Inertial navigation system aided positioning method and system for complex scenarios

Abstract

The application discloses a complex scene-oriented inertial navigation system auxiliary positioning method and system, and belongs to the technical field of navigation and positioning. The technical solution points are as follows: signals collected by an inertial measurement unit at the current time are acquired, the inertial measurement unit is located in an object to be positioned; whether the signals are interfered by electromagnetic interference is determined according to a preset convolutional neural network model; if yes, the signals are projected into a space of a preset dimension for adaptive filtering processing to obtain updated signals; and the predicted position of the object to be positioned is determined according to the updated signals. The original signals are filtered in the application, the electromagnetic interference can be effectively relieved, the relatively accurate signals can be recovered, the position of the object to be positioned can be determined based on the signals, and the rank of the covariance matrix is reduced through the projection mode in the application, the calculation amount is reduced, and the good filtering performance is maintained.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of navigation positioning, more particularly to an inertial navigation system auxiliary positioning method and system for complex scenes. BACKGROUND

[0002] An inertial navigation system (INS) is an autonomous navigation system that does not rely on external information and is based on Newtonian mechanics. It measures the acceleration and angular velocity of a carrier through an accelerometer and a gyroscope in an inertial measurement unit (IMU), and obtains the velocity, position and attitude information of the carrier through integration operation.

[0003] However, in actual applications, the inertial navigation system often faces various complex environmental disturbances, such as electromagnetic interference, vibration interference, temperature change, etc. These disturbances will affect the measurement accuracy of the IMU, resulting in large errors in the measured acceleration and angular velocity data, and further affecting the positioning accuracy.

[0004] For example, in the automated production scene of a mold workshop, the electrical equipment, motors, frequency converters, etc. in the workshop will generate a strong electromagnetic field, and the electronic components in the IMU are easily affected by electromagnetic interference, resulting in noise or distortion of the measurement signal. In addition, the high-temperature room and cooling room in the mold workshop will cause uneven and large changes in the environmental temperature, which will further increase the measurement error of the sensor components inside the IMU and affect the positioning accuracy. Therefore, the existing inertial navigation positioning method has deficiencies. SUMMARY

[0005] In view of the deficiencies of the prior art, the purpose of the present application is to provide an inertial navigation system auxiliary positioning method and system for complex scenes, which automatically identifies signals affected by electromagnetic interference through a deep learning method, filters and processes them, restores more accurate signals, and performs positioning analysis based on them.

[0006] To achieve the above-mentioned purpose, the present application provides the following technical solutions:

[0007] The present application provides an inertial navigation system auxiliary positioning method for complex scenes, comprising:

[0008] Obtaining signals collected by an inertial measurement unit at the current time, the inertial measurement unit being located in an object to be positioned;

[0009] Determining whether the signals are affected by electromagnetic interference according to a preset convolutional neural network model, if so, projecting the signals into a space of a preset dimension for adaptive filtering processing to obtain updated signals, the preset dimension being smaller than the dimension corresponding to the signals;

[0010] The predicted position of the object to be located is determined based on the update signal.

[0011] As a further improvement of the present invention, determining the predicted position of the object to be located based on the update signal includes:

[0012] Based on the update signal and the preset hidden Markov model, the motion mode of the object to be located is determined, including linear acceleration mode, turning mode and constant speed mode.

[0013] The first position of the object to be located is determined according to the positioning method corresponding to the motion mode;

[0014] Send a control command to the object to be located so that the object emits an infrared signal, and determine a neighboring object based on the infrared signal;

[0015] Obtain adjacent positions, and determine the predicted position of the object to be located based on the adjacent positions and the first position, wherein the adjacent positions are the first positions corresponding to the adjacent objects.

[0016] As a further improvement of the present invention, the step of projecting the signal onto a space of a preset dimension for adaptive filtering to obtain an updated signal includes:

[0017] The signal is projected into a space of a preset dimension through a preset projection matrix to obtain a first low-dimensional input vector;

[0018] The first low-dimensional input vector is input into the filter, and the updated signal is calculated through the low-dimensional filter coefficient vector, which is obtained from the signal acquired at the adjacent previous time step.

[0019] As a further improvement of the present invention, the low-dimensional filter coefficient vector is obtained from the signal acquired at the adjacent previous time step, including:

[0020] The signal acquired at the previous moment is projected into the space of the preset dimension through a preset projection matrix to obtain the second low-dimensional input vector;

[0021] The gain vector is calculated based on the second low-dimensional input vector and the covariance matrix;

[0022] The low-dimensional filter coefficient vector is calculated based on the gain vector.

