinear, a linearization procedure must be performed in the process of deriving the filtering equation. In this case, real-time linear Taylor approximation from the previously estimated state variable time linear Taylor approximation), and the Kalman filter obtained here is called Extended Kalman Filter. This method makes it possible to handle nonlinear models very well, and is very simple and effective, and is widely used in many real-time applications. Kalman filter applications are largely characterized by parameter identification and state variable estimation. Are distinguished.

Next, the particle filter is one of the simulation-based prediction techniques and is also called the continuous Monte Carlo method. Particle filters are important in econometrics. Particle filters are usually used to estimate the Bayesian model. This is a case where the latent variables are related to each other by the Markov chain. It is similar to the Hidden Markov Model (HMM), but the state space of variables that are not normally revealed is continuous and not limited enough to be accurately estimated. For example, in a linear dynamic system, the state space of the latent variable is limited to the Gaussian distribution, so an accurate estimate can be made effectively with a Kalman filter only. In the context of HMM, and related models, filtering determines the distribution of latent variables at a particular time, but observations are given up to that time. The reason for the name particle filter is that it uses a group of "particles" (examples of distributions with different weights) to "filter" the approximation.

In addition, the particle filter is made similarly to the Markov chain Monte Carlo (MCMC) batch processing method, which is sometimes similar to the import sampling method. Good particle filters are much faster than MCMC. Sometimes an extended Kalman filter (EKF) or an unscented Kalman filter (UKF) is used instead. The advantage over EKF or UKF is that it is more accurate than EKF or UKF, because if the sample is sufficient, the Bayesian optimal estimate is approached. However, problems can arise if the number of samples is not sufficient. A particle filter and a Kalman filter can also be used in combination. This is a method of using a kind of Kalman filter to propose a distribution for the particle filter.

Meanwhile, as Wi-Fi has recently become popular, research on location estimation technology has been actively conducted in various wireless service fields. In particular, the interest in technology for tracking a moving object and a technology for estimating a location in a room where a wireless network environment is established is very high.

The most commonly used signalprint scheme for indoor positioning and tracking has good reliability, but there is no guarantee that the received signal strength indicator (RSSI) value is reliably collected. If the collected RSSI value is outside the predicted range There is a possibility of different location estimation results. Therefore, various filters may be applied to solve this problem.

The Kalman filter (KF) is a technique that obtains the best results when it has a linear structure. When applied to the location estimation technique, any object is stationary without movement or can move in one direction at a constant speed. do. However, there are problems in applying the Kalman filter as it is because the moving patterns of the actual moving objects are various.

Also, in general, the signal map-based location tracking algorithm in an indoor Wi-Fi environment is composed of two stages as shown in FIG. 1, but has a problem that cannot be used depending on characteristics of a given radio wave environment. This method consists of the first step in which the device measures and records the RSSI signal from the neighboring AP, and the second step in which the RSSI data is corrected using a filter and then compared with a previously prepared signal map to estimate the nearest location.

KR 10-1252531 B1 KR 10-1326644 B1

The present invention seeks to improve the position tracking accuracy of a moving object by mixing a Kalman filter and a particle filter to overcome this problem. That is, in the indoor Wi-Fi environment, the Kalman filter and the particle filter are mixed to apply the filtering according to the characteristics of the moving pattern of the moving object, thereby reducing errors due to wireless environmental factors.

In the Wi-Fi environment according to the present invention, the method for tracking the position of a moving object using a Kalman filter and a particle filter in a computer system using a Wi-Fi environment uses a Kalman filter and a particle filter in accordance with the characteristics of the moving pattern of the moving object. A moving object location tracking method comprising: collecting an RSSI signal value from an AP of a Wi-Fi (S100); Measuring the angular velocity when the moving object is moving along an arbitrary path (S200); When the moving object is moving along an arbitrary movement path, determining a straight line or a curved section by comparing with a preset threshold (S300) and analyzing a pattern for a moving direction of the moving object and applying a filter suitable for this. It is characterized by including (S400).

The present invention discloses a method for tracking the position of a moving object to complement each other's shortcomings by mixing Kalman filters and particle filters with each other in an indoor Wi-Fi environment. In the straight section scenario, the error distance performance of the disclosed method was improved by 80% compared to the conventional filter in the later stage of measurement. The present invention has a more adaptive and large correction effect compared to a conventional filter in a section in which a moving path of a moving object is not constant as well as a result of application in a simple straight or curved section.

