An adaptive and accurate step counting method based on low-cost accelerometer

Abstract

The invention discloses an adaptive precise step counting method based on a low-cost accelerometer, which belongs to the technical field of step detection, and comprises the following steps: step 1, data preprocessing: comprising acceleration modulus calculation, gravity correction, data smoothing processing through a Butterworth low-pass filter, and initialization of a zero acceleration threshold and two length thresholds; step 2, adaptive zero acceleration threshold selection: using a preset zero acceleration threshold to perform a first zero-speed detection, then comparing the state interval length with the first length threshold, and then adaptively adjusting the zero acceleration threshold; step 3, state vector updating: using the adjusted zero acceleration threshold to perform a second zero-speed detection on a pedestrian acceleration sequence, comparing the state interval length with the second length threshold, and updating the state vector; step 4, counting the number of motion intervals and static intervals, and taking the counted numbers of the two intervals as the number of pedestrian steps.

Description

Technical Field

[0001] The present invention belongs to the technical field of step detection, and in particular to an adaptive and accurate step counting method based on a low-cost accelerometer. Background Art

[0002] Nowadays, people have an increasing demand for location services. In open areas, the Global Navigation Satellite System (GNSS) can meet people's needs, but it cannot provide accurate real-time positioning services in GNSS-denied environments. Pedestrian navigation systems (PNS) play a vital role in ensuring the safety of workers in extreme environments, especially in high-risk situations such as fires and emergency rescue. Among them, pedestrian inertial navigation (PDR) based on inertial measurement units (IMUs) stands out. Because it is not affected by prior information such as signal base stations or fingerprint databases, it has the advantages of independence, immediacy, high positioning accuracy, applicability and real-time performance, and has attracted more and more attention. PDR includes three parts: step detection, step length estimation and heading estimation. It realizes position estimation by identifying the movement process of pedestrians and estimating step length and direction. Among them, step detection determines whether PDR performs position estimation. It is the decision maker of PDR and plays a very important role. In the research on step detection, the mainstream is to obtain pedestrian gait information by placing accelerometers on the human body and using the acceleration performance of pedestrians during movement. The common sources of accelerometer carriers can be roughly divided into two categories: smartphones and wearable devices. Smartphones focus on the use of mobile phones by users, while wearable devices focus more on the need for free hands. Among them, smartphones are multifunctional mobile devices that integrate social, entertainment, payment, health, etc. and are indispensable items in life. The built-in sensors of smartphones are used to collect user walking data, such as number of steps, step length, step frequency, direction change and other information. Because the position of the device is very flexible, it is difficult to detect pedestrian gait using smartphones. Among them, the carrying position of the mobile phone is only one of the factors affecting the step counting. The complex and changeable walking patterns of pedestrians will also increase the difficulty of step counting. In the existing methods, SVM and FSM-DT classifiers are constructed to identify the user's motion state and mobile phone usage mode, and appropriate parameters are initialized for different situations. A high accuracy rate is achieved in normal walking and free walking, but the detection effect is not good in other activities. Wearable devices are often used on the feet, waist, arms, head, etc., and are bringing benefits to humans in various ways, such as convenient payment and health care health testing services. Among them, the gait information of pedestrians is an important indicator, which is usually captured by the accelerometer in the MEMS IMU embedded in the device. Many step detection algorithms have also emerged in this field, mainly including step cycle recognition method, autocorrelation analysis method, peak detection method, zero crossing detection method and machine learning algorithm. Although they can avoid the system's dependence on a predetermined threshold, they are easily affected by body shaking or the environment, and their applicability and robustness are relatively weak. Therefore, the present invention discloses an adaptive and precise step counting method based on a low-cost accelerometer, which achieves precise step counting and improves overall applicability and robustness. Summary of the invention

[0003] The purpose of the present invention is to provide an adaptive and precise pedometer based on a low-cost accelerometer, so as to solve the problems of low pedometer accuracy, poor applicability and robustness in the prior art.

