Pedestrian autonomous positioning joint calibration method based on waist-foot matching

Abstract

The invention discloses a pedestrian self-positioning joint calibration method based on waist-foot matching, and the method comprises the steps: firstly obtaining the wave crest and wave trough module values and stride frequency of an accelerometer in a pedestrian motion state through a pedestrian waist-loaded inertial measurement unit (IMU); and outputting a pedestrian foot zero-speed detector threshold value and a current pedestrian motion state after the prediction of the BP neural network model. And then zero speed point discrimination is carried out through a generalized likelihood ratio calculated by a pedestrian foot inertial measurement unit in combination with a motion state and a prediction threshold. Fixed threshold detection is used during normal walking, predicted threshold detection is used during fast walking / jogging, a generalized likelihood ratio minimum point is searched for for zero-speed detection under fast running, and in order to improve the model accuracy of the BP neural network, a zero-speed detector threshold is obtained through a zero-speed interval search method and is fed back to the BP neural network to update model parameters. And finally, performing online sensor error joint calibration by combining zero-speed updating with the waist course.

Description

Technical Field

[0001] The present invention belongs to the technical field of navigation and positioning, and relates to a pedestrian autonomous positioning and joint calibration method based on waist-foot matching. Background Art

[0002] Zero-velocity detection refers to the moment when the speed of a pedestrian's foot approaches zero after touchdown is detected using sensor data in a pedestrian foot-mounted inertial navigation system. It is an important step in pedestrian foot-mounted inertial positioning, and the accuracy of zero-velocity detection affects the accuracy of the foot-mounted inertial positioning system. Currently, relying solely on foot-mounted IMU data and using the generalized likelihood ratio with a fixed threshold for zero-velocity detection can achieve good positioning accuracy during slow walking. However, in cases of multiple motion states, variable-speed motion, fast running, etc., using a zero-velocity detector with a fixed threshold will result in a decrease in positioning accuracy due to a decrease in recognition accuracy.

[0003] To address this problem, the literature (Liu Siyu, Cui Jianmin, Liu Guodong, et al. An Adaptive Zero-Velocity Detection Pedestrian Navigation Method Based on Gait Characteristic Analysis [J]. Automation & Instrumentation, 2024, (06): 1-5.) proposed a method that uses a support vector machine (SVM) to determine the motion state, and then uses an RNN neural network to combine the original inertial data with the motion state to determine whether the output is a zero-velocity event. This method is superior to the zero-velocity detection method with a fixed threshold in complex motion states, but it requires pre-collecting data to train the neural network model, does not verify the generality of the model in multi-person tests, and increases the additional computational load of the system. The literature (Zhao Xiaoming, Deng Fangjin, Yang Songpu, et al. A Zero-Velocity Correction Method for Indoor Pedestrian Positioning Based on Pressure Sensor Assistance [J]. Chinese Journal of Inertial Technology, 2018, 26(01): 1-5.) proposed a multi-condition constraint zero-velocity correction method based on MIMU and pressure sensors. By using the specific force modulus, the sliding variance of the specific force, the angular velocity modulus, and the plantar pressure modulus, the zero-velocity interval is comprehensively detected. Compared with the zero-velocity correction method that relies solely on acceleration and angular velocity, this method improves the horizontal accuracy during normal walking and running. However, this method makes the installation complex due to the addition of a pressure sensor and does not consider the applicability of non-flat roads.

[0004] After retrieval, the application publication number is CN109297484A, which is a method for correcting the autonomous positioning error of pedestrians with gait constraints. By restricting the change of the pedestrian's heading (turning 0°, 90°, or 180°), the template matching method based on sequence detection is used to correct the heading error in real time, solving the problem of heading divergence caused by the long-term drift of inertial devices and improving the accuracy of pedestrian autonomous positioning. The method includes: (1) collecting IMU data and detecting static gait using the constraint conditions of pedestrian movement; (2) calculating the heading information of the pedestrian and determining the walking behavior. Different walking behaviors are matched with the trained gait sequence templates, and the heading error is corrected in real time according to the heading constraints under the templates; (3) using an intelligent filter to fuse the systematic error for error estimation and outputting the final positioning information of the pedestrian. This method uses the characteristics of the indoor positioning scenario of pedestrians to suppress the divergence of the heading error, but these scenario characteristics are not necessarily unsuitable for outdoor scenarios. For example, the sheltered street corners in urban canyons are not necessarily strictly 90°. At the same time, under the multi-gait movement of pedestrians, in addition to the positioning error caused by the heading, the speed error caused by the misdetection or missed detection of zero-speed detection will also affect the final positioning error. The present invention improves the correct rate of zero-speed detection under multi-gait to improve the overall positioning accuracy. Summary of the Invention