1 is a conceptual diagram of a location tracking algorithm using a conventional filter.
2 is a conceptual diagram showing the processing steps of the location tracking algorithm according to the present invention.
3 is a macro code of a location tracking algorithm according to an embodiment of the present invention.
4 is a diagram showing a plan view of an experimental space and a space used to construct a fingerprint map according to an embodiment of the present invention.
5 is a scenario of a moving and stationary object according to an embodiment of the present invention.
6A and 6B are signal map configurations of an experiment target area according to an embodiment of the present invention.
7 is a graph comparing error distances in a straight section according to an embodiment of the present invention.
8 is a graph comparing error distances in a mixing section according to an embodiment of the present invention.
9 is a graph showing a comparison of error distances at point 45 according to an embodiment of the present invention.
10 is a graph showing a comparison of error distances at point 91 according to an embodiment of the present invention.

Hereinafter, with reference to the drawings according to an embodiment of the present invention, this is for easier understanding of the present invention, and the scope of the present invention is not limited thereto.

When a part of the specification "includes" a certain component, this means that other components may be further included instead of excluding other components, unless specifically stated to the contrary. In addition, terms such as "... sub" and "module" described in the specification mean a unit that processes at least one function or operation, which may be implemented in hardware or software, or a combination of hardware and software. .

Throughout the specification, when a part is "connected" to another part, this includes not only "directly connected" but also "electrically connected" with another element in between. .

In the present specification, when one component'transmits' data or a signal to another component, the component may directly transmit the data or signal to another component, and through at least one other component This means that data or signals can be transmitted to other components.

Prior to the description, a number of aspects and embodiments are described herein, and these are merely illustrative and not limiting.

After reading this specification, skilled artisans will appreciate that other aspects and embodiments are possible without departing from the scope of the invention.

Before addressing the details of the embodiments described below, some terms will be defined or clarified.

AP stands for Access Point and has the meaning of a wireless relay base station. And it has the function of interworking with the antenna and wireless signal processing, management function, and wired network and wireless network.

RSSI stands for "Received Signal Strength Indication" and means received signal strength. It is a very common name for radio/RF received signal strength including noise, and it means the strength of a received radio signal that can be easily measured and checked.

The Kalman Filter is a filter developed by Rudolf E. Kalman, which is a recursive filter that tracks the dynamic state of noise and is based on measurements made over time. Kalman filtering is an optimal state estimation process applied to a dynamic system that includes irregular external disturbances, and Kalman filter carries noise measured for each discrete real time interval. It is a recursive algorithm of linear, unbiased, and minimum error variance for optimal estimation of unknown state variables of dynamic systems from data.

Particle filter is one of the simulation based prediction techniques, also known as the continuous Monte Carlo method. Particle filters are important in econometrics. Particle filters are usually used to estimate the Bayesian model. This is a case where the latent variables are related to each other by the Markov chain. It is similar to the Hidden Markov Model (HMM), but the state space of variables that are not normally revealed is continuous and not limited enough to be accurately estimated.

A parameter, also called a parameter in computer programming, is a special kind of variable, and is used to point to one of several data provided as input to a subroutine such as a function. Here, various data provided as inputs of a subroutine are called transfer factors. Normally, the list of parameters is included in the definition part of the subroutine, and each time the subroutine is called, the parameters used in the call are assigned to the corresponding parameters.

An AP (Access Point) is a wireless terminal, which means providing a wireless LAN service. A common internet sharer is a type of AP.

1 is a conceptual diagram of a location tracking algorithm using a conventional filter, FIG. 2 is a conceptual diagram showing a processing step of a location tracking algorithm according to the present invention, and FIG. 3 is a macro code of a location tracking algorithm according to an embodiment of the present invention , FIG. 4 is a diagram showing a plan view of an experimental space and a space used for constructing a fingerprint map according to an embodiment of the present invention, and FIG. 5 is a scenario of a moving and stationary object according to an embodiment of the present invention, and FIG. 6A ,b is a signal map configuration of an experimental target area according to an embodiment of the present invention, FIG. 7 is a graph comparing error distances in a straight section according to an embodiment of the present invention, and FIG. 8 is an embodiment of the present invention 9 is a graph comparing the error distances in the mixing section, and FIG. 9 is a graph showing the comparison of the error distances at point 45 according to an embodiment of the present invention, and FIG. 10 is at point 91 according to an embodiment of the present invention It is a graph showing the comparison of the error distance.