[0004] To achieve the above object, the present invention provides an adaptive and accurate step counting method based on a low-cost accelerometer, comprising the following steps:

[0005] Step 1: Data preprocessing: including acceleration modulus calculation, gravity correction, data smoothing through Butterworth low-pass filter, initialization of zero acceleration threshold and two length thresholds;

[0006] Step 2, adaptive zero acceleration threshold selection: Use the preset zero acceleration threshold to perform zero speed detection, the zero speed detection result is the state value 0 and 1, 0 represents motion, 1 represents stillness; then compare the state interval length LSI with the first length threshold, and then adaptively adjust the zero acceleration threshold; specifically: if the length of the still interval exceeds the first length threshold, it is considered that the zero acceleration threshold is too large, and the threshold is adjusted downward; similarly, if the length of the motion interval exceeds the first length threshold, it is considered that the zero acceleration threshold is too small, and it is adjusted upward; in all other cases, the threshold remains unchanged;

[0007] Step 3, state vector update: Use the adjusted zero acceleration threshold to perform a second zero-speed detection on the pedestrian acceleration sequence, compare the state interval length with the second length threshold to eliminate some minor misjudgments, and update the state vector;

[0008] Step 4: Count the number of moving intervals and static intervals, and use the number of the two intervals counted as the number of pedestrian steps, where the static interval and moving interval after the state vector is updated are a group of continuous 1s and 0s, respectively.

[0009] Preferably, the expression for calculating the acceleration modulus is as follows:

[0010]

[0011] In the formula, , , Represent the acceleration components of the X, Y, and Z axes respectively, It is the magnitude of acceleration, which indicates the instantaneous acceleration of the object measured by the IMU.

[0012] Preferably, the calculation expression of gravity correction is as follows:

[0013]

[0014] In the formula, is the sampling frequency, For t The acceleration modulus after separation of gravity acceleration, Indicates time.

[0015] Preferably, the process of performing data smoothing by using a Butterworth low-pass filter is as follows:

[0016] The cutoff frequency is normalized and the calculation expression is as follows:

[0017]

[0018] The order of the filter To calculate, the expression is as follows:

[0019]

[0020] Calculate the output signal after Butterworth low-pass filtering, the expression is as follows:

[0021]

[0022] In the formula, is the normalized cutoff frequency, is the cut-off frequency, is the sampling frequency, G is the low-pass filter gain, is the Butterworth low-pass filter function.

[0023] Preferably, the expressions for initializing the zero acceleration threshold and the two length thresholds are as follows:

[0024]

[0025]

[0026]

[0027] In the formula, represents the zero acceleration threshold, They represent the zero acceleration empirical thresholds when the walking frequency is slow, normal, and fast walking, respectively. Indicates the first length threshold, Indicates the second length threshold.

[0028] Preferably, the specific process of step 2 is as follows:

[0029] S21, the first zero speed detection, Set the zero acceleration threshold of normal walking frequency, and the calculation expression is as follows:

[0030]

[0031] In the formula, express k The state value of the point, express k The acceleration modulus of the point;

[0032] S22. Calculate the difference between the state values ​​of two adjacent points to obtain the starting point and the ending point of the static or moving interval. The calculation expression is as follows:

[0033]

[0034] In the formula, , , , They are the starting point of the static interval, the end point of the static interval, the starting point of the motion interval, and the end point of the motion interval. express k-1 The state value of the point, express k+1 The state value of the point;

[0035] S23, calculate the state interval length LSI, the calculation expression is as follows:

[0036]

[0037] In the formula, represents the length of the static interval, Indicates the length of the exercise interval;

[0038] S24, compare the state interval length LSI with the first length threshold , select the corresponding zero acceleration threshold, and the calculation expression is as follows:

[0039] .

[0040] Preferably, the specific content of step 3 is: after comparing the state interval length with the second length threshold, if the static interval length is less than the second length threshold, the state value in the static interval is replaced with 0; if the motion interval length is less than the second length threshold, the state value in the motion interval is replaced with 1.

[0041] Preferably, the calculation expression of step 4 is as follows:

[0042]

[0043] In the formula, represents the number of pedestrian steps, represents the number of static intervals, Indicates the number of motion intervals.