[0005] The present invention aims to solve the above problems of the prior art. A joint calibration method for pedestrian autonomous positioning based on waist-foot matching is proposed. The technical solution of the present invention is as follows:

[0006] A joint calibration method for pedestrian autonomous positioning based on waist-foot matching, which includes the following steps:

[0007] Step a, based on the pedestrian's waist-mounted IMU, calculate the modulus value of the accelerometer in the pedestrian's motion state, obtain the values at the peaks and valleys of the accelerometer modulus value, and take the reciprocal of the time interval between two valleys as the walking frequency;

[0008] Step b, based on the pedestrian's foot IMU, obtain the generalized likelihood ratio of the current window according to the measured gyroscope and accelerometer values;

[0009] Step c, based on the values at the peaks and valleys of the obtained accelerometer modulus value and the walking frequency, obtain the predicted zero-speed detector threshold and the current pedestrian motion state through a 4-layer BP neural network;

[0010] Step d, based on the obtained predicted zero-speed detector threshold, pedestrian motion state, and generalized likelihood ratio, perform zero-speed detection to obtain zero-speed points;

[0011] Step e, after a period of time after exiting zero-speed detection, based on the cached generalized likelihood ratio values, use zero-speed interval search to obtain the optimal zero-speed interval, thereby obtaining the current optimal zero-speed detector threshold, and feedback it to the BP neural network for updating the BP neural network model parameters;

[0012] Step f: Online calibration and compensation of sensor errors are performed by updating the gyro and accelerometer biases through zero-speed feedback, and its expression is:

[0013]

[0014] In the formula, is the three-dimensional gyro angular rate output vector after online compensation of the bias at time k, is the three-dimensional gyro angular rate output vector without bias compensation at time k, and δw k is the gyro three-axis bias vector at time k.

[0015] Furthermore, step a is based on the pedestrian waist-mounted IMU, calculates the modulus value of the accelerometer in the pedestrian's motion state, obtains the values at the peaks and valleys of the accelerometer modulus value, and takes the reciprocal of the time interval between two valleys as the step frequency. Specifically, it includes:

[0016] The walking and running motions of pedestrians are periodic. In each period, there is an up-and-down fluctuation of the pedestrian's waist. By analyzing the changes in the sensor output caused by the waist fluctuation of the pedestrian, the gait detection of the pedestrian can be realized; among them, the acceleration modulus value is used to reflect this change. The IMU carried on the waist has three-axis accelerometers, and the axes are right-front-up. Let the output values of the three-axis accelerometers be a x 、a y 、a z ;

[0017] Then the expression of the acceleration modulus value is as follows:

[0018]

[0019] In the formula, Acc norm is the modulus value output by the accelerometer in the pedestrian waist IMU. When the pedestrian goes from lifting the foot to putting it down once, Acc norm will successively appear peaks and valleys. The appearance of a valley reflects the moment when the pedestrian puts down the foot. Let the acceleration modulus values at the peak and valley be Acc norm_max 、Acc norm_min ; The modulus value at the peak minus the modulus value at the valley gives the height Acc norm_h of the wave. Acc norm_h can reflect the current motion intensity of the pedestrian. Take the reciprocal f step of the time interval between two adjacent valleys as the current instantaneous step frequency, which reflects the current motion speed of the pedestrian.

[0020] Furthermore, step b is based on the pedestrian foot IMU, and according to the measured gyro and accelerometer values, the generalized likelihood ratio of the current window is obtained. Specifically, it includes:

[0021] The IMU carried on the foot outputs the triaxial accelerometer and triaxial gyroscope at the current moment, and caches the accelerometer and gyroscope data at the previous w - 1 moments. Let the outputs of the triaxial accelerometer and gyroscope at the kth moment be a k =[a fx ,a fy ,a fz and ω k =[ω fx ,ω fy ,ω fz , with the axes being right - front - up, w being the window length, which is 3. First, calculate the average acceleration vector within the current window and the modulus

[0022]

[0023] Then, combine the gyroscope data within the window to calculate the generalized likelihood ratio T:

[0024]

[0025] In the above formula, and are the noise variances of the accelerometer and gyroscope, g is the acceleration due to gravity. i is the ith data within the window.

