CN111595344B - Multi-posture downlink pedestrian dead reckoning method based on map information assistance - Google Patents

Multi-posture downlink pedestrian dead reckoning method based on map information assistance Download PDF

Info

Publication number
CN111595344B
CN111595344B CN202010483376.8A CN202010483376A CN111595344B CN 111595344 B CN111595344 B CN 111595344B CN 202010483376 A CN202010483376 A CN 202010483376A CN 111595344 B CN111595344 B CN 111595344B
Authority
CN
China
Prior art keywords
pedestrian
algorithm
course
mode
information
Prior art date
Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
Active
Application number
CN202010483376.8A
Other languages
Chinese (zh)
Other versions
CN111595344A (en
Inventor
陈国良
王轩
靳赛州
张正华
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
China University of Mining and Technology CUMT
Original Assignee
China University of Mining and Technology CUMT
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by China University of Mining and Technology CUMT filed Critical China University of Mining and Technology CUMT
Priority to CN202010483376.8A priority Critical patent/CN111595344B/en
Publication of CN111595344A publication Critical patent/CN111595344A/en
Application granted granted Critical
Publication of CN111595344B publication Critical patent/CN111595344B/en
Active legal-status Critical Current
Anticipated expiration legal-status Critical

Links

Images

Classifications

    • G—PHYSICS
    • G01—MEASURING; TESTING
    • G01C—MEASURING DISTANCES, LEVELS OR BEARINGS; SURVEYING; NAVIGATION; GYROSCOPIC INSTRUMENTS; PHOTOGRAMMETRY OR VIDEOGRAMMETRY
    • G01C21/00—Navigation; Navigational instruments not provided for in groups G01C1/00 - G01C19/00
    • G01C21/20—Instruments for performing navigational calculations
    • G01C21/206—Instruments for performing navigational calculations specially adapted for indoor navigation
    • G—PHYSICS
    • G01—MEASURING; TESTING
    • G01C—MEASURING DISTANCES, LEVELS OR BEARINGS; SURVEYING; NAVIGATION; GYROSCOPIC INSTRUMENTS; PHOTOGRAMMETRY OR VIDEOGRAMMETRY
    • G01C21/00—Navigation; Navigational instruments not provided for in groups G01C1/00 - G01C19/00
    • G01C21/005—Navigation; Navigational instruments not provided for in groups G01C1/00 - G01C19/00 with correlation of navigation data from several sources, e.g. map or contour matching
    • G—PHYSICS
    • G01—MEASURING; TESTING
    • G01C—MEASURING DISTANCES, LEVELS OR BEARINGS; SURVEYING; NAVIGATION; GYROSCOPIC INSTRUMENTS; PHOTOGRAMMETRY OR VIDEOGRAMMETRY
    • G01C21/00—Navigation; Navigational instruments not provided for in groups G01C1/00 - G01C19/00
    • G01C21/10—Navigation; Navigational instruments not provided for in groups G01C1/00 - G01C19/00 by using measurements of speed or acceleration
    • G01C21/12—Navigation; Navigational instruments not provided for in groups G01C1/00 - G01C19/00 by using measurements of speed or acceleration executed aboard the object being navigated; Dead reckoning
    • G01C21/16—Navigation; Navigational instruments not provided for in groups G01C1/00 - G01C19/00 by using measurements of speed or acceleration executed aboard the object being navigated; Dead reckoning by integrating acceleration or speed, i.e. inertial navigation
    • G01C21/165—Navigation; Navigational instruments not provided for in groups G01C1/00 - G01C19/00 by using measurements of speed or acceleration executed aboard the object being navigated; Dead reckoning by integrating acceleration or speed, i.e. inertial navigation combined with non-inertial navigation instruments
    • Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
    • Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
    • Y02D—CLIMATE CHANGE MITIGATION TECHNOLOGIES IN INFORMATION AND COMMUNICATION TECHNOLOGIES [ICT], I.E. INFORMATION AND COMMUNICATION TECHNOLOGIES AIMING AT THE REDUCTION OF THEIR OWN ENERGY USE
    • Y02D30/00—Reducing energy consumption in communication networks
    • Y02D30/70—Reducing energy consumption in communication networks in wireless communication networks

Landscapes

  • Engineering & Computer Science (AREA)
  • Radar, Positioning & Navigation (AREA)
  • Remote Sensing (AREA)
  • Automation & Control Theory (AREA)
  • Physics & Mathematics (AREA)
  • General Physics & Mathematics (AREA)
  • Navigation (AREA)

Abstract

The invention relates to a multi-posture downlink pedestrian dead reckoning method based on map information assistance, and belongs to the technical field of indoor positioning. Firstly, the problem that the equipment posture is limited when a smart phone is used for carrying out pedestrian track calculation in the prior art is solved, and the heading estimation algorithm is improved by calculating the acceleration integral of the moving direction of a user. A lightweight algorithm, namely a finite state machine and a decision tree, is used for monitoring and identifying the use mode of the mobile phone, and the characteristic of a gyroscope at a corner is used for sensing a behavior characteristic point, so that the course estimation performance of a linear stage is improved. In addition, the problem of overlarge error accumulation caused by course angle deviation during positioning is solved. The course is constrained by introducing map matching and particle filtering technology, and the problems of improper transfer of particles, wall penetration of tracks and the like are limited by sensing behavior landmarks in indoor movement. The method can effectively improve the course estimation performance and the positioning robustness of the PDR system, and enhance the pedestrian experience.