Referring to FIG. 2, a method of tracking a moving object using a Kalman filter and a particle filter in a Wi-Fi environment according to the present invention is a moving pattern of a moving object using a Kalman filter and a particle filter in a computer system using a Wi-Fi environment. In accordance with the characteristics of the mobile object location tracking method, comprising: collecting the RSSI signal value from the AP of the Wi-Fi (S100); Measuring the angular velocity when the moving object is moving along an arbitrary path (S200); When the moving object is moving along an arbitrary path, determining a straight line or a curved section by comparing with a preset threshold (S300) and applying a suitable filter according to the result obtained in the step (S400) It is characterized by including.

In addition, after the step of applying the filter (S400), the step of correcting the position coordinates (S500) may be further included.

In addition, in the step of applying the appropriate filter (S400), the difference between the angle calculated by the angular velocity and the angle calculated by the currently measured angular velocity is calculated in the step S200 of measuring the angular velocity. If it is a curve, a particle filter is applied.

In addition, the particle filter is pre-stored in the computer system by defining noise and the number of particles, and the determining step (S300) includes applying the filter while the result is a curve (S400). If a particle filter is used, the process proceeds to step S500 of correcting the noise and the number of particles without an initialization process. Otherwise, the filter according to the movement pattern is applied after initializing the noise and the number of particles. Step (S400) and the step of correcting (S500) proceeds.

In addition, the Kalman filter sets parameters (A, H, Q, R) used in the computer system and stores them in advance in the computer system, and the determining step (S300) applies the filter while the result is a straight line. If the Kalman filter is used in the step S400, the Kalman filter using the parameters is passed to the correcting step S500 without an initialization process. Otherwise, the Kalman filter is initialized using the parameters and then moved. The step of applying the filter according to the pattern (S400) and the step of correcting (S500) are performed.

To describe the design of the moving object location tracking algorithm in more detail, we disclose a moving object location tracking algorithm that uses a Kalman filter and a particle filter in a Wi-Fi environment according to the characteristics of the moving pattern of the moving object. The processing step of the proposed location tracking algorithm is largely composed of three steps, as shown in FIG. 2, and the pattern for the moving direction of the moving object is analyzed and a filter suitable for this is selected to process the location tracking.

In the first step, the RSSI signal value is collected, and in the second step, when the moving object is moving along a random movement path, the angular velocity is obtained by using the gyro sensor for a specific movement pattern, and then compared with a specific threshold set in a straight line or curve Determine the interval. In the final third step, the position coordinates are corrected after applying a filter suitable for a specific movement pattern. 3 is a representation of the location tracking algorithm proposed by the present invention in macro code.

Next, to describe the filter initialization and RSSI data collection step (step 1), the Kalman filter is a filter that is frequently used in a section having a linear characteristic, and noise is first calculated in consideration of characteristics of the measured environment and used in the system model. Set the parameters (A, H, Q, R) to be stored in the system. In addition, as shown in Figure 3, the particle filter defines the number of noises and particles inside and stores them in the system in advance.

Since the initialization process of the Kalman filter and the particle filter does not end only once, but must be reused according to the result of step 2, the fixed parameters should be set so as not to change. The time interval at which the system collects RSSI data is set to 10 seconds to obtain stable data, which is the minimum measurement interval of the mobile device on which the location estimation algorithm is performed.

Next, the angular velocity calculation and filter determination step (step 2) will be described. In step 2, a 3-axis gyro is a method for determining which filter to apply to a specific movement pattern when the moving object is moving along a specific movement path. Use the angular velocity of the scope sensor. Equation 1 represents an expression for calculating the rotation angle using the angular velocity, and corresponds to the Algorithm Check_Angle module of FIG. 3. In Equation 1, w denotes the angular velocity and DPS denotes the amount of angular change per unit time. The selection of the Kalman filter or particle filter is determined by calculating the difference between the angle calculated at the angular velocity in the previous measurement and the angle calculated at the current measured angular velocity. That is, if the movement pattern is a straight line, a Kalman filter is applied, and if it is a curve, a particle filter is applied.

<Equation 1. Formula to calculate the rotation angle using angular velocity>

Next, to describe the step of applying the filter based on the movement pattern and correcting the position coordinate (step 3), when the processes of steps 1 and 2 are completed, apply the appropriate filter to the corresponding movement pattern to correct the position in step 3. First, if the result of step 2 is a curve, if the particle filter was used in step 1, then the correction step of step 3 is performed without an initialization process. Otherwise, the number of particles and noise defined in step 1 are reinitialized and the movement pattern is reset. The filter determination step and the position coordinate correction step are performed.

The estimation and correction steps in the particle filter are largely divided into prediction, update, and resampling steps, and are expressed as Equation 2. As shown in FIG. 3, the particle filter is performed when the result value of the Check_Angle module is 1, which is a value assigned to the k-th z in Equation 2.