[0023] As a further improvement of the present invention, the step of determining the first position of the object to be located according to the positioning method corresponding to the motion mode includes:

[0024] If the motion mode is linear acceleration mode, the acceleration value of the object to be located is determined according to the update signal, and the first position of the object to be located is determined according to the acceleration value;

[0025] If the motion mode is a constant speed mode, the velocity value of the object to be located is determined according to the update signal, and the first position of the object to be located is determined according to the velocity value.

[0026] If the motion mode is a turning mode, the angular velocity value of the object to be located is determined according to the update signal, and the first position of the object to be located is determined according to the angular velocity value and the circular motion model.

[0027] As a further improvement of the present invention, determining the first position of the object to be positioned based on the angular velocity value and the circular motion model includes:

[0028] The turning angle of the object to be positioned is calculated based on the angular velocity value;

[0029] The turning radius of the object to be positioned is calculated based on the angular velocity value and the turning angle.

[0030] The second position of the object to be positioned is determined based on the turning angle, the turning radius, and the circular motion model.

[0031] Based on the dynamic model, the second position is corrected to obtain the first position of the object to be located.

[0032] As a further improvement of the present invention, determining the predicted position of the object to be located based on the adjacent positions and the first position includes:

[0033] The distance between the adjacent object and the object to be located is determined based on the infrared signal;

[0034] The predicted position of the object to be located is determined based on the first position, the adjacent positions, and the distance between the adjacent objects and the object to be located.

[0035] As a further improvement of the present invention, determining the predicted position of the object to be located based on the first position, the adjacent positions, and the distance between the adjacent objects and the object to be located includes:

[0036] Based on a preset error standard, a first error range and a second error range are determined, wherein the first error range is the error range corresponding to the first position, and the second error range is the error range corresponding to the adjacent position;

[0037] The predicted position of the object to be located is determined based on the first error range, the second error range, and the distance between the adjacent objects and the object to be located.

[0038] As a further improvement of the present invention, determining the predicted position of the object to be located based on the first error range, the second error range, and the distance between the adjacent objects and the object to be located includes:

[0039] The circular equation is determined based on the first error range and the second error range;

[0040] Based on the circular equation and the distance between the adjacent objects and the object to be located, establish a system of inequalities;

[0041] The predicted position of the object to be located is determined by solving the system of inequalities.

[0042] This invention provides a deep learning-assisted positioning system for inertial navigation systems in complex scenarios, including an inertial measurement unit and a server;

[0043] The inertial measurement unit includes an accelerometer, a gyroscope, and an odometer;

[0044] The server includes:

[0045] The acquisition module is used to acquire the signal collected by the inertial measurement unit at the current moment, wherein the inertial measurement unit is located in the object to be located;

[0046] The calculation module is used to determine whether the signal is subject to electromagnetic interference based on a preset convolutional neural network model. If so, the signal is projected onto a space of a preset dimension for adaptive filtering to obtain an updated signal. The preset dimension is smaller than the dimension corresponding to the signal.

[0047] The positioning module determines the predicted position of the object to be located based on the update signal.

[0048] This invention effectively mitigates electromagnetic interference and recovers a more accurate signal by filtering the original signal, thereby determining the position of the object to be located. Furthermore, considering the high computational complexity of traditional filtering methods, this invention reduces the rank of the covariance matrix through projection, thus reducing computational load while maintaining good filtering performance. Attached Figure Description

[0049] Figure 1 This is a schematic diagram of the method steps of the present invention;

[0050] Figure 2 This is a schematic diagram of the scenario of the present invention;

[0051] Figure 3 A diagram of the predicted location Figure 1 ;

[0052] Figure 4 A diagram of the predicted location Figure 2 ;

[0053] Figure 5 This is a schematic diagram of the system structure of the present invention. Detailed Implementation

[0054] The technical solution of the present invention will be described in detail below with reference to the accompanying drawings and specific embodiments. It should be understood that the embodiments of the present invention and the specific features in the embodiments are detailed descriptions of the technical solution of the present invention, rather than limitations thereof.

[0055] The term "and / or" in the following text is merely a description of the relationship between related objects, indicating that three relationships can exist. For example, A and / or B can represent: A alone, A and B simultaneously, or B alone. Additionally, the character " / " generally indicates that the preceding and following related objects have an "or" relationship.

