A pedestrian dead reckoning method based on human motion state recognition

Abstract

The present invention seeks to protect a pedestrian dead reckoning method based on pedestrian motion state recognition. In view of the fact that the traditional pedestrian dead reckoning (PDR) algorithm can only be used for a single state of normal walking and is difficult to meet the needs of actual applications, an improved PDR algorithm based on human motion state recognition is proposed. The algorithm divides the pedestrian motion mode into two states: walking and running. It fully considers the relationship between the peak acceleration and the motion state during the pedestrian motion process, obtains the peak acceleration under different motion states through experiments, and thus sets dynamic thresholds to achieve step detection and stride length estimation under different states. The improved PDR algorithm is applied to pedestrian positioning: the motion data of pedestrians while walking is obtained by the micro inertial measurement unit (IMU), and the improved peak detection method is used to perform step detection and state recognition on pedestrians. The step length is estimated using an adaptive step length estimation formula according to the pedestrian's motion state. Finally, the pedestrian's position information is obtained in combination with the calculated heading.

Description

Technical Field

[0001] The present invention belongs to the technical field of inertial navigation and relates to a pedestrian dead reckoning method based on human motion state recognition. Background Art

[0002] Pedestrian dead reckoning (PDR) uses sensor data from an IMU's accelerometer, gyroscope, and magnetometer to infer a pedestrian's real-time progress and direction, and then calculates the current position through accumulation, thereby achieving positioning. With the rapid development of micro-electro-mechanical systems (MEMS), IMUs, due to their low cost, small size, low energy consumption, and ease of integration, have gradually gained attention, and IMU-based pedestrian navigation has become a research hotspot. IMU-based positioning systems do not require advance information about unknown environments, nor do they require the deployment of other hardware equipment in the positioning area. They have important application value in emergency disaster relief efforts such as fires and earthquakes.

[0003] Currently, most research on improving pedestrian positioning and navigation accuracy focuses on reducing IMU measurement errors, with less research on the pedestrian's motion state. The PDR algorithm, based on a known initial position, uses real-time accelerometer and gyroscope data to determine whether the pedestrian has taken a step. The algorithm then calculates the pedestrian's step length and heading angle, accumulating these data to determine the pedestrian's current position. Conventional PDR algorithms perform poorly in pedestrian positioning for motion states such as running and a combination of walking and running. Building on existing research, this method proposes a pedestrian dead reckoning method based on human motion state recognition. By analyzing the corresponding acceleration peaks in walking and running states, different acceleration thresholds and time windows are set to improve the existing peak detection algorithm for step counting. Furthermore, an adaptive step length calculation formula is employed for different motion states to reduce the PDR algorithm's overall positioning error. Experimental results demonstrate that the proposed algorithm achieves good positioning for pedestrians in walking, running, and a combination of walking and running.

[0004] CN113239803A, a dead reckoning positioning method based on pedestrian motion state recognition, includes the steps of: building a pedestrian motion state recognition classification model, identifying pedestrian motion state, performing cadence detection, step length estimation, heading estimation and dead reckoning.

[0005] The method in this patent requires sufficient gait data for training the gait model. It also needs to process the time domain features of five types of motion states and establish an effective feature matrix. The algorithm is relatively complex and requires high hardware processing capabilities, making it difficult to popularize in practical applications. There are too many motion state classifications, which can easily lead to misjudgments and larger errors. The algorithm of the present invention does not require gait data to be obtained in advance. The motion state classification is based on acceleration thresholds and time window thresholds. It has low time and space complexity and minimal hardware overhead. Summary of the Invention

[0006] The present invention aims to solve the above problems in the prior art. It proposes a pedestrian dead reckoning method based on human motion state recognition. The technical solution of the present invention is as follows:

[0007] A pedestrian dead reckoning method based on human motion state recognition comprises the following steps:

[0008] Step a: Wear the IMU on the pedestrian's waist, obtain the accelerometer and gyroscope data in real time during the pedestrian's movement, and pre-process the data;

[0009] Step b: using the collected motion data, using an adaptive peak detection algorithm to classify the motion state of the pedestrian and perform step detection, and using a heading calculation model to calculate the heading;

[0010] Step c: using adaptive step length estimation to fit the actual motion step length of the pedestrian according to the different motion states of the pedestrian;

[0011] In step d, the position information of the pedestrian during movement is obtained in real time based on step counting detection and step length estimation.