[0026] Furthermore, step c is based on the values at the peaks and valleys of the obtained accelerometer modulus and the step frequency, and through a 4 - layer BP neural network, the predicted zero - speed detector threshold and the current pedestrian motion state are obtained. Specifically, it includes:

[0027] Based on the current pedestrian motion intensity Acc norm_h , the current instantaneous step frequency f step is fed into the BP neural network for inference to obtain the predicted threshold T p of the zero - speed detector; among them, the BP neural network has a 4 - layer structure, including an input - output layer and 2 hidden layers, and each hidden layer contains 3 nodes; this neural network needs to be trained offline first: the pedestrian first collects data from normal walking to normal running, uses the optimal zero - speed interval search method to obtain the optimal zero - speed interval threshold for each step, normalizes the input data Acc norm_h and f step , and the ideal output threshold, and then conducts neural network training. The number of training batches is more than 1e4 to obtain the pre - trained model parameters of the neural network; this model will be used as the general model parameters of the neural network prediction algorithm; at the same time, the current motion state is predicted according to the current motion intensity Acc norm_h . The specific method is:

[0028]

[0029] Further, step d performs zero-speed detection to obtain zero-speed points based on the obtained predicted zero-speed detector threshold, pedestrian motion state, and generalized likelihood ratio, specifically including:

[0030] Based on the trough signal, current motion state, and predicted threshold T obtained in steps a, b, and c p , the current generalized likelihood ratio T at the current moment, perform zero-speed point discrimination, including three strategies for three motion states, a threshold-based detector, and the specific process is as follows: In the initial state, the pedestrian motion state is considered slow walking, and a fixed threshold T is used F for detection, and the detection method is:

[0031]

[0032] If the current is a candidate zero-speed point and is the first candidate zero-speed point of this gait, it is necessary to judge the quality of the current threshold. If the quality of the current threshold does not meet the standard, the current threshold is not used, and the strategy under fast running is selected; among them, the quality of the current threshold is judged by the speed calculated by the foot. While performing zero-speed detection of the foot, the speed v at the current moment will be output based on the foot sensor data k for pedestrian navigation and positioning; according to the pedestrian foot-mounted inertial navigation method, when the pedestrian's foot lands, the true speed of the foot-mounted IMU is considered zero speed plus random perturbation speed; and the speed calculated by the foot-mounted IMU at this moment is the true speed plus a certain speed error. At least one zero-speed point update is performed every time the foot lands. Let the pedestrian calculate the northeast sky speed as v E 、v N 、v U ; the judgment condition is:

[0033]

[0034] When the motion state is fast walking / jogging, use the medium-speed detection strategy. The medium-speed detection strategy uses the predicted threshold T p , first use the waist parameters to assist the foot for zero-speed detection. According to the continuity of the pedestrian's motion, first use the predicted threshold obtained from the previous step parameters for zero-speed point detection and threshold quality judgment. When the parameters of the next step arrive, use the predicted threshold given by the next step for zero-speed point detection and threshold quality judgment; if the quality of the predicted thresholds given by the two steps of the waist does not meet the standard, use the strategy under fast running;

[0035] When the motion state is the fast running motion state, use the fast detection strategy; the specific process of the fast detection strategy is that when the signal of the subsequent wave trough of the waist sensor arrives, start to find the moment of the generalized likelihood ratio wave trough. When the moment of the wave trough is found, judge whether the generalized likelihood ratio at this moment is less than the upper limit fixed threshold. If the condition is met, it is considered that this moment is the zero speed point, and record the calculated speed at this point as the speed error for zero speed update.

[0036] Further, step e specifically includes:

[0037] When performing zero speed point discrimination based on step d, cache the generalized likelihood ratio data at the same time. The cache window contains the data of the previous 0.7 s. To execute the zero speed interval search process, it is necessary to first identify the zero speed interval exit signal, which specifically includes: when using the fixed threshold or the predicted threshold, after the zero speed point has been detected, if there are 5 consecutive moments that do not meet the zero speed point discrimination, it is considered that the zero speed interval ends; under the fast motion strategy, when the generalized likelihood ratio wave trough is detected, it is considered that the zero speed interval ends; after the zero speed interval ends, cache the data of 0.2 s, and then perform the zero speed interval search on the cache window. The specific process is as follows:

[0038] First, find the minimum point in the buffer area as the initial zero speed interval average value T aver , taking two points to the left and right in turn with the minimum point as the center, and naming them T R2 , T R1 , T L2 , T L2 ;

[0039] Take the minimum point as the zero speed interval, and now perform the zero speed interval expansion operation, expanding one point each time, and expanding in two directions respectively each time;

[0040] When expanding to the left, T L1 is the nearest point to the zero speed interval. If T L1 <1.1T aver or T L1 shows a downward trend, then it is considered that T L1 is the zero speed interval, and the zero speed interval expands to the left, and T L1 is involved in the zero speed interval to take the average and update T aver ;

[0041] If reaching the buffer boundary, or when T L1 >1.1T aver and T L2 >1.2T aver , it is considered that the zero speed interval boundary is reached, and stop expanding in this direction;

[0042] When both directions stop expanding, find the maximum value of the generalized likelihood ratio within the zero-velocity interval as the zero-velocity interval, and send this value into the neural network for backpropagation to update the network parameters.