Description

Multi-posture downlink pedestrian dead reckoning method based on map information assistance
Technical Field
The invention relates to the technical field of indoor positioning, in particular to a multi-posture downlink pedestrian dead reckoning method based on map information assistance.
Background
In recent years, with the popularization of intelligent mobile devices, people have increasingly strong demand for indoor location services. Because the traditional satellite positioning technology is not available indoors and is difficult to provide reliable position service for indoor environments, finding a reliable and accurate indoor navigation scheme is extremely challenging.
Many researchers have made great progress in the research of indoor positioning technology, such as inertial sensors, bluetooth, geomagnetism, radio frequency identification, ultra wide band, wireless local area network, computer vision, etc. However, most indoor positioning technologies rely on a particular infrastructure and are costly to deploy and maintain. The PDR based on the smart phone is continuously positioned, the data updating is quick, infrastructure is not needed, and the PDR is widely concerned by students. At present, gait detection and step length estimation are greatly explored, but the course estimation of a person descending by using postures is still in a serious challenge, and the problem of error accumulation caused by inertia elements of the person descending by using postures also severely limits the wide popularization of the person descending by using the posture estimation.
Pedestrian heading estimation faces two major challenges. On the one hand, the inertial sensor mounted on the mobile device is not only affected by drift caused by its own characteristics, but also easily affected by environmental disturbances (such as ambient magnetic field and irregular motion of human body). On the other hand, the PDR algorithm has many limitations, such as the direction in which the mobile phone must be directed toward the pedestrian without changing the posture. In addition, error accumulation caused by low-cost sensors makes the PDR method unable to reach acceptable accuracy, which all results in that the traditional PDR is difficult to adapt to the actual requirements of users, and directly influences the popularization of the PDR system.
Disclosure of Invention
In view of the above analysis, although methods based on smartphone pedestrian heading estimation and methods of cumulative elimination of positioning errors have been explored, there are difficulties with the PDR algorithm in multi-mode and multi-pose. Aiming at the problem, the invention aims to provide a multi-posture downlink pedestrian bit calculation method based on map information assistance, which is used for solving the problems of positioning and error accumulation of the existing smart phone under a typical posture and improving the accuracy and robustness of a PDR system. The method mainly comprises two aspects:
1. according to the habit of using the mobile phone by pedestrians in daily life, five typical modes of carrying the smart phone are defined, including five modes of holding by hand, recording, talking, swinging and pocket. In the prior art, the mode of the equipment is mostly identified by continuously extracting sensor data, so that a large amount of memory is consumed, and the service life of a battery is greatly shortened. By matching a predefined threshold, a finite vector machine (FSM) algorithm can respond to mode switching in time, avoiding data feature extraction. Therefore, the invention combines FSM and Decision Tree (DT) algorithm to monitor and identify the mobile phone mode in real time. The method uses the DT algorithm to identify the user's current pattern only after the FSM algorithm detects a change in the device's pose.
2. By introducing gyroscope information, the stepping course estimation method is improved, and the global course of the pedestrian in the linear stage is calculated. On the basis, the invention introduces a particle filter-based map matching algorithm to improve the performance of course estimation and positioning.
The invention is mainly realized by the following technical scheme:
a multi-posture downlink pedestrian dead reckoning method based on map information assistance comprises the following steps:
step 1, low-pass filtering and smoothing preprocessing are carried out on an acceleration original signal acquired by a low-cost sensor embedded in a smart phone, so that adverse effects of factors such as sensor noise and jitter on pattern recognition and positioning are weakened. Correcting the equipment magnetometer in an indoor environment to reduce the influence of internal errors and external interference;
step 2, recognizing the use mode of the mobile phone in a typical scene through a lightweight equipment posture monitoring and recognizing algorithm which is mainly divided into a feature extraction module and a mode monitoring and classifying module;
3, discriminating a pseudo wave peak/pseudo wave valley based on a wave crest detection algorithm by utilizing the preprocessed signals and the equipment use posture output by the mode recognition algorithm, and realizing the gait recognition of the PDR algorithm;
step 4, calculating the global course of the pedestrian linear stage by introducing gyroscope information, obtaining a fusion course with strong robustness in an auxiliary manner based on a particle filter algorithm matched with a map, and correcting the problem of track through-wall by perceiving the behavior characteristic landmarks in the movement of the pedestrian;
and 5, carrying out pedestrian dead reckoning through the gait detection, the mode identification and the course estimation to finally obtain pedestrian position information with high reliability.
Further, in the step 1, a fourth-order Butterworth low-pass filter with the cut-off frequency of 8Hz is used for eliminating the influence of high-frequency noise, and then a moving average algorithm is adopted for smoothing data and eliminating unnecessary burrs; and then, rotating the smart phone around three axes of a mobile phone coordinate system in a space under an indoor environment, and correcting the magnetometer signals by using a least square fitting ellipsoid method for the obtained magnetometer sample data.
The beneficial effect of adopting the further scheme is that: the preprocessed acceleration signal can more clearly reflect the motion characteristics of the pedestrian and weaken the adverse effect of the noise of the sensor on user mode identification, gait identification and positioning; the magnetometer correction reduces the influence of internal errors and external interference and improves the performance of system attitude calculation.
Further, in step 2, in the feature extraction, data slicing is performed by using a sliding window, data features of each window are extracted, the window size is set to 128 samples, and the overlapping degree is 50%.
The classification of the equipment posture is realized through real-time average acceleration information of the pedestrian, and the calculation formula is as follows:
Figure BDA0002518187700000041
in the formula (I), the compound is shown in the specification,