[0056] like Figure 1 As shown, embodiments of this application provide an inertial navigation system-assisted positioning method for complex scenarios, including:

[0057] The signal acquired by the inertial measurement unit at the current moment is obtained, and the inertial measurement unit is located in the object to be located;

[0058] The system determines whether the signal is subject to electromagnetic interference based on a pre-defined convolutional neural network model. If so, the signal is projected into a space of a pre-defined dimension for adaptive filtering to obtain an updated signal. The pre-defined dimension is smaller than the dimension corresponding to the signal.

[0059] The predicted position of the object to be located is determined based on the updated signal.

[0060] The object to be located can be a train, car, public transportation vehicle, or robot, for example, such as... Figure 2 As shown, in the automated production scenario of the mold workshop, there are multiple work areas such as high temperature room, cooling room, processing area, assembly area, and testing area. Each robot needs to travel to a specific work area to complete its work. In order to enable the robot to accurately reach the designated work area and avoid the inability to perform the work accurately due to position deviation, it is necessary to locate the robot's position during travel.

[0061] Specifically, the inertial measurement unit (IMU) includes multiple components, such as an accelerometer, gyroscope, odometer, and microcontroller module. The accelerometer measures acceleration, the gyroscope measures angular velocity, the odometer measures distance traveled and speed, and the microcontroller module triggers data acquisition at fixed short time intervals (e.g., 0.5 seconds). During each acquisition, the IMU reads the currently measured values ​​from the accelerometer, gyroscope, and odometer and stores them internally. At fixed long time intervals, the data acquired during this period is sent to the server for subsequent signal processing. This long time interval is the interval between two adjacent moments, and the data acquired during this period is the signal acquired by the IMU at the current moment. Each trigger point is used as a sampling point.

[0062] Therefore, the data collected by the inertial measurement unit includes multiple types, each corresponding to a component, and each component corresponding to a pre-defined convolutional neural network model. Thus, electromagnetic interference (EMI) judgment needs to be performed on each type of signal separately. Each pre-defined EMI model has the same structure, including an input layer, convolutional layer, pooling layer, fully connected layer, and output layer. The last fully connected layer has two output neurons, corresponding to the cases of electromagnetic interference and no electromagnetic interference. The output layer uses an activation function to convert the output of the fully connected layer into a probability distribution, outputting the probability that the signal is interfered with or not interfered with. If the probability of interference exceeds a fixed threshold, the signal is considered to be subject to EMI and requires subsequent filtering. If the signal is not subject to EMI, no filtering is required, and the motion pattern is directly determined by comparing the collected signal with the pre-defined Hidden Markov Model.

[0063] This embodiment effectively mitigates electromagnetic interference and recovers a more accurate signal by filtering the original signal, thereby determining the position of the object to be located. Furthermore, considering the high computational complexity of traditional filtering methods, this invention reduces the rank of the covariance matrix through projection, thus reducing computational load while maintaining good filtering performance.

[0064] Furthermore, embodiments of this application provide a step of projecting a signal onto a space of a preset dimension for adaptive filtering to obtain an updated signal, including:

[0065] The signal is projected into a space of a preset dimension through a preset projection matrix to obtain the first low-dimensional input vector;

[0066] The first low-dimensional input vector is input into the filter, and the updated signal is calculated through the low-dimensional filter coefficient vector, which is obtained from the signal acquired in the previous adjacent time step.

[0067] Further, this embodiment provides a step of obtaining a low-dimensional filter coefficient vector from signals collected at the previous adjacent moment, including:

[0068] Project the signal collected at the previous moment into a space of a preset dimension through a preset projection matrix to obtain a second low-dimensional input vector;

[0069] Calculate a gain vector based on the second low-dimensional input vector and the covariance matrix;

[0070] Calculate the low-dimensional filter coefficient vector based on the gain vector.

[0071] Specifically, for each moment, the following steps need to be repeated:

[0072] First, for each collected signal, construct its corresponding full-order input vector X(n) = [x(m), x(m - 1),..., x(m - N + 1)] T , where n represents the current moment, m represents the number of elements in the vector after representing the collected signal as a vector, that is, the number of sampling points within two adjacent moments, x(m) represents the mth element in the vector, N represents the order of the filter (i.e., the dimension corresponding to X(n)), and T represents the transpose. Then calculate the output result of the filter, that is, the current updated signal y(n) = w T (n)X(n), w(n) = Uw K (n), where w K (n) is the low-dimensional filter coefficient vector, calculated from the signal collected at the previous moment, and w(n) is the filter coefficient vector, obtained by transforming the low-dimensional filter coefficient vector through the projection matrix.