<Equation 2>

Second, if the result of step 2 is a straight line, if the Kalman filter was used in step 1, then the correction step of step 3 is performed without the initialization process as in the first case, otherwise the parameters used in the system model defined in step 1 After initializing the Kalman filter using (A, H, Q, R), the filter determination step according to the moving pattern and the position coordinate correction step are performed.

)는 시간 t의 위치에 따라 계속 변화하며 추정 값(x_t)의 신뢰도에 영향을 준다.After calculating the internal variable Kalmin gain (K _t ) for predicting data in the Kalman filter, the estimated value (x _t ) of the corresponding section is calculated as in Equation 3. The calculation process of the Kalman filter is performed when the result of the Check_Angle module is 0 in the macro code of FIG. 3, wherein Equation 3 means recursively calculating the estimated value (x _t ) in the time t section. do. And the measured value (z _t ) and the previous predicted value (

) Continuously changes with the position of time t and affects the reliability of the estimated value (x _t ).

<Equation 3>

This process is recursively repeated when the moving object moves along the moving path.

The following will be described with respect to the embodiment recommended in the present invention. The moving pattern of the moving object was collected based on the results of the gyroscope sensor. Based on this, simulation was performed using MATLAB and error distance and localization accuracy for each scenario were set. ) Was analyzed.

The arbitrarily set area is the same as in FIG. 4, and the size of the space applied as an example is 46m X 34m in the part shown in red in the right figure of FIG. The AP consisted of three external APs and two private APs, and the RSSI collection equipment was one laptop.

As shown in FIG. 5, the scenarios used in the examples were reviewed as 4 (1-4) and 3 (5-7) targets of moving objects and stationary objects, respectively. Among them, 4 types of scenarios of moving objects and stationary objects were selected in each of 2 types, and RSSI values were collected at least 30 times to improve the reliability of the data collected in the selected scenarios.

In order to construct the signal map, the 2m X 2m area was first defined as one reference point. For reference, the narrower the distance between each zone, the more precise the location tracking. The signal map production method is divided into 141 reference zones in the test area, as shown in Figure 6.

7 and 8 show the results of applying each filter to a moving object, and the cumulative distribution function over time (measurement unit: 10 sec.). FIGS. 9 and 10 show the results of applying each filter to a specific point while the moving object is stopped. Is compared with the error distance over time (measurement unit: 10sec). The proposed algorithm is indicated as “Hybrid” in FIGS. 7 to 10.

7 shows an error distance for a straight path scenario of a moving object. In the initial stage of measurement, when the value of the random variable x is less than or equal to 2, the proposed algorithm's F(x) (cumulative distribution for error distance) is about 50% lower than other filters in the same section, but in the latter stage of measurement, It can be seen that it is about 80% higher than the particle filter. The reason is that there is no change of direction in the linear movement path, so it has the same result as the Kalman filter.

In the linear section, the F(x) value reached 1 when the performance of the particle filter was 8, so it can be seen that this performance is the worst. In addition, as a result of analyzing the pattern of RSSI when collecting data in the same embodiment, it can be seen that an abnormal F(x) value within a range of +-18% was measured every 20-30 measurements. However, this does not affect the overall result due to the low frequency of occurrence, and the maximum error distance of the proposed algorithm is about 4m, which shows that the performance is improved by 2 times compared to the maximum error distance of the particle filter of 8.5m.

8 shows an error distance for an embodiment of a mixed section path in which a straight section and a curved section, which are one of the typical moving patterns of a moving object, are present together. Therefore, it can be seen that the performance of the algorithm proposed by the present invention is improved by 2 m to 6 m compared to the Kalman filter and the particle filter. In addition, the average error distance in the straight section of the mixing section is 8m and 12m respectively for the Kalman filter and the particle filter, whereas the proposed algorithm has a performance improvement of at least 25% at 6m.

In addition, the error distance in the curve section among the mixing sections was 3.5 m while the algorithm of the present invention was 3.5 m while the Kalman filter and the particle filter were 10 m and 7.5 m, respectively. Therefore, it can be seen that the algorithm disclosed in the present invention has a performance of at least 2 times or more than each filter in a curve section.

It can be seen that the correction performance of the proposed algorithm in all sections of the mixed section path scenario is 3 times that of the Kalman filter and 2 times that of the particle filter, as shown in FIG. 8. The error distance is 3m~12m for the Kalman filter and 5m~14m for the particle filter, but the method using the algorithm according to the embodiment proposed by the present invention is 3m~5m, and it can be seen that the average performance is better than 2m~13m.