[0044] Therefore, the present invention adopts the above-mentioned adaptive accurate pedometer method based on low-cost accelerometer, which has the following beneficial effects:

[0045] (1) Accurate step counting is achieved: Through two zero-speed detections and two length comparisons, the zero acceleration threshold is adaptively selected and the state vector is updated; thus accurately representing the state of the pedestrian. Finally, the number of state interval lengths is counted to achieve accurate step counting;

[0046] (2) Reduced algorithm complexity and improved accuracy and robustness: The magnitude of acceleration is used to identify the zero-speed interval, and only one length judgment is required to select the optimal threshold, which greatly reduces the complexity of the algorithm;

[0047] (3) The method provided by the present invention has good applicability and has very good accuracy on different data sets containing many pedestrians performing various activities.

[0048] The technical solution of the present invention is further described in detail below through the accompanying drawings and embodiments. BRIEF DESCRIPTION OF THE DRAWINGS

[0049] Figure 1 It is an overall flow chart of an adaptive and accurate step counting method based on a low-cost accelerometer of the present invention;

[0050] Figure 2 A schematic diagram of state vector updating according to an embodiment of the present invention;

[0051] Figure 3 The graphs are result graphs of five algorithms of the embodiment of the present invention under different walking types; wherein (a) is a result graph of walking type of going up / down stairs; (b) is a result graph of walking type of fast walking in the playground; (c) is a result graph of walking type of stepping on the spot; (d) is a result graph of walking type of "8" shape; (e) is a result graph of walking type of running in circles; (f) is a result graph of walking type of long-distance mixed walking; (g) is a result graph of walking type of going up stairs; (h) is a result graph of walking type of going down stairs; (i) is a result graph of walking type of corridor walking;

[0052] Figure 4 The figure is a statistical graph of the number of steps of different walking types using the five algorithms of the embodiments of the present invention under three data sets. DETAILED DESCRIPTION

[0053] The following detailed description of the embodiments of the present invention provided in the accompanying drawings is not intended to limit the scope of the invention claimed for protection, but merely represents selected embodiments of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0054] See also Figure 1-4 , an adaptive and accurate step counting method based on a low-cost accelerometer, comprising the following steps:

[0055] Step 1: Data preprocessing: including acceleration modulus calculation, gravity correction, data smoothing through Butterworth low-pass filter, initialization of zero acceleration threshold and two length thresholds;

[0056] The expression for calculating the acceleration modulus is as follows:

[0057]

[0058] In the formula, , , Represent the acceleration components of the X, Y, and Z axes respectively, It is the magnitude of acceleration, which indicates the instantaneous acceleration of the object measured by the IMU.

[0059] The calculation expression of gravity correction is as follows:

[0060]

[0061] In the formula, is the sampling frequency, is the acceleration modulus after separation of gravity acceleration at time t, Indicates time.

[0062] The process of data smoothing through the Butterworth low-pass filter is as follows:

[0063] The cutoff frequency is normalized and the calculation expression is as follows:

[0064]

[0065] The order of the filter To calculate, the expression is as follows:

[0066]

[0067] Calculate the output signal after Butterworth low-pass filtering, the expression is as follows:

[0068]

[0069] In the formula, is the normalized cutoff frequency, is the cut-off frequency, is the sampling frequency, G is the low-pass filter gain, is the Butterworth low-pass filter function.

[0070] The expressions for initializing the zero acceleration threshold and the two length thresholds are as follows:

[0071]

[0072]

[0073]

[0074] In the formula, represents the zero acceleration threshold, They represent the zero acceleration empirical thresholds when the walking frequency is slow, normal, and fast walking, respectively. Indicates the first length threshold, Indicates the second length threshold.