[0012] Furthermore, the data preprocessing in step a specifically includes:

[0013] The collected three-axis acceleration data are combined according to the following formula to obtain the combined acceleration modulus:

[0014]

[0015] Where a x 、a y 、a z is the collected three-axis acceleration value;

[0016] For the preprocessing of the original data, various noises are eliminated and the features of the original data are retained as much as possible; the moving mean filtering method is used to preprocess the combined acceleration data with a window size of 5.

[0017] Furthermore, the step b, using the collected motion data, uses an adaptive peak detection algorithm to classify the motion state of the pedestrian and detect the steps, and uses a heading calculation model to calculate the heading, specifically includes:

[0018] The peak detection method uses the characteristic that the acceleration signal of a person walking presents a sine wave-like characteristic. When a peak is detected, it is considered that the pedestrian has taken a step, and the peak time is recorded, as shown in the following formula:

[0019]

[0020] Where S represents the set of moments marked as acceleration peaks, and Respectively represent t peak-k and t peak+k The magnitude of the acceleration at the moment, where k is the conditional parameter of the peak;

[0021] The peak detection method sets a threshold. Only when the peak acceleration value exceeds the threshold will it be considered that the pedestrian has taken a step; as shown in the following formula:

[0022]

[0023] The pedestrian's motion state is divided into three types: walking, stationary and running. An adaptive peak detection algorithm is proposed based on the different acceleration peaks.

[0024] The heading calculation in PDR mainly relies on the heading dead reckoning model, namely

[0025]

[0026] in, and The headings at time t-1 and t, and the initial heading Initial calibration is performed using the information from the magnetometer; ΔT is the interval between gyroscope data; ∑ω is the sum of the angular velocities output by the gyroscope from time t-1 to time t.

[0027] Furthermore, in step c, according to the different motion states of the pedestrian, an adaptive step size is used to estimate the actual motion step size of the pedestrian, specifically including:

[0028] The step length is estimated by fusing linear and nonlinear step length models. For walking, the linear step length model of formula (5) is used to estimate the step length. For running, the Weinberg nonlinear step length model determined by the maximum and minimum acceleration values ​​shown in formula (8) is used.

[0029] S k =A+B×LF k +C×LV k (5)

[0030] In the formula, parameters A, B, and C are constants, and LF k and LV k are the step frequency and acceleration variance of the kth step, respectively, and the calculation formula is as follows:

[0031]

[0032]

[0033] Among them, t k and t k+1 are the starting and ending times of step k, respectively. t is the acceleration at time t, is the average acceleration of the kth step, N k Indicates the number of acceleration samples included in the k-th step;

[0034]

[0035] Where K is a constant, a max and a min Indicates the maximum and minimum values ​​of single-step acceleration;

[0036] Combining Equations (5) and (8), an adaptive step length estimation algorithm is proposed to meet the requirements of accurate step counting in both normal walking and running walking modes, as shown below:

[0037]

[0038] The advantages and beneficial effects of the present invention are as follows:

[0039] The present invention discloses a pedestrian dead reckoning method based on human motion state recognition, the innovations, advantages and beneficial effects of which are:

[0040] 1. It is proposed to preprocess the original data to eliminate the interference of noise and the pseudo peaks that appear near the peaks of the original acceleration values, thereby avoiding misjudgment and improving the accuracy of step counting detection.

[0041] 2. By analyzing the corresponding acceleration peaks of pedestrians walking and running, different acceleration thresholds and time windows are set to improve the existing peak detection algorithm for step counting. Furthermore, an adaptive step length calculation formula is used for different motion states to reduce the overall positioning error of the PDR algorithm. Experimental results show that this method achieves good positioning results in pedestrians walking, running, and walking and running combined.

[0042] The innovations of the present invention mainly include: the data preprocessing in step a uses moving mean filtering to eliminate the interference of noise and the pseudo-peaks that appear near the peak of the original acceleration value, thereby avoiding misjudgment and improving the accuracy of step counting detection; the adaptive peak detection algorithm in step b performs threshold judgment on the maximum acceleration generated by each step to achieve the classification of motion state. The threshold is set according to the parameters corresponding to the walking state and the running state, and the neighborhood window is changed to achieve adaptive judgment and accurate step counting in different motion states. The algorithm has good adaptability to the sensor posture and the pedestrian's motion state, and achieves a step counting accuracy rate of more than 99% under various sensor postures, which makes up for the deficiency of the conventional peak detection algorithm that cannot perform abnormal step counting; the adaptive step length estimation model in step c adjusts the estimated value of each step in real time according to the different motion states of the pedestrian, thereby improving the accuracy of positioning. BRIEF DESCRIPTION OF THE DRAWINGS