Figure BDA0002518187700000042
representing the three-axis acceleration mean value of the kth sliding window, and WinSize is the size of the sliding window.
The beneficial effect of adopting the above further scheme is: data output by a multi-source sensor assembly built in the smart phone reflects various behaviors of a user to a great extent. However, the discrete data is not enough to analyze the behavior characteristics of the pedestrian, and the sliding window is used to extract the three-axis acceleration mean value in the typical mode, so that the three-axis spatial relationship of the device (as shown in fig. 5) can be reflected, and the posture of the device can be effectively classified.
Further, in step 2, the mode monitoring and classifying module utilizes a finite vector machine (FSM) and a Decision Tree (DT) algorithm to realize real-time monitoring and identification of the equipment attitude, and utilizes the DT algorithm to identify the current mode of the pedestrian only when the FSM algorithm detects that the equipment attitude changes. In the FSM algorithm, the states of the device are totally divided into 6 types, including 5 main modes: message, video, calling, swing, pocket and 1 transition mode TRANS bridging the main mode, wherein the starting mode is the Message mode, when the switching of the device mode is not monitored, the device is always in the current main state, otherwise, the device is switched to the TRANS state; when the system enters a TRANS Mode, the algorithm automatically extracts the acceleration information of the next sliding window, and the accurate Mode information DT Mode after switching is obtained by utilizing the decision tree algorithm, and the pedestrian state is temporarily considered as the RoughMode in the period.
The beneficial effect of adopting the further scheme is that: the FSM algorithm can respond to mode switching in time by matching a predefined amplitude threshold, reducing the large computational load associated with continuous sensor data processing. The DT is a tree-structured nonparametric classifier which can directly reflect the characteristics of signals, if an observed value is given, a corresponding logic expression can be easily introduced according to a generated DT model, and the algorithm is simple and has low complexity. The FSM + DT algorithm can be used for effectively realizing real-time monitoring and identification of the equipment, and the problems of high equipment calculation cost, battery life damage and the like caused by continuous data feature extraction for classification in the prior art are solved.
Further, in the step 3, the preprocessed signals and the device use postures output by the pattern recognition algorithm are used for extracting the acceleration signals preprocessed by the vertical axis of the device for gait detection, and a peak value threshold, an adjacent peak time threshold and an adjacent peak selection mechanism are adjusted according to different postures, and pseudo wave peaks/pseudo wave troughs are discriminated by using multiple conditions to realize gait adaptive recognition;
the beneficial effect of adopting the further scheme is that: acceleration signals of vertical shafts under different postures are extracted, and the influence of irrelevant shaft noise on step counting is weakened; the threshold value of the step-counting algorithm under the typical posture is adjusted in a self-adaptive mode, and compared with the fixed threshold value in the prior art, the adaptive capacity of the step-counting algorithm is improved.
Further, the step 4 comprises two steps of calculating a global heading and estimating the heading assisted by particle filtering and map matching.
Calculating the global course, and acquiring the posture of the intelligent equipment held by the pedestrian in the walking process in real time by adopting a Madgwick-AHRS algorithm; calculating a conversion matrix from the carrier coordinate system to the reference coordinate system by using the equipment attitude to obtain the horizontal acceleration of the pedestrian under the reference coordinate system:
Figure BDA0002518187700000061
in the formula
Figure BDA0002518187700000062
Indicating the acceleration information of the user i at the moment in the coordinate system of the device,
Figure BDA0002518187700000063
representing acceleration information of the user i under a reference coordinate system at the moment;
and calculating the course information in one step according to the pedestrian gait information obtained by the gait detection algorithm:
Figure BDA0002518187700000064
Figure BDA0002518187700000065
in the formula (I), the compound is shown in the specification,
Figure BDA0002518187700000066
representing the stepping course of the pedestrian in the jth step;
Figure BDA0002518187700000067
a velocity vector obtained by integrating the horizontal acceleration; step j And Step j+1 Respectively representing the time instants of the j and j +1 steps acquired by the gait detection algorithm.
Judging the motion characteristics of the pedestrian by the angular speed information of a specific axis of the smart phone under different postures; the method comprises the following steps of obtaining turning information of a pedestrian by utilizing a quartile abnormal value detection algorithm, dividing a pedestrian track into a plurality of linear stages, obtaining a real-time global heading of the pedestrian by utilizing the integral of horizontal acceleration of the linear stages, and realizing continuous correction of the heading of the pedestrian:
Figure BDA0002518187700000068
in the formula (I), the compound is shown in the specification,
Figure BDA0002518187700000071
represents the global heading, step, of Step j k Indicating the time of the last user turn.
The method comprises the steps of particle filtering and map matching assisted course estimation, according to boundary information contained in a map, constraining the transfer of particles and reducing the number of invalid particles, and when the particles are transferred to an unreachable area, assigning the weight of the particles to be 0; otherwise, the weight of the particle is calculated using the right side of the bell-shaped curve of the gaussian distribution:
Figure BDA0002518187700000072
in the formula, N (l) and E (l) represent coordinate values of north and east directions obtained by observing the equation at time l, and N i (l) And E i (l) A priori coordinate values representing the north and east directions of the ith particle. The particle weight calculation also includes resampling of the particles: the particle resampling adopts a random resampling method, high-weight particles are kept as far as possible, and low-weight particles are removed.
To minimize trajectory fluctuations, the update of the pedestrian's position can be calculated by:
Figure BDA0002518187700000073
in the formula,. DELTA.E global And Δ N global The user's displacement in the east and north directions is calculated for the global heading.
And sensing the nearest landmark information by using the turning angle moment provided by the gyroscope, and obtaining a correction value of each step by taking the edge cutting of the track as a basis, thereby correcting the track and avoiding the track from passing through an impassable area.