[0075] Step 2, adaptive zero acceleration threshold selection: Use the preset zero acceleration threshold to perform zero speed detection, the zero speed detection result is the state value 0 and 1, 0 represents motion, 1 represents stillness; then compare the state interval length LSI with the first length threshold, and then adaptively adjust the zero acceleration threshold; specifically: if the length of the still interval exceeds the first length threshold, it is considered that the zero acceleration threshold is too large, and the threshold is adjusted downward; similarly, if the length of the moving interval exceeds the first length threshold, it is considered that the zero acceleration threshold is too small, and the threshold is adjusted upward; in all other cases, the threshold remains unchanged; the specific process is as follows:

[0076] S21, the first zero speed detection, Set the zero acceleration threshold of normal walking frequency, and the calculation expression is as follows:

[0077]

[0078] In the formula, express k The state value of the point, express k The acceleration modulus of the point;

[0079] S22. Calculate the difference between the state values ​​of two adjacent points to obtain the starting point and the ending point of the static or moving interval. The calculation expression is as follows:

[0080]

[0081] In the formula, , , , They are the starting point of the static interval, the end point of the static interval, the starting point of the motion interval, and the end point of the motion interval. express k-1 The state value of the point, express k+1 The state value of the point;

[0082] S23, calculate the state interval length LSI, the calculation expression is as follows:

[0083]

[0084] In the formula, represents the length of the static interval, Indicates the length of the exercise interval;

[0085] S24, compare the state interval length LSI with the first length threshold , select the corresponding zero acceleration threshold, and the calculation expression is as follows:

[0086] .

[0087] Step 3, state vector update: Use the adjusted zero acceleration threshold to perform a second zero-speed detection on the pedestrian acceleration sequence, compare the state interval length with the second length threshold to eliminate some minor misjudgments, and update the state vector; the specific content is: after comparing the state interval length with the second length threshold, if the length of the static interval is less than the second length threshold, the state value in the static interval is replaced by 0; if the length of the moving interval is less than the second length threshold, the state value in the moving interval is replaced by 1.

[0088] Step 4: Count the number of moving intervals and static intervals, and use the number of the two intervals as the number of pedestrian steps. The static interval and moving interval after the state vector is updated are a group of continuous 1s and 0s, respectively. The calculation expressions are as follows:

[0089]

[0090] In the formula, represents the number of pedestrian steps, represents the number of static intervals, Indicates the number of motion intervals.

[0091] Example

[0092] The present invention uses a fixed threshold algorithm (Fixed Threshold), an LSTM algorithm, an angular rate energy detector (ARED) algorithm, a SVM algorithm and the proposed algorithm to perform ZVD (zero velocity detection) performance analysis on a local foot inertial dataset and a representative subset of two publicly available inertial datasets (IPIN2010 foot inertial dataset, University of Toronto foot inertial dataset). The algorithm parameter settings are shown in Table 1:

[0093] Table 1 Parameter settings

[0094]

[0095] The relevant information of the IMU of the local foot inertial dataset and two publicly available inertial datasets is compared. The performance of the IMU determines the quality of the dataset, making ZVD more challenging in the dataset built with ultra-low-cost IMU. The collective performance comparison is shown in Table 2:

[0096] Table 2 IMU performance comparison

[0097]

[0098] Among them, WT9011DCL-bt50 is the sensor used by us, VectorNav VN-100 is the sensor used by the University of Toronto foot inertial dataset, and MTx-28A53G25 is the sensor used by the IPIN2010 foot inertial dataset.