[0073] Next, project the full-order input vector X(n) into a low-dimensional space (a space of a preset dimension) through the projection matrix U. The dimension of the projection matrix U is N×K, to obtain the first low-dimensional input vector X K (n) = U T X(n). And calculate the output result of the filter in the low-dimensional space where is the ith element of the low-dimensional filter coefficient vector w K (n), is the ith element of the first low-dimensional input vector X K (n), K represents the dimension of the low-dimensional space, K < N, and the value of K can be determined according to the computing resources and time cost.

[0074] After that, obtain the expected signal q(n) of y d (n) according to the prior distribution, and calculate the error between them e(n) = q(n) - y d(n). The gain vector k is calculated based on the first low-dimensional input vector. K (n):

[0075]

[0076] Where λ represents the forgetting factor, P K (n-1) is the covariance matrix, which can be represented by the gain vector k from the previous time step (n-1). K (n-1) is calculated.

[0077] Then, based on the gain vector k K The low-dimensional filter coefficient vector and covariance matrix are updated by the error e(n) and the error e(n), respectively:

[0078] w K (n+1)=w K (n)+k K (n)e(n)

[0079]

[0080] Among them, I K Given a K-dimensional identity matrix, the final value of w is... K (n+1) is restored to w(n+1)=Uw through the projection matrix U. K (n+1) is used for calculations in the next adjacent time step.

[0081] This embodiment updates the filter coefficient vector and covariance matrix in a low-dimensional space and restores them to a high-dimensional space through a projection matrix for use in calculating the updated signal in the next time step. Compared with traditional filtering methods, the vector and matrix dimensions involved are lower, and the number of operations such as multiplication and addition is greatly reduced, thus reducing computational complexity.

[0082] Furthermore, this embodiment provides a method for determining the predicted position of an object to be located based on an update signal, including:

[0083] Based on the updated signal and the preset Hidden Markov Model, the motion mode of the object to be located is determined. The motion modes include linear acceleration mode, turning mode and constant speed mode.

[0084] The first position of the object to be located is determined according to the positioning method corresponding to the motion mode;

[0085] Send control commands to the object to be located so that the object emits infrared signals, and determine an adjacent object based on the infrared signals;

[0086] Obtain adjacent positions, and determine the predicted position of the object to be located based on the adjacent positions and the first position. The adjacent positions are the first positions corresponding to adjacent objects.

[0087] Specifically, after obtaining the first position, the server sends a control command to the object to be located, causing the object to emit an infrared signal through its own infrared communication module. The object receiving the infrared signal immediately returns an infrared signal. The distance between the object to be located and the object receiving the infrared signal can be determined by the time and speed of the infrared signal propagation. However, there may be multiple objects receiving infrared signals. In this embodiment, the closest object is selected as the neighboring object, and the neighboring position is obtained.

[0088] The preset Hidden Markov Model is trained based on historical data. The principle of the Hidden Markov Model is that the motion pattern is assumed to consist of a series of hidden states. The hidden states cannot be directly observed, but each hidden state will generate an observable feature vector with a certain probability, forming an observation sequence.

[0089] Applying this to the present embodiment, in the classification of motion patterns of the object to be located, the hidden state is different motion patterns (linear acceleration mode, turning mode, and constant speed mode). The observation sequence is a motion data vector obtained by converting the update signal from an electrical signal to a digital signal. For example, after filtering the signals collected by the accelerometer, gyroscope, and odometer, three update signals are obtained. After converting these three update signals into digital signals, the acceleration vector, angular velocity vector, and velocity vector of the object to be located at the current moment can be obtained. Each value in the acceleration vector, angular velocity vector, and velocity vector is normalized and combined to obtain the observation sequence [a, ω, v], where a, ω, and v are the normalized acceleration vector, angular velocity vector, and velocity vector, respectively.

[0090] Based on this, the observed sequence is input into a preset hidden Markov model, and the likelihood probability of the observed sequence under each motion mode can be calculated. The likelihood probability represents the probability that the observed sequence is generated by a specific motion mode. The motion mode with the highest likelihood probability is selected as the final classification result.

[0091] Furthermore, this embodiment provides a positioning method based on a motion pattern to determine the first position of an object to be positioned, including:

[0092] If the motion mode is linear acceleration mode, the acceleration value of the object to be located is determined according to the update signal, and the first position of the object to be located is determined according to the acceleration value.

[0093] If the motion mode is uniform speed mode, the velocity value of the object to be located is determined according to the update signal, and the first position of the object to be located is determined according to the velocity value.

[0094] If the motion mode is turning mode, the angular velocity value of the object to be located is determined according to the update signal, and the first position of the object to be located is determined according to the angular velocity value and the circular motion model.