According to the probability position determined by filtering, the filtered course angle can be reversely calculated:
Figure BDA0002518187700000081
the robustness of the course is further improved, and the fused course can be calculated by the following formula:
Figure BDA0002518187700000082
in the formula, theta global And theta stepwise Representing the global and step headings provided in the above section; ρ is a unit of a gradient 1 And ρ 2 The representative weight value can be adaptively adjusted according to the actual situation so as to improve the heading estimation capability of the model. Delta represents the threshold value of course deviation determined by particle filtering, and is set to be 15 degrees; d 1 And d 2 Represents theta particle And theta global And theta stepwise The difference of (c).
The beneficial effect of adopting the above further scheme is: the global course of the PDR system is calculated, the problems of course deviation accumulation and improper particle transfer in the traditional technology are restrained by combining an ion filtering scheme of map matching and behavior perception, and the course determined by the particles is obtained. The provided fusion course combines the advantages of high global course stability and high particle filter course accuracy, and improves the robustness of course estimation.
In the invention, the technical schemes can be combined with each other to realize more preferable combination schemes. Additional features and advantages of the invention will be set forth in the description which follows, and in part will be obvious from the description, or may be learned by practice of the invention. The objectives and other advantages of the invention will be realized and attained by the structure particularly pointed out in the written description and claims hereof as well as the appended drawings.
The invention finally achieves the following beneficial effects: the method considers the problems existing in course calculation of the PDR system under different using postures, solves the problem that the course under multiple postures is difficult to calculate in the prior art by calculating the acceleration integral of the moving direction of the user, and improves the course estimation performance in a linear stage by sensing the behavior characteristic point by utilizing the characteristics of a gyroscope at a corner. The method utilizes the information of the sensor inside the mobile phone, does not need additional equipment, and is convenient and quick; the monitoring and the recognition of the user use mode are realized by utilizing a lightweight algorithm, the uninterrupted feature extraction for classification in the prior art is avoided, and the algorithm is low in complexity, simple and effective; the particle filter algorithm is utilized to effectively combine the building information and PDR position updating, and the robustness of positioning and course estimation is improved.
Drawings
The drawings, in which like reference numerals refer to like parts throughout, are for the purpose of illustrating particular embodiments only and are not to be considered limiting of the invention.
Fig. 1 is an overall flowchart of a pedestrian dead reckoning method according to an embodiment of the present invention.
FIG. 2 is a schematic diagram of accelerometer and gyroscope data preprocessing in an embodiment of the invention.
FIG. 3 is a schematic diagram of a least squares magnetometer calibration in an embodiment of the invention.
Fig. 4 is a spatial relationship table of the device coordinate system in different scenarios according to the embodiment of the present invention.
FIG. 5 is a diagram illustrating an average acceleration value under different postures according to an embodiment of the present invention.
Fig. 6 is a schematic diagram of time-frequency characteristics of angular velocity when switching the device attitude in the embodiment of the present invention.
Fig. 7 is a schematic diagram of monitoring and identifying a device mode according to an embodiment of the present invention.
Fig. 8 is a schematic diagram of an adaptive gait detection algorithm in an embodiment of the invention.
Fig. 9 is a schematic diagram of an output value of a gyroscope in a handheld gesture according to an embodiment of the present invention.
Fig. 10 is a schematic diagram of a particle filtering algorithm based on map matching according to an embodiment of the present invention.
Detailed Description
The accompanying drawings, which are incorporated in and constitute a part of this application, illustrate preferred embodiments of the invention and together with the description, serve to explain the principles of the invention and not to limit the scope of the invention.
The specific embodiment of the invention discloses a map information-assisted multi-posture downlink pedestrian dead reckoning method, and the specific flow is shown in fig. 1, and the method comprises the following steps:
a multi-posture downlink pedestrian dead reckoning method based on map information assistance comprises the following steps:
and S1, processing data.
When the low-cost sensor embedded in the smart phone is used, the system error caused by the characteristics of the low-cost sensor is easy to interfere by the environment. Therefore, the raw signals obtained need to be pre-processed before using the data to attenuate the adverse effects of sensor noise and errors on user pattern recognition and localization.
Step S101, low-pass filtering and smoothing.
By analyzing sensor data of different device poses during actual use by a user, it is found that the frequency of most of the raw signals is below 8Hz. Therefore, the invention firstly uses a fourth-order Butterworth low-pass filter with the cut-off frequency of 8Hz to eliminate the influence of high-frequency noise, and then adopts a moving average algorithm to further smooth data and eliminate unnecessary burrs. As shown in fig. 2, the filtered acceleration signal and the gyroscope signal contain less noise and are smoother, so that the characteristics of the pedestrian motion can be reflected more clearly.
And step S102, ellipsoid correction.
Magnetometers carried by smartphones are often influenced by internal errors and external interference, and the internal errors can be generally divided into non-orthogonal errors, sensitivity errors, sensor noise and zero offset errors. Magnetometers are important sources of information for computing device heading, and therefore magnetometer calibration is necessary. The invention rotates the smart phone around three axes of the device in the space under the indoor environment, corrects the magnetometer signals by the obtained magnetometer sample data by adopting a least square fitting ellipsoid method, and fig. 3 shows the comparison results before and after correction of the magnetometer data.
And S2, behavior recognition.
In an actual use scene, different device postures and abnormal representation behaviors of pedestrians, data output by the built-in inertial element of the mobile phone often show different characteristics. The invention provides a lightweight equipment attitude monitoring and recognition algorithm which is mainly divided into two steps of feature extraction and mode monitoring and classification.