[0099] The experimental results of the fixed threshold algorithm, LSTM algorithm, angular rate energy detector (ARED) algorithm, SVM algorithm and the method proposed in this invention on three data sets are shown in Figure 2. Figure 3As shown in Figure 1, (b), (d), (e), and (f), the scenes depict pedestrians navigating on flat terrain with relatively simple and low-amplitude movements, resulting in minimal gyroscope accumulated error and drift, and showing good ZVD performance of the ARED algorithm. However, in stair environments or situations involving multiple turns, such movements pose significant challenges to both the gyroscope and the entire IMU. In these situations, the human foot movement may become more complex, including actions such as lifting the leg and walking, as shown in (a), (g), (h), and (i). In this case, the effectiveness of the ARED algorithm for ZVD is significantly reduced compared to the flat terrain condition. In addition, as shown in (c), when standing still with large foot movements, the ARED algorithm has difficulty accurately identifying the zero velocity phase, resulting in a large number of misjudgments, making the ZVD results almost worthless (the Fixed Threshold algorithm, LSTM algorithm, and SVM algorithm also face similar challenges). When training the model using the SVM algorithm, we curated different pedestrian activities from the three datasets, including climbing stairs, normal walking, fast walking, running, and standing still. The SVM algorithm uses the RBF kernel function with a gamma value set to 0.001, such as Figure 3 As shown in (d), (e), (f), and (i), the results show accurate identification of pedestrian motion types and precise ZVD in the test set with less complex motion. The LSTM algorithm also shows similar capabilities, but the LSTM algorithm is superior to the SVM algorithm in both modeling capabilities and sequence data processing. Figure 3 (b) in the figure shows that although the fixed threshold can skillfully detect single and regular pedestrian movements and display accurate values, this situation is rarely encountered in actual walking. Figure 3 (e) in the figure can reflect that the acceleration of pedestrians may cause the zero-speed phase detection to fail. On the contrary, if there is jitter around the set zero-speed threshold, it may cause a large number of false alarms. Therefore, it is actually not feasible to rely on a fixed threshold to identify the flexible pedestrian foot activities. Therefore, the present invention realizes the accurate identification of the pedestrian's movement state by performing two ZVD and length comparisons, ensuring that the pedestrian's foot state is accurate at every moment, such as going up and down stairs. Figure 3 (a), (g), (h); multiple turning actions such as Figure 3 (d) and (i) in the figure, and pedestrians walking at different speeds, including mixed speeds such as Figure 3 (b), (e), and (f) in the figure.

[0100] The step counts of different walking types under three data sets using the fixed threshold algorithm, LSTM algorithm, angular rate energy detector (ARED) algorithm, SVM algorithm and the method proposed in this invention are shown in Figure 2. Figure 4As shown, the results show that the method proposed in the present invention is consistent with the actual number of steps, and the number of steps generated by the LSTM algorithm, the angular rate energy detector (ARED) algorithm, and the SVM algorithm are always significantly higher than the actual number of steps.

[0101] At the same time, the detection accuracy and average detection accuracy of the five algorithms on the three data sets are compared, as shown in Table 3:

[0102] Table 3 Comparison of step counting accuracy (unit: %)

[0103]

[0104] The results show that the detection accuracy of the methods we use is 100%, which proves the robustness of the present invention. Compared with the fixed threshold algorithm (Fixed Threshold), SVM algorithm, angular rate energy detector (ARED) algorithm, and LSTM algorithm, the step count accuracy of the present invention is improved by 37.69%, 17.21%, 14.97% and 13.05%, respectively.

[0105] Finally, the present invention also makes a comparison in processing time, as shown in Table 4:

[0106] Table 4 Comparison of processing time (unit: s)

[0107]

[0108] The results show that the processing time of the LSTM algorithm, the angular rate energy detector (ARED) algorithm, and the SVM algorithm are significantly affected by the size of the training data, resulting in a significantly longer processing time than other algorithms. The processing time of the fixed threshold algorithm (FixedThreshold) is comparable to that of the processing method of the present invention, thereby ensuring accuracy and greatly reducing processing time.

[0109] Therefore, the present invention adopts the above-mentioned adaptive precise pedometer method based on low-cost accelerometer, and adaptively selects zero acceleration threshold and updates state vector through two zero speed detections and two length comparisons; thereby accurately representing the state of pedestrians, and finally counting the number of state interval lengths, thereby achieving accurate step counting while reducing the complexity of the algorithm.

[0110] Finally, it should be noted that the above embodiments are only used to illustrate the technical solution of the present invention rather than to limit it. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that they can still modify or replace the technical solution of the present invention with equivalents, and these modifications or equivalent replacements cannot cause the modified technical solution to deviate from the spirit and scope of the technical solution of the present invention.