[0095] Furthermore, this embodiment provides a method for determining the first position of an object to be located based on angular velocity values ​​and a circular motion model, including:

[0096] Calculate the turning angle of the object to be positioned based on the angular velocity value;

[0097] The turning radius of the object to be positioned is calculated based on the angular velocity value and the turning angle.

[0098] Based on the turning angle, turning radius, and circular motion model, determine the second position of the object to be positioned;

[0099] Based on the dynamic model, the second position is corrected to obtain the first position of the object to be located.

[0100] Specifically, assume the calculated angular velocity vector of the object to be located at the current moment is W1=(ω1,ω2,...,ω τ ), where τ represents the dimension corresponding to the angular velocity vector, and the interval between two adjacent sampling points is Δt, the turning angle can be calculated. The unit of Δt can be seconds. Then, based on the velocity vector and angular velocity vector of the object to be located at the current moment, the turning radius can be calculated. v avg and ω avg These are the mean values ​​of all elements in the velocity vector and angular velocity vector, respectively.

[0101] Assuming the predicted position of the object to be located was (x0, y0) calculated at the previous moment, according to the circular motion model, the second position (x1, y1) of the object to be located at the current moment is:

[0102]

[0103] Where θ0 represents the heading angle of the object to be located before the turn, and the heading angle represents the initial direction of travel of the object to be located before the turn.

[0104] To further improve the accuracy of positioning, a dynamic model can be used to correct the second position. For example, assuming the object to be positioned is a car, considering factors such as the vehicle's mass, tire lateral characteristics, and steering system characteristics, a dynamic model of the vehicle can be established as follows:

[0105]

[0106] Where M represents the mass of the vehicle, β(n) represents the lateral angle of the vehicle's center of gravity at the current moment, t(n) represents the road curvature at the current moment, h(n) and l(n) represent the lateral forces of the front and rear tires at the current moment, respectively, and ξ is the stability factor, which can be determined based on wheelbase, tire characteristics, etc.

[0107] Next, the lateral and longitudinal displacements of the vehicle are calculated using both the dynamic model and the circular motion model. The difference between the two lateral and two longitudinal displacements is used to obtain the lateral position error Δy and the longitudinal displacement error Δx. Based on the lateral position error Δy and the longitudinal displacement error Δx, the second position is corrected to obtain the first position as (x1+Δx, y1+Δy).

[0108] This embodiment uses a preset Hidden Markov Model to accurately determine the current motion mode of the object to be located, and selects different calculation methods for different motion modes, which can obtain the first position of the object to be located more accurately. Furthermore, in the turning mode, this embodiment considers the errors caused by factors such as vehicle mass and tire side slip characteristics through the dynamic model, which can obtain the first position of the object to be located more accurately.

[0109] Furthermore, this embodiment provides a step for determining the predicted position of an object to be located based on adjacent positions and a first position, including:

[0110] The distance between adjacent objects and the object to be located is determined based on infrared signals;

[0111] The predicted position of the object to be located is determined based on the first position, adjacent positions, and the distance between adjacent objects and the object to be located.

[0112] Furthermore, this embodiment provides a step for determining the predicted position of an object to be located based on a first position, adjacent positions, and the distance between adjacent objects and the object to be located, including:

[0113] Based on the preset error standard, a first error range and a second error range are determined. The first error range is the error range corresponding to the first position, and the second error range is the error range corresponding to the adjacent position.

[0114] The predicted position of the object to be located is determined based on the first error range, the second error range, and the distance between adjacent objects and the object to be located.

[0115] Furthermore, this embodiment provides a step for determining the predicted position of an object to be located based on a first error range, a second error range, and the distance between adjacent objects and the object to be located, including:

[0116] The circular equation is determined based on the first and second error ranges.

[0117] Based on the equation of a circle and the distances between adjacent objects and the object to be located, establish a system of inequalities;

[0118] The predicted position of the object to be located is determined by solving a system of inequalities.

[0119] For example, such as Figure 3 As shown, the two solid coils represent the first error range and the second error range, respectively. The centers of the two solid coils represent the first position and the adjacent position, respectively. The ranges within the two dashed lines represent the predicted positions of the object to be located and the adjacent object, respectively. d represents the distance between the adjacent object and the object to be located.

[0120] Specifically, the above filtering steps only solve the problem of electromagnetic interference. There is still a possibility that the first position obtained by the inertial measurement unit may be incorrect due to uneven temperature in the workshop. However, the degree of error between the first position and the accurate position in this temperature environment can be obtained through experiments or historical positioning data. Based on the degree of error, the position range of the accurate position can be determined. This range is the first error range. Similarly, the second error range corresponding to adjacent objects can be determined.