[0043] Further, step f specifically includes:

[0044] Perform online calibration compensation for sensor errors through the gyro zero bias feedback by zero-velocity update, and its expression is:

[0045]

[0046] In the formula, is the three-dimensional gyro angular rate output vector after online compensation of the zero bias at time k, is the three-dimensional gyro angular rate output vector without zero bias compensation at time k, and δw k is the gyro three-axis zero bias vector at time k.

[0047] An electronic device, which includes a memory, a processor, and a computer program stored on the memory and executable on the processor. When the processor executes the program, it implements the pedestrian autonomous positioning joint calibration method based on waist-foot matching as described in any one of the above.

[0048] A non-transitory computer-readable storage medium, on which a computer program is stored. When the computer program is executed by a processor, it implements the pedestrian autonomous positioning joint calibration method based on waist-foot matching as described in any one of the above.

[0049] The advantages and beneficial effects of the present invention are as follows:

[0050] The present invention uses a dual-node combination of a waist sensor node and a foot sensor to enhance gait perception ability and improve positioning accuracy. For example, in step b, the waist sensor data is used to identify the current gait of the pedestrian, and the BP neural network is used to output a prediction threshold. Since the waist movement can accurately reflect the movement state of the pedestrian, the waist sensor can be used to sense the gait. In step c, three strategies are selected through the output result of the waist sensor to handle zero-velocity detection under multiple gaits, and at the same time, the prediction threshold can be used to handle each gait under medium exercise intensity. When running fast, in the face of fewer zero-velocity intervals and unfixed thresholds, using the scheme of selecting the minimum generalized likelihood ratio can ensure zero-velocity update for each step while reducing the probability of false detection. Steps e and f correct the BP neural network model and sensor zero bias by caching past data. Description of the Drawings

[0051] Figure 1 is a system block diagram of foot zero-velocity detection assisted by a waist node in a preferred embodiment provided by the present invention;

[0052] Figure 2It is the flowchart of the zero-speed detection method in step d of the present invention;

[0053] Figure 3 It is the flowchart of the optimal zero-speed threshold search algorithm in step e of the present invention;

[0054] Figure 4 It is the effect diagram of the online calibration of the gyro zero bias in step f of the present invention.

[0055] Figure 5 It is the comparison diagram of the positioning effects of using a fixed threshold for zero-speed detection and using the zero-speed detection method described in the present invention. Specific implementation mode

[0056] Next, the technical solutions in the embodiments of the present invention will be clearly and detailedly described in conjunction with the accompanying drawings in the embodiments of the present invention. The described embodiments are only a part of the embodiments of the present invention.

[0057] The technical solution of the present invention to solve the above technical problems is:

[0058] See Figure 1 , a pedestrian autonomous positioning and joint calibration method based on waist-foot matching, including the following steps:

[0059] In step a, the walking and running movements of a pedestrian are periodic. In each period, there is an up-and-down fluctuation of the pedestrian's waist. By analyzing the changes in the sensor output caused by the waist fluctuation of the pedestrian, the gait detection of the pedestrian can be realized. Among them, the acceleration modulus can more obviously reflect this change. The IMU carried by the waist has three-axis accelerometers, and the axes are right-front-up. Let the output values of the three-axis accelerometers be a x , a y , a z .

[0060] Then the acceleration modulus expression is as follows:

[0061]

[0062] In the formula, Acc norm is the modulus of the output of the accelerometer in the IMU of the pedestrian's waist. When a pedestrian goes through the process of lifting the foot and then putting it down once, Acc norm will successively appear wave peaks and wave valleys. The appearance of a wave valley reflects the moment when the pedestrian puts down the foot. Let the acceleration moduli at the wave peak and wave valley be Acc norm_max , Acc norm_min . The modulus at the wave peak minus the modulus at the wave valley gives the height Acc norm_h of the wave. Acc norm_h can reflect the current movement intensity of the pedestrian. Take the reciprocal f step of the time interval between two adjacent wave valleys.As the current instantaneous step frequency, it reflects the current pedestrian movement speed.

[0063] Step b: Based on the IMU carried by the foot, output the triaxial accelerometer and triaxial gyroscope at the current moment, and cache the accelerometer and gyroscope data at the previous w - 1 moments. Assume that the output of the triaxial accelerometer and gyroscope at time k is a k =[a fx ,a fy ,a fz and ω k =[ω fx ,ω fy ,ω fz , with the axial directions being right - front - up, w being the window length, which is 3. First, calculate the average acceleration vector within the current window and its modulus

[0064]

[0065] Then, combine the gyroscope data within the window to calculate the generalized likelihood ratio T:

[0066]

[0067] In the above formula, and are the noise variances of the accelerometer and gyroscope, and g is the acceleration due to gravity.