Although the present invention has a rather large error distance compared to other filters, it was found that the value of the random variable x is greater than or equal to 10. The reason is that the smaller the measurement time interval of RSSI, the more frequently this result occurs. When looking at the results, the error distance correction effect of the proposed algorithm is most effective when the ratio of straight lines and curves in a scenario consisting of mixed sections has a 1:1 or 1:n structure. It was confirmed to be low.

9 is an example according to the present invention to compare the error distance when the stationary object is at an arbitrary point in the experiment space, No. 45. This is the point where there are no obstacles around and the LOS (line of sight) condition is satisfied everywhere. At this point, when collecting RSSI data, it can be seen that the measured signal value fluctuates once every 30 to 60 seconds at 3, 5 to 9, and 15 to 17 (unit: 10 sec).

As can be seen in FIG. 9, at 45 points, the Kalman filter and the proposed algorithm have the highest average error distance of 1 m, which is the best performance. The error distance of the particle filter was found to be up to 7 m at the specific time points 3, 5 to 9 and 15 to 17 above when collecting RSSI data, and it can be seen that performance is very low compared to the algorithm disclosed in the present invention.

10 is an example according to the present invention to compare the error distance when the stationary object is at an arbitrary point 91. This point is a place where two directions are blocked by a wall, so the NLOS (non LOS) condition is satisfied. Despite the NLOS environment, it was confirmed that the average error distance of the algorithm according to the present invention and the Kalman filter and particle filter are both 2 m on average, and the range of the average error distance is within +-15%. In this embodiment as well, when collecting the RSSI data, it can be seen that the measurement signal value fluctuates at 3, 6, 8, 11, 1, as in FIG. 9.

It can be seen from FIG. 10 that the error distance of the particle filter is uniformly 2.2 m at most time points, but is confirmed to be the highest 3.8 m at time 2 and the lowest 1 m at time 12. However, it can be seen that the algorithm according to the present invention not only has an equal error distance at all time points, but also has an average value of 2.2 m, which is more stable than the particle filter.

Predetermined predetermined numerical values described above are described to help understanding of the present invention, which may vary depending on the operator's setting, and it is natural that the aforementioned numerical values are not limited to the present invention.

Although described above with reference to the drawings according to embodiments of the present invention, those skilled in the art to which the present invention pertains will be able to perform various applications and modifications within the scope of the present invention based on the above.

Claims (5)

In a computer system using a Wi-Fi environment, the Kalman filter and the particle filter are mixed to use the moving object location tracking method according to the characteristics of the moving object,
Collecting RSSI signal values from the AP of the Wi-Fi (S100);
Measuring the angular velocity when the moving object is moving along an arbitrary path (S200);
Determining whether a straight line or a curved section is compared with a preset threshold when the moving object is moving along an arbitrary path (S300);
Analyzing the pattern of the moving direction of the moving object and applying a filter suitable therefor (S400) and
After the step of applying the filter (S400), including the step of correcting the position coordinates (S500),
The angular velocity calculation and filter determination of the step (S200) and the step (S400) use the angular velocity of the 3-axis gyroscope sensor, and the selection of the Kalman filter or particle filter is the angle calculated by the angular velocity during the previous measurement and the currently measured angular velocity. It is determined by calculating the difference between the angles calculated by Equation 1 below, and a Kalman filter and a particle filter are mixed in a Wi-Fi environment characterized by applying a Kalman filter if the movement pattern is a straight line and a particle filter if it is a curved line. How to locate a moving object.

w is the angular velocity, DPS is the amount of angular change per unit time
<Equation 1. Formula to calculate the rotation angle using angular velocity>
delete delete The method according to claim 1,
The particle filter defines noise and the number of particles and stores them in advance in the computer system,
In the determining (S300), if the result is a curve and a particle filter is used in the applying of the filter (S400), the noise and the number of particles are corrected without the initialization process (S500).
Otherwise, after initializing the noise and the number of particles, applying the filter according to the movement pattern (S400) and performing the correcting step (S500), the Kalman filter in a Wi-Fi environment. A method of tracking the position of a moving object using a particle filter.
The method according to claim 1,
The Kalman Filter sets parameters (A, H, Q, R) used in the computer system and stores them in advance in the computer system,
In the determining (S300), if the result is a straight line and the Kalman filter is used in the step (S400) of applying the filter at the same time, the process proceeds to the correcting (S500) of the Kalman filter using the parameters without an initialization process.
Otherwise, after initializing the Kalman filter using the parameters, applying the filter according to the movement pattern (S400) and performing the correcting step (S500), Kalman in a Wi-Fi environment. A method of tracking the position of a moving object using a filter and a particle filter.

Images