[0121] Assume the coordinates of the first position O3 are (x3, y3), the radius of the first error range is r3, the coordinates of the adjacent position O4 are (x4, y4), and the radius of the second error range is r4. r3 and r4 are determined based on the degree of error. Furthermore, this embodiment assumes that the diameters of both solid coils are greater than d, and that the accurate position A of the object to be located is (x3, y3). A ,y A The exact position B of the adjacent object is (x B ,y B ),but Known According to the formula for the magnitude vector, we know that:

[0122] (x B -x A ) 2 +(y B -y A ) 2 =d 2

[0123] Since A is within the first error range and B is within the second error range, we have the equation for a circle:

[0124] (x A -x3) 2 +(y A -y3) 2 ≤r3 2

[0125] (x B-x4) 2 +(y B -y4) 2 ≤r4 2

[0126] Substituting the two inequalities above into the formula for the magnitude vector, we get:

[0127] 2x3x A +2y3y A -2x A x B +2x4x B +2y4y B -2y A y B

[0128] ≥d 2 +x3 2 +y3 2 +x4 2 +y4 2 -r3 2 -r4 2

[0129] Combining the three inequalities above, we obtain a system of inequalities. By solving this system of inequalities, we can obtain x. A y A x B y B The range of values ​​is used to obtain the predicted positions of the object to be located and its adjacent objects.

[0130] In the prior art, the first error range and the second error range are usually used as the predicted positions of the object to be located and the adjacent objects, respectively. In this embodiment, the distance between the two objects is determined by infrared signals, and based on the first error range and the second error range, the range of the accurate position is further narrowed by the distance to obtain a more accurate predicted position.

[0131] Furthermore, in addition to the methods described above, this embodiment provides another method for determining the predicted position of an object to be located. For example, such as... Figure 4 As shown, the two solid ellipses represent the first and second error ranges, respectively, while the ranges within the dashed lines represent the predicted positions of the object to be located and adjacent objects, respectively.

[0132] Specifically, the first error range can be represented by the equation of the error ellipse:

[0133]

[0134] Where, σ A and σ BLet denot Δx and Δx represent the standard deviations in the horizontal and vertical directions, respectively. Assume the error between the first position and the accurate position in the horizontal and vertical directions are Δx and Δx, respectively. A and Δy A Furthermore, assuming that the measurement error follows a standard normal distribution, it can be approximated that the error range has a certain multiple relationship with the standard deviation. According to the 3 sigma principle, we can obtain...

[0135] The rotation angle of the ellipse, i.e. the tilt direction relative to the coordinate axes, is used to accurately match the ellipse with the direction of large measurement error. Specifically, based on the above filtering process, the covariance matrix corresponding to the updated signal can be obtained. Since the values ​​in the covariance matrix represent the dispersion of the data, the greater the dispersion, the greater the fluctuation of the data, i.e., the larger the measurement error, the larger the diagonal element corresponding to a certain direction in the covariance matrix is. Therefore, in this embodiment, when the diagonal element corresponding to a certain direction in the covariance matrix is ​​large, this direction is taken as the tilt direction of the ellipse, and thus the rotation angle is determined.

[0136] γ represents a constant related to the confidence level, determining the probability that the accurate location lies within the first error range. Since the error ellipse is typically constructed based on a two-dimensional normal distribution, for a given confidence level, such as 95% or 99%, the corresponding λ value can be obtained by consulting relevant statistical tables of the two-dimensional normal distribution. Similarly, the expression for the second error range can be derived.

[0137] After determining the distance d between adjacent objects and the object to be located, the predicted positions of the object to be located and adjacent objects are obtained by solving the system of inequalities as described above.

[0138] In a circular error range, every point within the circular area has the same probability of being considered the accurate location, failing to reflect differences in different directions. In contrast, this embodiment determines the predicted location using an elliptical error range. Compared to a circular error range, the elliptical error range is based on a two-dimensional Gaussian distribution to construct a probability model. At the center of the ellipse, the probability density is the highest, indicating that this is where the accurate location is most likely to occur. The probability density gradually decreases as one moves away from the center. The major and minor axes can quantify the magnitude of the error in different directions. The major axis of the ellipse corresponds to the direction with larger measurement errors, and the rate of change of probability density differs along the major and minor axes.