[0068] Step c: Based on the pedestrian's current movement intensity Acc norm_h , and the current instantaneous step frequency f step , feed them into the BP neural network for inference to obtain the prediction threshold T p of the zero - speed detector. Among them, the BP neural network has a 4 - layer structure, including an input - output layer and 2 hidden layers, and each hidden layer contains 3 nodes. This neural network needs to be trained offline first: The pedestrian first collects data from normal walking to normal running, uses the optimal zero - speed interval search method to obtain the optimal zero - speed interval threshold for each step, normalizes the input data Acc norm_h and f step , and the ideal output threshold, and then conducts neural network training with more than 1e4 training batches to obtain the pre - trained model parameters of the neural network. This model will be used as the general model parameters for the neural network prediction algorithm. At the same time, predict the current movement state according to the current movement intensity Acc norm_h . The specific method is:

[0069]

[0070] Step d: Based on the trough signal, current movement state, and prediction threshold T obtained in steps a, b, and c p, at the current moment, the generalized likelihood ratio T is used to determine the zero-velocity point. The specific content includes three strategies for three motion states, a threshold-based detector, and a principle of minimizing the false detection rate and increasing the detection rate. The specific process is as follows: In the initial state, the pedestrian's motion state is considered to be slow walking, and a fixed threshold T is used F for detection. The detection method is as follows:

[0071]

[0072] If the current is a candidate zero-velocity point and it is the first candidate zero-velocity point of this gait, it is necessary to judge the quality of the current threshold. If the quality of the current threshold does not meet the standard, in order to reduce the false detection rate, the current threshold is not used, and the strategy under fast running is selected. Among them, the quality of the current threshold is judged by the speed calculated by the foot. While performing zero-velocity detection of the foot, based on the foot sensor data, the speed v at the current moment will be output k for pedestrian navigation and positioning. According to the pedestrian foot-mounted inertial navigation method, when the pedestrian's foot lands, the true speed of the foot-mounted IMU can be considered as zero speed plus a random perturbation speed. And the speed calculated by the foot-mounted IMU at this moment is the true speed plus a certain speed error. Since the zero-velocity point is updated at least once every time the foot lands, the speed error is suppressed, and it is feasible to judge the quality of the current threshold by the currently calculated speed. Let the pedestrian calculate the northeast sky speed as v E 、v N 、v U . The judgment condition is:

[0073]

[0074] When the motion state is fast walking / jogging, the medium-speed detection strategy is used. The medium-speed detection strategy uses the predicted threshold T p , because for one step of foot movement, while the waist sensor can sense two steps. The parameters of the previous step can be generated prior to the true zero-velocity interval of the foot, and the parameters of the latter step may be generated later than the start time of the true zero-velocity interval of the foot. This situation is common in the slow motion state. In order to use the waist parameters to assist the foot in zero-velocity detection first, according to the continuity of pedestrian motion, the predicted threshold obtained from the parameters of the previous step is first used for zero-velocity point detection and threshold quality judgment. When the parameters of the latter step arrive, the predicted threshold given by the latter step is used for zero-velocity point detection and threshold quality judgment. If the quality of the predicted thresholds given by the two steps of the waist does not meet the standard, the strategy under fast running is used.

[0075] When the motion state is a fast running motion state, a fast detection strategy is used. When a pedestrian runs fast, the trough signal sensed by the waist-mounted sensor will precede the start time of the true zero-velocity interval of the foot, and the true zero-velocity interval will be shorter during fast movement. However, due to the contact of the pedestrian's foot with the ground, there is theoretically a moment when the true velocity of the foot IMU is zero. The fast detection strategy considers the moment of true zero as the trough moment of the generalized likelihood ratio in one step. Therefore, the specific process of the fast detection strategy is that when the trough signal of the next step of the waist sensor arrives, start looking for the trough moment of the generalized likelihood ratio. When the trough moment is found, judge whether the generalized likelihood ratio at this moment is less than the upper fixed threshold. If the condition is met, it is considered that this moment is the zero-velocity point, and record the calculated velocity at this point as the velocity error for zero-velocity update. Using the zero-velocity detection method described in the present invention, the positioning accuracy can be maintained under fast running.

[0076] Step e, when performing zero-velocity point discrimination based on step d, cache the generalized likelihood ratio data at the same time. The cache window contains the data of the previous 0.7 s. To execute the zero-velocity interval search process, it is necessary to first identify the zero-velocity interval exit signal. Specifically, when using a fixed threshold or a predicted threshold, after a zero-velocity point has been detected, if there are 5 consecutive moments that do not meet the zero-velocity point discrimination, it is considered that the zero-velocity interval ends. Under the fast movement strategy, when the trough of the generalized likelihood ratio is detected, it is considered that the zero-velocity interval ends. After the zero-velocity interval ends, cache the data for 0.2 s, and perform a zero-velocity interval search on the cache window. The specific process is as follows:

[0077] First, find the minimum point in the buffer as the initial zero-velocity interval average T aver , and take two points to the left and right of the minimum point as the center, and name them T R2 , T R1 , T L2 , T L2 .