Claims

1. An adaptive and accurate step counting method based on a low-cost accelerometer, characterized in that: The following steps are involved: Step 1: Data preprocessing: including acceleration modulus calculation, gravity correction, data smoothing through Butterworth low-pass filter, initialization of zero acceleration threshold and two length thresholds; Step 2, adaptive zero acceleration threshold selection: Use the preset zero acceleration threshold to perform the first zero speed detection, and the zero speed detection result is the state value 0 and 1, 0 represents motion, and 1 represents stillness; then compare the state interval length LSI with the first length threshold, and then adaptively adjust the zero acceleration threshold; specifically: if the length of the still interval exceeds the first length threshold, it is considered that the zero acceleration threshold is too large, and the threshold is adjusted downward; similarly, if the length of the motion interval exceeds the first length threshold, it is considered that the zero acceleration threshold is too small, and it is adjusted upward; in all other cases, the threshold remains unchanged; Step 3, state vector update: Use the adjusted zero acceleration threshold to perform a second zero speed detection on the pedestrian acceleration sequence, compare the state interval length with the second length threshold, and update the state vector; Step 4: Count the number of moving intervals and static intervals, and take the sum of the two types of intervals as the number of pedestrian steps, where the static interval and moving interval after the state vector is updated are a group of continuous 1s and 0s, respectively.

2. The adaptive accurate pedometer method based on a low-cost accelerometer according to claim 1, characterized in that: The expression for calculating the acceleration modulus is as follows: In the formula, , , Represent the acceleration components of the X, Y, and Z axes respectively, It is the magnitude of acceleration, which indicates the instantaneous acceleration of the object measured by the inertial measurement unit IMU.

3. The adaptive accurate pedometer method based on a low-cost accelerometer according to claim 2, characterized in that: The calculation expression of gravity correction is as follows: In the formula, is the sampling frequency, is the acceleration modulus after separation of gravity acceleration at time t, Indicates time.

4. The adaptive accurate pedometer method based on a low-cost accelerometer according to claim 3, characterized in that: The process of data smoothing through the Butterworth low-pass filter is as follows: The cutoff frequency is normalized and the calculation expression is as follows: The order of the filter To calculate, the expression is as follows: Calculate the output signal after Butterworth low-pass filtering, the expression is as follows: In the formula, is the normalized cutoff frequency, is the cut-off frequency, is the sampling frequency, G is the low-pass filter gain, is the Butterworth low-pass filter function.

5. The adaptive accurate pedometer method based on low-cost accelerometer according to claim 4, characterized in that: The expressions for initializing the zero acceleration threshold and the two length thresholds are as follows: In the formula, represents the zero acceleration threshold, They represent the zero acceleration empirical thresholds when the walking frequency is slow, normal, and fast walking, respectively. Indicates the first length threshold, Indicates the second length threshold.

6. The adaptive accurate step counting method based on low-cost accelerometer according to claim 5, characterized in that: The specific process of step 2 is as follows: S21, the first zero speed detection, Set the zero acceleration threshold of normal walking frequency, and the calculation expression is as follows: In the formula, express k The state value of the point, express k The acceleration modulus of the point; S22. Calculate the difference between the state values ​​of two adjacent points to obtain the starting point and the ending point of the static or moving interval. The calculation expression is as follows: In the formula, , , , They are the starting point of the static interval, the end point of the static interval, the starting point of the motion interval, and the end point of the motion interval. express k-1 The state value of the point, express k+1 The state value of the point; S23, calculate the state interval length LSI, the calculation expression is as follows: In the formula, represents the length of the static interval, Indicates the length of the exercise interval; S24, compare the state interval length LSI with the first length threshold , select the corresponding zero acceleration threshold, and the calculation expression is as follows: 。 7. The adaptive accurate pedometer method based on a low-cost accelerometer according to claim 6, characterized in that: The specific content of step 3 is: after comparing the state interval length with the second length threshold, if the static interval length is less than the second length threshold, the state value in the static interval is replaced by 0; if the motion interval length is less than the second length threshold, the state value in the motion interval is replaced by 1.

8. The adaptive accurate pedometer method based on a low-cost accelerometer according to claim 7, characterized in that: The calculation expression of step 4 is as follows: In the formula, represents the number of pedestrian steps, represents the number of static intervals, Indicates the number of motion intervals.

Images