[0139] Therefore, the non-uniform probability distribution within the elliptical error range better reflects the distribution of errors in actual measurements and more accurately reflects the probability of different locations being the accurate location. Based on this, if it is necessary to further select a point within the predicted location as the accurate location, the point with the highest probability value can be chosen according to the probability distribution within the elliptical range to ensure positioning accuracy. Furthermore, since the area of ​​the elliptical error range is smaller than that of the circular error range, the range of the accurate location can be further narrowed, resulting in a more accurate predicted location.

[0140] Furthermore, such as Figure 5 As shown, this application embodiment provides a deep learning-assisted positioning system for inertial navigation systems in complex scenarios, including an inertial measurement unit and a server;

[0141] The inertial measurement unit includes an accelerometer, a gyroscope, and an odometer;

[0142] The servers include:

[0143] The acquisition module is used to acquire the signal collected by the inertial measurement unit at the current moment. The inertial measurement unit is located in the object to be located.

[0144] The calculation module is used to determine whether the signal is subject to electromagnetic interference based on the preset convolutional neural network model. If so, the signal is projected into a space of preset dimensions for adaptive filtering to obtain an updated signal.

[0145] The positioning module determines the predicted position of the object to be located based on the update signal.

[0146] The inertial navigation system-assisted positioning method and system for complex scenarios provided in this application firstly filters the original signal to alleviate electromagnetic interference and recover a more accurate signal; then, based on the recovered signal, the current motion mode of the object to be located is determined, and an appropriate calculation method is selected according to different motion modes to obtain the first position of the object to be located and determine the error range; finally, the range of the accurate position is narrowed down by the distance between the object to be located and adjacent objects to obtain the predicted position.

[0147] Those skilled in the art will understand that embodiments of the present invention can be provided as methods, systems, or computer program products. Therefore, the present invention can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention can take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0148] This invention is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the invention. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart illustrations and / or block diagrams. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0149] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0150] The above description is merely a preferred embodiment of the present invention. The scope of protection of the present invention is not limited to the above embodiments. All technical solutions falling within the scope of the present invention's concept are within the scope of protection of the present invention. It should be noted that for those skilled in the art, any improvements and modifications made without departing from the principles of the present invention should also be considered within the scope of protection of the present invention.

Claims

1. An inertial navigation system-assisted positioning method for complex scenarios, characterized in that, include: The signal collected by the inertial measurement unit at the current moment is acquired, and the inertial measurement unit is located in the object to be located; The system determines whether the signal is subject to electromagnetic interference based on a preset convolutional neural network model. If so, the signal is projected into a space of a preset dimension for adaptive filtering to obtain an updated signal. The preset dimension is smaller than the dimension corresponding to the signal. The predicted position of the object to be located is determined based on the update signal; The step of determining the predicted position of the object to be located based on the update signal includes: Based on the update signal and the preset hidden Markov model, the motion mode of the object to be located is determined, including linear acceleration mode, turning mode and constant speed mode. The first position of the object to be located is determined according to the positioning method corresponding to the motion mode; Send a control command to the object to be located so that the object emits an infrared signal, and determine a neighboring object based on the infrared signal; Obtain adjacent positions, and determine the predicted position of the object to be located based on the adjacent positions and the first position, wherein the adjacent positions are the first positions corresponding to the adjacent objects; The step of determining the predicted position of the object to be located based on the adjacent positions and the first position includes: The distance between the adjacent object and the object to be located is determined based on the infrared signal; The predicted position of the object to be located is determined based on the first position, the adjacent positions, and the distance between the adjacent objects and the object to be located. The step of determining the predicted position of the object to be located based on the first position, the adjacent positions, and the distance between the adjacent objects and the object to be located includes: Based on a preset error standard, a first error range and a second error range are determined, wherein the first error range is the error range corresponding to the first position, and the second error range is the error range corresponding to the adjacent position; The predicted position of the object to be located is determined based on the first error range, the second error range, and the distance between the adjacent objects and the object to be located. The first error range is represented by the equation of an ellipse: , Indicates the exact position of the object to be located. This represents the coordinates of the first position. and These represent the standard deviations in the horizontal and vertical directions, respectively. The rotation angle of the ellipse is represented by the direction of the diagonal element corresponding to a certain direction in the covariance matrix of the updated signal. When the diagonal element corresponding to a certain direction is large, that direction is taken as the tilt direction of the ellipse, thus determining the rotation angle of the ellipse. This represents a constant related to the confidence level.

2. The inertial navigation system-assisted positioning method for complex scenarios according to claim 1, characterized in that, The step of projecting the signal onto a space of a preset dimension for adaptive filtering to obtain an updated signal includes: The signal is projected into a space of a preset dimension through a preset projection matrix to obtain a first low-dimensional input vector; The first low-dimensional input vector is input into the filter, and the updated signal is calculated through the low-dimensional filter coefficient vector, which is obtained from the signal acquired at the adjacent previous time step.