Step S201, feature extraction.
Since it is difficult to directly obtain accurate attitude information of the device by using the filtered data, a sliding window is required to slice the data and extract data features of each window. The inventive window size was set to 128 samples (2.5 s) with an overlap of 50%.
The pose of the device is typically identified and resolved using window features such as mean, variance, maximum, minimum, kurtosis, skewness, and energy density. Fig. 4 is a spatial relationship of a coordinate system of the device in different usage modes, and fig. 5 is a change curve of acceleration mean values of all axes extracted by the smart phone through a sliding window, which lasts 40s in five modes of telephone, hand-holding, video recording, pocket and swinging. It can be seen that the acceleration characteristics of three axes of the equipment are obviously different due to different spatial relations among axes of the equipment coordinate system. Therefore, the real-time average acceleration information of the user can be used for classifying the equipment posture, and the calculation formula is as follows:
Figure BDA0002518187700000111
in the formula (I), the compound is shown in the specification,
Figure BDA0002518187700000112
denotes the kthThe three-axis acceleration mean value of the sliding window, and WinSize is the size of the sliding window.
Step S202, mode monitoring and classification.
In actual use, a user often places the smart phone in a pocket or holds the smart phone in a hand to check and read, and user information is continuously extracted in real time and used for classification to judge the state of the user equipment, so that a large amount of memory is occupied, and the cruising ability is greatly reduced.
Therefore, the invention realizes the real-time monitoring and identification of the device posture by combining a Finite vector Machine (FSM) and a Decision Tree (DT) algorithm.
In the FSM algorithm, the states of the device are totally divided into 6 types, including 5 main modes (Message, video, calling, swing, pocket) and 1 transition mode (TRANS) bridging the main modes, the start mode is generally a short Message mode (assuming that navigation is always turned on through a handheld interaction mode), when the switching of the device mode is not monitored, the device is always in the current main state, otherwise, the device is switched to the TRANS state.
The device is switched in different modes, and the angular velocity characteristics may have similarity, and as shown in fig. 6, the device is switched from a video mode to a swing mode (130 s, 250 s) and a pocket mode (220 s). The simple utilization of rough mode information is difficult to meet the actual requirement, and the system has weak tolerance. Therefore, when the system enters TRANS, the algorithm automatically extracts the acceleration information of the next sliding window, and the decision tree algorithm is used to obtain the accurate Mode information (DT Mode) after switching, during which the state of the user is temporarily considered as RoughMode, and the algorithm flow is shown in fig. 7.
And S3, gait detection.
Since the acceleration has a sine wave characteristic when the human body is walking, the pedestrian steps can be detected by detecting the wave crests or the wave troughs of the sine wave. In order to avoid the influence of equipment posture switching on gait detection, a multi-posture multi-condition constraint-based wave crest-trough self-adaptive detection method is used for acquiring the gait information of a user. As shown in fig. 8, the systemIntegrating the current mode information, automatically selecting the acceleration information of the vertical coordinate axis (figure 4) of the equipment, and setting different peak value threshold values lambda θ Time threshold of adjacent peak Δ t k And the adjacent peak selection mechanism discriminates the pseudo wave crest/pseudo wave trough under a plurality of conditions to realize gait adaptive identification under multiple postures.
And S4, estimating the course.
The accuracy of the heading estimation will directly affect the final positioning accuracy. In practical use, the relative relationship between the equipment and the pedestrian is often changeable, and the practical requirement is difficult to adapt by replacing the direction of the pedestrian with the heading of the equipment or introducing heading compensation.
The invention provides an improved course calculation method, which is characterized in that the global course of a pedestrian linear stage is calculated by introducing gyroscope information, and then a fused course with strong robustness is obtained by utilizing a particle filter algorithm based on map matching.
Step S401, step-by-step heading.
Step S40101, AHRS (Attitude and Heading Reference System) is a Attitude Reference System with strong robustness and high precision provided for PDR by fusing measured values of three-axis accelerometer, three-axis gyroscope and three-axis magnetometer, the invention adopts a Madgwick-AHRS algorithm to obtain the Attitude of intelligent equipment held by a user in the walking process in real time, and utilizes the equipment Attitude to calculate a conversion matrix from a carrier coordinate System to a Reference coordinate System
Figure BDA0002518187700000131
Obtaining the horizontal acceleration of the user under a reference coordinate system, wherein the formula is as follows:
Figure BDA0002518187700000132
in the formula
Figure BDA0002518187700000133
Indicating the acceleration information of the user i at the moment in the coordinate system of the device,
Figure BDA0002518187700000134
indicating acceleration information of the user i in the reference coordinate system at the moment. The heading angle can pass
Figure BDA0002518187700000135
Calculated, the formula is as follows:
Figure BDA0002518187700000136
step S40102, in fact, due to poor quality of the mobile phone sensor, limited calculation capability and shaking of the body during walking, it is difficult to obtain an accurate heading from equation (3), so that the heading information in one step can be obtained by equation (5) using the gait information of the pedestrian obtained by the gait detection algorithm:
Figure BDA0002518187700000141
Figure BDA0002518187700000142
in the formula (I), the compound is shown in the specification,
Figure BDA0002518187700000143
representing the stepping course of the pedestrian in the jth step;
Figure BDA0002518187700000144
a velocity vector obtained by integrating the horizontal acceleration; step j And Step j+1 Respectively representing the time instants of the j and j +1 steps acquired by the gait detection algorithm.
Step S402, global heading.
Due to the limitation of buildings in the indoor environment, the randomness of the pedestrian track is greatly reduced and always shows a certain rule. The gyroscope data represents the variation of the angular velocity of the device along a certain direction, so that the angular velocity information of a specific axis of the smart phone in different postures can be utilized to judge the motion characteristics of the user, such as turning or straight going. Fig. 9 shows the three-axis characteristics of the gyroscope in the sms mode, and the Z-axis shows the variation of the angular velocity of the user in the direction of travel, where four singular points appear, representing four turns of the user during movement.