[0043] Figure 1 This is a comparison diagram of acceleration data before and after filtering according to a preferred embodiment of the present invention;

[0044] Figure 2 This is the PDR algorithm positioning principle diagram;

[0045] Figure 3 This is a graph of acceleration changes during walking and running;

[0046] Figure 4 It is a flowchart of the adaptive peak detection algorithm of the present invention;

[0047] Figure 5 This is a flowchart of a pedestrian dead reckoning method based on human motion state recognition according to the present invention;

[0048] Figure 6 It is an experimental roadmap for validating the algorithm;

[0049] Figure 7 These are the walking trajectory reproduction results under different algorithms. DETAILED DESCRIPTION

[0050] The following will describe the technical solutions in the embodiments of the present invention in detail with reference to the accompanying drawings. The described embodiments are only a part of the embodiments of the present invention.

[0051] The technical solution of the present invention to solve the above technical problems is:

[0052] Step 1: Wear the IMU module on the pedestrian's waist, obtain the accelerometer and gyroscope data in real time during the pedestrian's movement, and pre-process the data. The comparison effect before and after acceleration filtering is shown in the attached figure. Figure 1 shown.

[0053] The principle diagram of PDR algorithm positioning is as shown in the attached figure. Figure 2 As shown in (10), let the pedestrian’s starting position at the initial time t0 be (x0, y0). When moving to the next time t1, the pedestrian’s step length is d1, the heading angle is θ1, and the pedestrian’s coordinate information at time t1 is (x1, y1). Then the relationship between (x0, y0) and (x1, y1) is shown in (10).

[0054]

[0055] From the above formula, we can see that after multiple moments of accumulation, t k The coordinate information of the pedestrian at the moment (x k ,y k ) satisfies the formula:

[0056]

[0057] Step 2: Adopt the adaptive peak detection algorithm to classify the pedestrian's motion state and detect the steps. Figure 3 The following is a graph showing the acceleration changes during walking and running. As can be seen from the figure, the acceleration changes vary greatly under different motion states. Based on the acceleration characteristics of walking and running, an adaptive peak detection algorithm is used to identify the motion state of pedestrians and perform step counting. The process of the adaptive peak detection algorithm is shown in the attached figure. Figure 4 As shown in the figure, the data is first filtered to calculate the current state acceleration peak, also known as the potential peak. The current peak is then determined to be greater than 1.5g, thereby determining the pedestrian's current motion state. The time difference between the potential peak and the previous peak is then calculated to determine whether the time difference satisfies the corresponding cadence time window. Finally, the potential peak is determined to be the maximum peak, and spurious peaks are removed. If the potential peak is larger than all other peaks in the neighborhood, the pedestrian is considered to have taken a step.

[0058] Step 3: Estimate step length using a fusion of linear and nonlinear step length models, i.e., adaptive step length estimation. For walking, the acceleration peak value varies little and the stride frequency is low, so the linear step length model of Equation (5) is suitable for estimating step length. For running, the acceleration peak value is large and the stride frequency is high, so the Weinberg nonlinear step length model, determined by the maximum and minimum acceleration values, as shown in Equation (8), is suitable.

[0059] Step 4: The traditional PDR algorithm is only applicable to the single motion state of normal walking. It has poor positioning effect on pedestrians in states such as running, and the navigation route is quite different from the actual trajectory. Therefore, combining the adaptive peak detection and adaptive step length estimation proposed above, a pedestrian dead reckoning method based on pedestrian motion state recognition is proposed and applied to IMU to achieve pedestrian positioning. The system framework is shown in the attached figure. Figure 5 shown.

[0060] The pedestrian dead reckoning method based on human motion state recognition and the traditional PDR algorithm are applied to navigation equipment respectively. Figure 6 The experimental results of different positioning methods are shown in the attached figure. Figure 7 The trajectory errors of the two algorithms are shown in Table 1. It can be seen intuitively that the trajectory positioning error obtained by the improved PDR algorithm is smaller, and the algorithm improves the positioning accuracy.