3. The inertial navigation system-assisted positioning method for complex scenarios according to claim 2, characterized in that, The low-dimensional filter coefficient vector is obtained from the signal acquired at the adjacent previous time step, and includes: The signal acquired at the previous moment is projected into the space of the preset dimension through a preset projection matrix to obtain the second low-dimensional input vector; The gain vector is calculated based on the second low-dimensional input vector and the covariance matrix; The low-dimensional filter coefficient vector is calculated based on the gain vector.

4. The inertial navigation system-assisted positioning method for complex scenarios according to claim 1, characterized in that, Determining the first position of the object to be located according to the positioning method corresponding to the motion pattern includes: If the motion mode is linear acceleration mode, the acceleration value of the object to be located is determined according to the update signal, and the first position of the object to be located is determined according to the acceleration value; If the motion mode is a constant speed mode, the velocity value of the object to be located is determined according to the update signal, and the first position of the object to be located is determined according to the velocity value. If the motion mode is a turning mode, the angular velocity value of the object to be located is determined according to the update signal, and the first position of the object to be located is determined according to the angular velocity value and the circular motion model.

5. The inertial navigation system-assisted positioning method for complex scenarios according to claim 4, characterized in that, Determining the first position of the object to be located based on the angular velocity value and the circular motion model includes: The turning angle of the object to be positioned is calculated based on the angular velocity value; The turning radius of the object to be positioned is calculated based on the angular velocity value and the turning angle. The second position of the object to be positioned is determined based on the turning angle, the turning radius, and the circular motion model. Based on the dynamic model, the second position is corrected to obtain the first position of the object to be located.

6. The inertial navigation system-assisted positioning method for complex scenarios according to claim 1, characterized in that, Determining the predicted position of the object to be located based on the first error range, the second error range, and the distance between the adjacent objects and the object to be located includes: The circular equation is determined based on the first error range and the second error range; Based on the circular equation and the distance between the adjacent objects and the object to be located, establish a system of inequalities; The predicted position of the object to be located is determined by solving the system of inequalities.

7. A deep learning-assisted positioning system for inertial navigation systems in complex scenarios, characterized in that: Includes inertial measurement unit and server; The inertial measurement unit includes an accelerometer, a gyroscope, and an odometer; The server includes: The acquisition module is used to acquire the signal collected by the inertial measurement unit at the current moment, wherein the inertial measurement unit is located in the object to be located; The calculation module is used to determine whether the signal is subject to electromagnetic interference based on a preset convolutional neural network model. If so, the signal is projected onto a space of a preset dimension for adaptive filtering to obtain an updated signal. The preset dimension is smaller than the dimension corresponding to the signal. The positioning module determines the predicted position of the object to be located based on the update signal; The step of determining the predicted position of the object to be located based on the update signal includes: Based on the update signal and the preset hidden Markov model, the motion mode of the object to be located is determined, including linear acceleration mode, turning mode and constant speed mode. The first position of the object to be located is determined according to the positioning method corresponding to the motion mode; Send a control command to the object to be located so that the object emits an infrared signal, and determine a neighboring object based on the infrared signal; Obtain adjacent positions, and determine the predicted position of the object to be located based on the adjacent positions and the first position, wherein the adjacent positions are the first positions corresponding to the adjacent objects; The step of determining the predicted position of the object to be located based on the adjacent positions and the first position includes: The distance between the adjacent object and the object to be located is determined based on the infrared signal; The predicted position of the object to be located is determined based on the first position, the adjacent positions, and the distance between the adjacent objects and the object to be located. The step of determining the predicted position of the object to be located based on the first position, the adjacent positions, and the distance between the adjacent objects and the object to be located includes: Based on a preset error standard, a first error range and a second error range are determined, wherein the first error range is the error range corresponding to the first position, and the second error range is the error range corresponding to the adjacent position; The predicted position of the object to be located is determined based on the first error range, the second error range, and the distance between the adjacent objects and the object to be located. The first error range is represented by the equation of an ellipse: , Indicates the exact position of the object to be located. This represents the coordinates of the first position. and These represent the standard deviations in the horizontal and vertical directions, respectively. The rotation angle of the ellipse is represented by the direction of the diagonal element corresponding to a certain direction in the covariance matrix of the updated signal. When the diagonal element corresponding to a certain direction is large, that direction is taken as the tilt direction of the ellipse, thus determining the rotation angle of the ellipse. This represents a constant related to the confidence level.

Images