The method has better robustness in abnormal detection of the quartile, obtains the turning information of the pedestrian by using an abnormal value detection algorithm of the quartile, and divides the user track into a plurality of linear stages (the user is considered to walk along a straight line approximately). The real-time course of the user can be obtained by utilizing the integral of the horizontal acceleration in the linear stage, so that the course of the user is continuously corrected, and the formula is as follows:
Figure BDA0002518187700000151
in the formula (I), the compound is shown in the specification,
Figure BDA0002518187700000152
indicating the global heading, step, of Step j k Indicating the time of the last turn by the user.
Step S403, a course estimation algorithm assisted by particle filtering and map matching.
In step S40301, the particle filter approximates the probability density function by finding a set of samples (called particles) in the state space to obtain a minimum variance estimate of the state, which is expressed by the following formula:
Figure BDA0002518187700000153
Figure BDA0002518187700000154
in the formula, p (x) l |z 1:l ) The probability of a posterior is represented by,
Figure BDA0002518187700000155
is a dirac function;
Figure BDA0002518187700000156
representing the weight corresponding to the ith particle; x is the number of l Representing an observed value, namely a coordinate value calculated by using the global course;
Figure BDA0002518187700000157
representing the prior estimate of the ith particle at time i. This can be obtained by converting equation 9:
Figure BDA0002518187700000158
in the formula, N i (l) And E i (l) A priori coordinate values representing a north direction and an east direction;
Figure BDA0002518187700000159
and
Figure BDA00025181877000001510
representing north and east velocity values; delta t is a two-step time interval obtained by a gait detection algorithm; a is N (l) And a E (l) Representing north and east acceleration velocity values, following a normal distribution N — (0, σ) 2 )。
In step S40302, the particle filter algorithm is advantageous in solving the non-linear and noise non-gaussian problems, but it has problems such as improper particle transfer, for example, the particles are transferred to an unreachable area or pass through a wall to another indoor area. Building maps typically contain a large amount of useful information, such as the position of doors, walls or corners. This information can be used not only to constrain the transfer of particles, but also to effectively reduce the number of invalid particles, thereby providing more reliable position information. As shown in fig. 10a, when a particle is transferred to an unreachable area, its weight is set to 0. Instead, the right side of the bell-shaped curve of the gaussian distribution can be used to calculate the weight of the particle, resulting in a more realistic trajectory, which is formulated as:
Figure BDA0002518187700000161
step S40303, for the problem that the trajectory easily appears at the corner passes through the wall, the nearest landmark information (e.g., point P shown in fig. 10 b) is sensed by using the rotation angle time provided by the gyroscope, and the correction value of each step is obtained based on the trimming of the trajectory, so as to correct the trajectory and prevent the trajectory from passing through the impassable area.
Step S40304, after the particle weight is determined, normalization processing is also required, and the formula is:
Figure BDA0002518187700000162
Figure BDA0002518187700000163
step S40305, the particle screening uses a random resampling method, high-weight particles are kept as much as possible, and the removal of the current position of the low system can be updated as follows:
Figure BDA0002518187700000164
in the formula,. DELTA.E global And Δ N global And representing the displacement values of the pedestrian with the global heading calculation in the east and north directions, so that the main advancing direction of the pedestrian can be judged.
Step S40306, in which the course angle of the step can be calculated according to the probabilistic location determined by filtering:
Figure BDA0002518187700000171
step S40307, the course angle determined by the method can be matched with the real pedestrian direction in most positions, except for the existence of large fluctuations in corners and few points. Although the global course has accuracy lower than the course angle calculated by the particle filter based on map matching, the global course has better stability, so the calculation method for fusing courses can be obtained by the following formula:
Figure BDA0002518187700000172
in the formula, theta global And theta stepwise Representing the global and step headings mentioned in the previous section; rho 1 And ρ 2 The representative weight value can be adaptively adjusted according to the actual situation so as to improve the heading estimation capability of the model. Delta represents the threshold value of course deviation determined by particle filtering, and is set to be 15 degrees; d is a radical of 1 And d 2 Represents theta particle And theta global And theta stepwise The difference of (a).
The overall flow of the method is shown in figure 1, and the overall structure of the PDR system consists of data preprocessing, gait detection, mode identification and course estimation. The low-cost inertial sensor built in the smart phone is influenced by drift caused by characteristics of the low-cost inertial sensor and environmental interference. Therefore, the resulting raw signal needs to be filtered and corrected before use. The characteristic data reflecting the behaviors are extracted from the inertial sensor and used as the input of the classifier, and the finite state machine FSM and the decision tree DT algorithm are combined to carry out real-time monitoring and identification on the mobile phone mode. According to the method, after FSM algorithm detects the change of the attitude of the equipment, DT algorithm is used for identifying the current mode of the user. And finally, adjusting parameters of gait detection and corner detection according to the result of the classifier, and updating the position of the PDR through a particle filter algorithm.
Those skilled in the art will appreciate that all or part of the processes for implementing the methods of the embodiments described above can be implemented by a computer program, which can be stored in a computer-readable storage medium, for instructing the relevant hardware. The computer readable storage medium is a magnetic disk, an optical disk, a read-only memory or a random access memory, etc.
The above description is only a preferred embodiment of the present invention, and the scope of the present invention is not limited to the above embodiment, but equivalent modifications or changes made by those skilled in the art according to the present disclosure should be included in the scope of the present invention as set forth in the appended claims.