[0061] Table 4 Comparison of errors of different algorithms

[0062]

[0063] The systems, devices, modules, or units described in the above embodiments may be implemented by computer chips or entities, or by products having certain functions. A typical implementation device is a computer. Specifically, the computer may be, for example, a personal computer, a laptop computer, a cellular phone, a camera phone, a smartphone, a personal digital assistant, a media player, a navigation device, an email device, a game console, a tablet computer, a wearable device, or a combination of any of these devices.

[0064] It should also be noted that the terms "comprises," "includes," or any other variations thereof are intended to encompass non-exclusive inclusion, such that a process, method, commodity, or apparatus that includes a series of elements includes not only those elements but also other elements not explicitly listed, or includes elements inherent to such process, method, commodity, or apparatus. In the absence of further limitations, an element defined by the phrase "comprises a ..." does not exclude the presence of other identical elements in the process, method, commodity, or apparatus that includes the element.

[0065] The above embodiments should be understood as merely illustrating the present invention and not as limiting the scope of protection of the present invention. After reading the contents of the present invention, technicians may make various changes or modifications to the present invention, and these equivalent changes and modifications also fall within the scope defined by the claims of the present invention.

Claims

1. A pedestrian dead reckoning method based on human motion state recognition, characterized in that: The following steps are involved: Step a: Wear the IMU on the pedestrian's waist, obtain the accelerometer and gyroscope data in real time during the pedestrian's movement, and pre-process the data; Step b: using the collected motion data, using an adaptive peak detection algorithm to classify the motion state of the pedestrian and perform step detection, and using a heading calculation model to calculate the heading; Step c: using adaptive step length estimation to fit the actual motion step length of the pedestrian according to the different motion states of the pedestrian; Step d: obtaining the position information of the pedestrian in real time during movement based on step counting detection and step length estimation; The step b, using the collected motion data, uses an adaptive peak detection algorithm to classify the motion state of the pedestrian and perform step detection, and uses a heading calculation model to calculate the heading, specifically includes: The peak detection method uses the characteristic that the acceleration signal presents a sine wave when the human body is walking. When a peak is detected, it is considered that the pedestrian has taken a step, and the peak time is recorded, as shown in the following formula: Where S represents the set of moments marked as acceleration peaks, and Respectively represent t peak-k and t peak+k The magnitude of the acceleration at the moment, where k is the conditional parameter of the peak; The peak detection method sets a threshold. Only when the peak acceleration value exceeds the threshold will it be considered that the pedestrian has taken a step; as shown in the following formula: The pedestrian's motion state is divided into three types: walking, stationary and running. An adaptive peak detection algorithm is proposed based on the different acceleration peaks. The heading calculation in PDR relies on the heading dead reckoning model, namely in, and The headings at time t-1 and t, and the initial heading Initial calibration is performed using the information from the magnetometer; ΔT is the interval between gyroscope data; ∑ω is the sum of the angular velocities output by the gyroscope from time t-1 to time t; In step c, according to the different motion states of the pedestrian, the adaptive step length is used to estimate the actual motion step length of the pedestrian, specifically including: The step length is estimated by fusing linear and nonlinear step length models. For walking, the linear step length model of formula (4) is used to estimate the step length. For running, the Weinberg nonlinear step length model determined by the maximum and minimum acceleration values ​​shown in formula (7) is used. S k =A+B×LF k +C×LV k (4) In the formula, parameters A, B, and C are constants, and LF k and LV k are the step frequency and acceleration variance of the kth step, respectively, and the calculation formula is as follows: Among them, t k and t k+1 are the starting and ending times of step k, respectively. t is the acceleration at time t, is the average acceleration of the kth step, N k Indicates the number of acceleration samples included in the k-th step; Where K is a constant, a max and a min Indicates the maximum and minimum values ​​of single-step acceleration; Combining Equation (4) and Equation (7), an adaptive step length estimation algorithm is proposed to meet the requirements of accurate step counting in both normal walking and running walking modes, as shown below:

2. The pedestrian dead reckoning method based on human motion state recognition according to claim 1, characterized in that: The data preprocessing in step a specifically includes: The collected three-axis acceleration data are combined according to the following formula to obtain the combined acceleration modulus: Where a x 、a y 、a z is the collected three-axis acceleration value; Preprocess the original data to eliminate various noises and retain as many features of the original data as possible; The moving mean filtering method is used to preprocess the combined acceleration data with a window size of 5.

Images