[0078] Take the minimum point as the zero-velocity interval, and now perform the zero-velocity interval expansion operation, expanding one point each time, and expanding in two directions respectively each time.

[0079] Here, taking the leftward expansion as an example, T L1 is the closest point to the zero-velocity interval. If T L1 <1.1T aver or T L1 shows a downward trend, then it is considered that T L1 is the zero-velocity interval, and the zero-velocity interval expands to the left, and T L1 is involved in taking the average of the zero-velocity interval to update T aver .

[0080] If the buffer boundary is reached, or when T L1 >1.1T aver and T L2> 1.2T aver When it reaches this value, it is considered to reach the boundary of the zero-speed interval, and the expansion in this direction stops.

[0081] When the expansion stops in both directions, find the maximum value of the generalized likelihood ratio within the zero-speed interval as the zero-speed interval, and send this value into the neural network for backpropagation to update the network parameters.

[0082] Step f: Through the gyroscope and accelerometer zero biases feedback by zero-speed update, perform online calibration compensation for sensor errors, and its expression is:

[0083]

[0084] In the formula, is the three-dimensional gyroscope angular rate output vector after online compensation of the zero bias at time k, is the three-dimensional gyroscope angular rate output vector without zero bias compensation at time k, and δw k is the gyroscope three-axis zero bias vector at time k.

[0085] Furthermore, in order to verify the effectiveness of the method of the present invention in pedestrian autonomous positioning and joint calibration, a field experiment on hybrid gait pedestrian positioning is carried out for verification. The performance parameters of the IMU in the experiment are listed in Table 1.

[0086] Table 1:

[0087]

[0088] Three groups of repeated positioning experiments are carried out on the IMU acquisition data corresponding to Table 1, and then the pedestrian positioning solutions are carried out by using the method of the present invention and the classical zero-speed update method respectively.

[0089] As Figure 4 shown is the online calibration effect diagram of the gyroscope zero bias by using the method of the present invention. It can be seen from the figure that the errors of the sensors are effectively estimated in real time. As Figure 5 shown is the comparison diagram of the positioning effects by using the fixed threshold for zero-speed detection and the zero-speed detection method described in the present invention in one of the groups of experiments. It can be clearly seen from Figure 5 that the positioning accuracy of the method of the present invention is higher. Therefore, the pedestrian autonomous positioning and joint calibration method of this application has a better sensor error compensation effect.

[0090] The comparison of the pedestrian positioning errors by using the two methods is shown in Table 2.

[0091] Table 2:

[0092]

[0093] As can be seen from the comparison results of the pedestrian positioning errors in Table 2, the method provided by the present invention can achieve zero-velocity detection with higher precision and online compensation for sensor errors. Compared with the traditional fixed-threshold zero-velocity update method, the pedestrian positioning error of the method of the present invention can still be maintained within 5 m in the case of hybrid gait. It can be seen that the method of the present invention is effective and correct in high-precision zero-velocity detection and online compensation for sensor errors, and has higher calibration accuracy than the traditional fixed-threshold zero-velocity detection method.

[0094] The systems, devices, modules or units illustrated in the above embodiments can be specifically implemented by computer chips or entities, or by products with certain functions.

[0095] Computer-readable media includes permanent and non-permanent, removable and non-removable media and can be implemented by any method or technology for information storage. The information can be computer-readable instructions, data structures, program modules or other data. Examples of computer storage media include, but are not limited to, phase change memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory or other memory technologies, compact disc read-only memory (CD-ROM), digital versatile disc (DVD) or other optical storage, magnetic cassette tapes, magnetic tape magnetic disk storage or other magnetic storage devices, or any other non-transmission media that can be used to store information accessible by a computing device. As defined herein, computer-readable media does not include transitory computer-readable media such as modulated data signals and carrier waves.

[0096] It should also be noted that the term "comprises", "comprising" or any other variation thereof is intended to cover a non-exclusive inclusion, such that a process, method, commodity or device comprising a series of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, commodity or device. Without further limitation, an element defined by the statement "comprising an..." does not exclude the presence of additional identical elements in the process, method, commodity or device comprising the element.