Claims (5)

1. A multi-pose downlink pedestrian dead reckoning method based on map information assistance is characterized by comprising the following steps of:
step 1, carrying out low-pass filtering and smoothing pretreatment on an acceleration original signal acquired by a low-cost sensor embedded in a smart phone to weaken the adverse effect of noise and jitter of the sensor on mode identification and positioning; correcting a magnetometer of the smart phone in an indoor environment, and reducing the influence of internal errors and external interference;
step 2, recognizing the use mode of the mobile phone in a typical scene through a lightweight equipment posture monitoring and recognizing algorithm which is mainly divided into a feature extraction module and a mode monitoring and classifying module;
in step 2, the feature extraction module uses a sliding window to perform data slicing, extracts data features of each window, sets the window size to be 128 samples, and has an overlap degree of 50%;
the classification of the equipment attitude is realized through real-time average acceleration information of the pedestrian, and the calculation formula is as follows:
Figure FDA0003866763740000011
in the formula (I), the compound is shown in the specification,
Figure FDA0003866763740000012
representing the three-axis acceleration mean value of the kth sliding window, wherein WinSize is the size of the sliding window;
in step 2, the mode monitoring and classifying module realizes real-time monitoring and identification of the equipment attitude by combining a finite vector machine (FSM) and a Decision Tree (DT) algorithm, and identifies the current mode of the pedestrian by using the DT algorithm only when the FSM algorithm detects that the equipment attitude changes;
step 3, discriminating a pseudo wave peak/pseudo wave valley based on a wave crest detection algorithm by utilizing the preprocessed acceleration signal and the equipment use attitude output by the equipment attitude monitoring and recognition algorithm, so as to realize the gait recognition of the PDR algorithm;
step 4, calculating the global course of the pedestrian linear stage by introducing gyroscope information, obtaining a fusion course with strong robustness by assisting a particle filter algorithm based on map matching, and correcting the problem of track through-the-wall by perceiving the behavior characteristics landmark in the movement of the pedestrian;
the step 4 comprises two steps of calculating a global course and estimating the course assisted by particle filtering and map matching;
the global course is calculated, and specifically, the posture of the intelligent equipment held by the pedestrian in the walking process is obtained in real time by adopting a Madgwick-AHRS algorithm; calculating a conversion matrix from the carrier coordinate system to the reference coordinate system by using the equipment attitude to obtain the horizontal acceleration of the pedestrian under the reference coordinate system:
Figure FDA0003866763740000021
in the formula
Figure FDA0003866763740000022
Indicating the acceleration information of the user i at the moment in the coordinate system of the device,
Figure FDA0003866763740000023
representing acceleration information of the user i under a reference coordinate system at the moment;
and calculating the course information in one step according to the pedestrian gait information obtained by the gait detection algorithm:
Figure FDA0003866763740000024
Figure FDA0003866763740000025
in the formula (I), the compound is shown in the specification,
Figure FDA0003866763740000026
representing the stepping course of the pedestrian in the jth step;
Figure FDA0003866763740000027
a velocity vector obtained by integrating the horizontal acceleration; step j And Step j+1 Respectively representing the time of the j step and the j +1 step acquired by a gait detection algorithm;
judging the motion characteristics of the pedestrian by the angular speed information of a specific axis of the smart phone under different postures; the method comprises the following steps of obtaining turning information of a pedestrian by utilizing a quartile abnormal value detection algorithm, dividing a pedestrian track into a plurality of linear stages, obtaining a real-time global heading of the pedestrian by utilizing the integral of horizontal acceleration of the linear stages, and realizing continuous correction of the heading of the pedestrian:
Figure FDA0003866763740000031
in the formula (I), the compound is shown in the specification,
Figure FDA0003866763740000032
indicating the global heading, step, of Step j k Indicating the time of the last turn of the user;
the heading estimation assisted by particle filtering and map matching is specifically characterized in that the particle filtering approximates a probability density function by searching a group of samples, namely particles, in a state space to obtain minimum variance estimation of a state, and north direction and east direction acceleration values are obtained through the north direction and east direction acceleration values;
according to boundary information contained in the map, restricting the transfer of the particles and reducing the number of invalid particles, and when the particles are transferred to an unreachable area, assigning the weight value of the particles to be 0; otherwise, the weight of the particle is calculated with the right side of the bell-shaped curve of the gaussian distribution:
Figure FDA0003866763740000033
in the formula, N (l) and E (l) represent coordinate values of north and east directions obtained by observation equation at time l, and N i (l) And E i (l) A priori coordinate values representing the north and east directions of the ith particle;
the particle weight calculation also includes resampling of the particles: the particle resampling adopts a random resampling method, high-weight particles are kept as far as possible, and low-weight particles are removed;
to minimize trajectory fluctuations, the update of the pedestrian's position can be calculated by:
Figure FDA0003866763740000041
in the formula,. DELTA.E global And Δ N global The displacement of the user in the east and north directions calculated for the global heading;
and step 5, carrying out pedestrian dead reckoning through gait detection, mobile phone use mode identification and course estimation to finally obtain pedestrian position information with high reliability.
2. The method of claim 1, wherein in step 1, the preprocessing of the data collected by the accelerometer and the magnetometer comprises:
eliminating the influence of high-frequency noise by using a fourth-order Butterworth low-pass filter with the cut-off frequency of 8Hz, smoothing data by adopting a moving average algorithm and removing unnecessary burrs;
the smart phone is rotated around three axes of a mobile phone coordinate system in a space under an indoor environment, and the obtained magnetometer sample data is corrected to magnetometer signals by adopting a least square fitting ellipsoid method, so that the influence of internal errors and external interference of the sensor is reduced.
3. The map information-assisted multi-pose downlink pedestrian dead reckoning method according to claim 1, wherein: in the FSM algorithm of step 2, the states of the device are divided into 6 types in total, including 5 typical main modes: message, video, calling, swing, pocket and 1 transition mode TRANS bridging the main mode, the initial mode is the Message mode, when not monitoring the switching of the device mode, the device is always in the current main state, otherwise, the device is switched to the TRANS state; when the system enters a TRANS Mode, the algorithm automatically extracts the acceleration information of the next sliding window, and the accurate Mode information DT Mode after switching is obtained by utilizing the decision tree algorithm, and the pedestrian state is temporarily considered as the Rough Mode in the period.
4. The multi-pose downlink pedestrian dead reckoning method based on map information assistance according to claim 1, characterized in that in step 3, the preprocessed signals and the device use pose output by a pattern recognition algorithm are utilized, the acceleration signals preprocessed by the device vertical axis are extracted for gait detection, a peak threshold, an adjacent peak time threshold and an adjacent peak selection mechanism are adjusted according to different poses, and pseudo wave peaks/pseudo wave troughs are screened by utilizing multiple conditions to realize gait adaptive recognition.
5. The map information-assisted multi-pose downlink pedestrian dead reckoning method according to claim 1, wherein: step 4, sensing the nearest landmark information by using the turning angle time provided by the gyroscope, and obtaining a correction value of each step by taking the trimming of the track as a basis, so as to correct the track and avoid the track from passing through an impassable area;
and according to the probability position determined by filtering, reversely calculating a filtered course angle:
Figure FDA0003866763740000051
the robustness of the course is further improved, and the fused course is calculated by the following formula:
Figure FDA0003866763740000052
in the formula, theta global And theta stepwise Representing the global and step headings mentioned in the previous section; rho 1 And ρ 2 The representative weight is adaptively adjusted according to the actual situation so as to improve the heading estimation capability of the model; delta represents a threshold value of course deviation determined by particle filtering and is set to be 15 degrees; d is a radical of 1 And d 2 Represents theta particle And theta global And theta stepwise The difference of (c).
CN202010483376.8A 2020-06-01 2020-06-01 Multi-posture downlink pedestrian dead reckoning method based on map information assistance Active CN111595344B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202010483376.8A CN111595344B (en) 2020-06-01 2020-06-01 Multi-posture downlink pedestrian dead reckoning method based on map information assistance

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202010483376.8A CN111595344B (en) 2020-06-01 2020-06-01 Multi-posture downlink pedestrian dead reckoning method based on map information assistance

Publications (2)

Publication Number Publication Date
CN111595344A CN111595344A (en) 2020-08-28
CN111595344B true CN111595344B (en) 2023-02-24

Family

ID=72189829

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202010483376.8A Active CN111595344B (en) 2020-06-01 2020-06-01 Multi-posture downlink pedestrian dead reckoning method based on map information assistance

Country Status (1)

Country Link
CN (1) CN111595344B (en)

Families Citing this family (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN112819916B (en) * 2020-12-29 2022-08-19 同济大学 Method for drawing probability peak-valley graph of stay time of workers in workshop and application of method
CN113218357A (en) * 2021-05-19 2021-08-06 广西电网有限责任公司电力科学研究院 On-line monitoring system of high-voltage isolating switch
CN114577210B (en) * 2022-02-24 2024-07-09 哈尔滨工程大学 Cross-region detection algorithm based on map information matrix
CN115420288B (en) * 2022-07-19 2024-05-17 北京航空航天大学 Indoor positioning method for lightweight map information robust constraint PDR
CN115235455B (en) * 2022-09-19 2023-01-13 中国人民解放军国防科技大学 Pedestrian positioning method based on smart phone PDR and vision correction

Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN108225304A (en) * 2018-01-26 2018-06-29 青岛美吉海洋地理信息技术有限公司 Based on method for rapidly positioning and system in Multiple Source Sensor room
CN109164411A (en) * 2018-09-07 2019-01-08 中国矿业大学 A kind of personnel positioning method based on multi-data fusion
CN110487273A (en) * 2019-07-15 2019-11-22 电子科技大学 A kind of indoor pedestrian track projectional technique of level meter auxiliary
CN111189443A (en) * 2020-01-14 2020-05-22 电子科技大学 Pedestrian navigation method for online step length calibration, motion deviation angle correction and adaptive energy management

Family Cites Families (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US10508920B2 (en) * 2014-11-04 2019-12-17 Xsens Holding B.V. Pedestrian dead reckoning position tracker

Patent Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN108225304A (en) * 2018-01-26 2018-06-29 青岛美吉海洋地理信息技术有限公司 Based on method for rapidly positioning and system in Multiple Source Sensor room
CN109164411A (en) * 2018-09-07 2019-01-08 中国矿业大学 A kind of personnel positioning method based on multi-data fusion
CN110487273A (en) * 2019-07-15 2019-11-22 电子科技大学 A kind of indoor pedestrian track projectional technique of level meter auxiliary
CN111189443A (en) * 2020-01-14 2020-05-22 电子科技大学 Pedestrian navigation method for online step length calibration, motion deviation angle correction and adaptive energy management

Also Published As

Publication number Publication date
CN111595344A (en) 2020-08-28

Similar Documents

Publication Publication Date Title
CN111595344B (en) Multi-posture downlink pedestrian dead reckoning method based on map information assistance
CN107635204B (en) Indoor fusion positioning method and device assisted by exercise behaviors and storage medium
Pei et al. Optimal heading estimation based multidimensional particle filter for pedestrian indoor positioning
CN107396321B (en) Unsupervised indoor positioning method based on mobile phone sensor and iBeacon
CN108632761B (en) Indoor positioning method based on particle filter algorithm
EP3014476B1 (en) Using movement patterns to anticipate user expectations
WO2018068771A1 (en) Target tracking method and system, electronic device, and computer storage medium
JP6411917B2 (en) Self-position estimation apparatus and moving body
CN110579207B (en) Indoor positioning system and method based on combination of geomagnetic signals and computer vision
CN104180805A (en) Smart phone-based indoor pedestrian positioning and tracking method
CN105674984A (en) Method for acquiring advancing directions of pedestrians by aid of mobile terminals without constraints
CN108106613B (en) Positioning method and system based on visual assistance
Ren et al. Movement pattern recognition assisted map matching for pedestrian/wheelchair navigation
CN106606407A (en) High-precision indoor navigation system for blind person
CN112729301A (en) Indoor positioning method based on multi-source data fusion
CN109540143B (en) Pedestrian unconventional action direction identification method based on multi-sensing-source dynamic peak fusion
CN114593735B (en) Pose prediction method and device
CN112087728B (en) Method and device for acquiring Wi-Fi fingerprint spatial distribution and electronic equipment
Wu et al. CWIWD-IPS: A crowdsensing/walk-surveying inertial/Wi-Fi data-driven indoor positioning system
CN111435083A (en) Pedestrian track calculation method, navigation method and device, handheld terminal and medium
US10551195B2 (en) Portable device with improved sensor position change detection
CN115235455B (en) Pedestrian positioning method based on smart phone PDR and vision correction
Yao et al. A magnetic interference detection-based fusion heading estimation method for pedestrian dead reckoning positioning
CN116465412A (en) Improved PDR indoor positioning method based on LSTM and attention mechanism
CN114608560B (en) Passive combined indoor positioning system and method based on intelligent terminal sensor

Legal Events

Date Code Title Description
PB01 Publication
PB01 Publication
SE01 Entry into force of request for substantive examination
SE01 Entry into force of request for substantive examination
GR01 Patent grant
GR01 Patent grant