[0097] The above embodiments should be understood as being only for illustrative purposes of the present invention and not for limiting the protection scope of the present invention. After reading the content recorded in the present invention, those skilled in the art can 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 autonomous positioning joint calibration method based on waist-foot matching, characterized in that: The following steps are involved: Step a, based on the pedestrian's waist-mounted IMU, calculate the accelerometer modulus in the pedestrian's motion state, obtain the values ​​at the peak and trough of the accelerometer modulus, and use the reciprocal of the time interval between two troughs as the step frequency; Step b, based on the pedestrian's foot IMU, obtain the generalized likelihood ratio of the current window according to the measured gyro and accelerometer values; Step c, based on the obtained values ​​of the accelerometer modulus peaks and troughs and the step frequency, a predicted zero-speed detector threshold and the current pedestrian motion state are obtained through a 4-layer BP neural network; Step d, performing zero-speed detection to obtain a zero-speed point based on the predicted zero-speed detector threshold, pedestrian motion state, and generalized likelihood ratio; Step e, after exiting the zero-speed detection for a period of time, based on the cached generalized likelihood ratio value, the zero-speed interval is searched to obtain the optimal zero-speed interval, thereby obtaining the current optimal zero-speed detector threshold, which is fed back to the BP neural network to update the BP neural network model parameters; Step f, by updating the feedback gyro at zero speed and adding the table zero bias, the sensor error is calibrated and compensated online, and the expression is: In the formula, is the three-dimensional gyro angular rate output vector after online compensation of zero bias at time k, is the uncompensated zero-bias three-dimensional gyro angular rate output vector at time k, δw k is the gyro three-axis zero bias vector at time k.

2. According to claim 1, a pedestrian autonomous positioning joint calibration method based on waist-foot matching is characterized by: The step a calculates the accelerometer modulus of the pedestrian in motion based on the pedestrian's waist-mounted IMU, obtains the values ​​at the peak and trough of the accelerometer modulus, and uses the inverse of the time interval between two troughs as the step frequency, specifically including: Pedestrians' walking and running movements are cyclical. In each cycle, the pedestrian's waist will float up and down. By analyzing the changes in sensor output caused by the floating of the pedestrian's waist, pedestrian gait detection can be achieved. The acceleration modulus is used to reflect this change. The IMU carried on the waist has three axial accelerometers, which are right-front-up. Suppose the output values ​​of the three-axis accelerometers are a x 、a y 、a z ; Then the acceleration modulus expression is as follows: Where Acc norm is the modulus of the accelerometer output in the pedestrian's waist IMU. When the pedestrian lifts his foot and puts it down, Acc norm There will be peaks and troughs in sequence. The troughs reflect the moment when pedestrians land. The acceleration moduli at the peaks and troughs are Acc respectively. norm_max 、Acc norm_min ; The modulus value at the peak minus the modulus value at the trough gives the wave height Acc norm_h .Accel norm_h It can reflect the current pedestrian movement intensity, and the inverse of the time interval between two adjacent troughs is f step As the current instantaneous step frequency, it reflects the current pedestrian movement speed.

3. The pedestrian autonomous positioning and joint calibration method based on waist-foot matching according to claim 1 is characterized by: The step b is based on the pedestrian's foot IMU and obtains the generalized likelihood ratio of the current window according to the measured gyro and accelerometer values, specifically including: Based on the IMU carried by the foot, the three-axis accelerometer and three-axis gyroscope are output at the current moment, and the accelerometer and gyroscope data of the previous w-1 moments are cached. Assume that the output of the three-axis accelerometer and gyroscope at moment k is a k =[a fx ,a fy ,a fz ] and ω k =[ω fx ,ω fy ,ω fz ], the axis is right-front-up, w is the window length, which is 3, first calculate the average acceleration vector in the current window and modulus Calculate the generalized likelihood ratio T from the gyroscope data within the combination window: In the above formula, and is the noise variance of the accelerometer and gyroscope, g is the acceleration of gravity, and i is the i-th data in the window.

4. The pedestrian autonomous positioning joint calibration method based on waist-foot matching according to claim 2 is characterized by: The step c is based on the obtained values ​​of the peak and trough of the accelerometer modulus value and the step frequency, and obtains the predicted zero-speed detector threshold and the current pedestrian motion state through a 4-layer BP neural network, specifically including: Based on the pedestrian's current motion intensity Acc obtained in step a norm_h , current instantaneous step frequency f step The prediction threshold T of the zero-speed detector is obtained by sending it to the BP neural network for inference. p ; Among them, the BP neural network has a 4-layer structure, including input and output layers, 2 hidden layers, and each hidden layer contains 3 nodes; the neural network needs to be trained offline first: pedestrians first collect normal walking to normal running data, and use the optimal zero-speed interval search method to obtain the optimal zero-speed interval threshold for each step, and the input data Acc norm_h and f step , the ideal output threshold is normalized, and then the neural network is trained. The training batch is more than 1e4, and the neural network pre-training model parameters are obtained; this model will be used as the general model parameters of the neural network prediction algorithm; at the same time, according to the current exercise intensity Acc norm_h Predict the current motion state. The specific method is:

5. The pedestrian autonomous positioning and joint calibration method based on waist-foot matching according to claim 1 is characterized by: The step d performs zero-speed detection to obtain a zero-speed point based on the predicted zero-speed detector threshold, pedestrian motion state, and generalized likelihood ratio, and specifically includes: Based on the trough signal obtained in steps a, b, and c, the current motion state, and the predicted threshold T p , the generalized likelihood ratio T at the current moment, and the zero-speed point discrimination, including three strategies for three motion states, a threshold-based detector, the specific process is as follows: in the initial state, the pedestrian motion state is considered to be slow walking, using a fixed threshold T F Detection, the detection method is: If the current point is a candidate for zero speed, and it is the first candidate for zero speed in this gait, the quality of the current threshold needs to be judged. If the quality of the current threshold does not meet the standard, the current threshold is not used, and the strategy of fast running is selected; the quality of the current threshold is judged by the speed calculated by the foot. While performing the foot zero speed detection, the current speed v is output based on the foot sensor data. k Used for pedestrian navigation and positioning; according to the pedestrian foot-mounted inertial navigation method, when the pedestrian's foot lands, the real speed of the foot-mounted IMU is considered to be zero speed plus random perturbation speed; at this moment, the speed calculated by the foot-mounted IMU is the real speed plus a certain speed error. Each time the foot lands, at least one zero speed point update is performed. Suppose the pedestrian calculates the northeast sky speed as v E 、v N 、v U ; The judgment conditions are: When the motion state is fast walking / jogging, the medium speed detection strategy is used, and the medium speed detection strategy uses the prediction threshold T p First, use the waist parameters to assist the foot in zero-speed detection. According to the continuity of pedestrian movement, first use the prediction threshold obtained by the previous step parameters to perform zero-speed point detection and threshold quality judgment. When the next step parameters arrive, use the prediction threshold given by the next step to perform zero-speed point detection and threshold quality judgment. If the quality of the prediction threshold given by the waist in the two steps does not meet the standard, use the strategy of fast running. When the motion state is a fast running motion state, a fast detection strategy is used; the specific process of the fast detection strategy is that when the trough signal of the waist sensor arrives, start looking for the trough moment of the generalized likelihood ratio. When the trough moment is found, determine whether the generalized likelihood ratio at this moment is less than the upper fixed threshold. If the condition is met, the moment is considered to be the zero speed point, and the calculated speed of the point is recorded as the speed error for zero speed update.

6. The pedestrian autonomous positioning and joint calibration method based on waist-foot matching according to claim 5 is characterized by: The step e specifically comprises: When the zero-speed point is judged based on step d, the generalized likelihood ratio data is cached at the same time, and the cache window contains the data of the previous 0.7s. In order to execute the zero-speed interval search process, it is necessary to first identify the zero-speed interval exit signal, which specifically includes: when using a fixed threshold or a predicted threshold, after the zero-speed point has been detected, if there are 5 consecutive moments that do not meet the zero-speed point judgment, the zero-speed interval is considered to be over; under the fast motion strategy, the generalized likelihood ratio trough is detected, and the zero-speed interval is considered to be over; after the zero-speed interval ends, 0.2s of data is cached, and then the zero-speed interval search is performed on the cache window. The specific process is as follows: First find the minimum point of the buffer zone as the average value of the initial zero speed interval T aver , take the minimum point as the center, and select two points to the left and right, named T R2 , T R1 , T L2 , T L2 ; The minimum point is taken as the zero-speed interval, and the zero-speed interval expansion operation is now performed, expanding one point at a time, and expanding in two directions each time; When expanding to the left, T L1 is the closest point in the zero speed interval. If T L1 <1.1T aver Or T L1 If the trend is downward, it is considered that T L1 The zero speed interval is extended to the left. L1 Participate in the zero speed interval to take the average update T aver ; If the buffer boundary is reached, or when T L1 >1.1T aver and T L2 >1.2T aver When , it is considered that the boundary of the zero-speed interval has been reached, and expansion in this direction is stopped; When the expansion in both directions stops, the maximum value of the generalized likelihood ratio in the zero-speed interval is found as the zero-speed interval, and the value is sent to the neural network for back propagation to update the network parameters.

7. According to claim 1, a pedestrian autonomous positioning joint calibration method based on waist-foot matching is characterized in that: The step f specifically comprises: The sensor error is compensated online by updating the gyro bias feedback at zero speed. The expression is: In the formula, is the 3D gyro angular rate output vector after online compensation of zero bias at time k, is the uncompensated zero-bias three-dimensional gyro angular rate output vector at time k, δw k is the gyro three-axis zero bias vector at time k.

8. An electronic device, characterized in that: It includes a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the program, it implements the pedestrian autonomous positioning and joint calibration method based on waist-foot matching as described in any one of claims 1 to 7.

9. A non-transitory computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the pedestrian autonomous positioning and joint calibration method based on waist-foot matching as described in any one of claims 1 to 7 is implemented.

Images