Method for dynamic error prediction correction for electromagnetic positioning systems

Abstract

The application discloses a method for dynamic error prediction correction of an electromagnetic positioning system, which unifies multiple data points collected at different time points in a measurement period to a same target timestamp by establishing a dynamic prediction model to generate a correction data frame with completely aligned timestamps, and uses the correction data frame for pose solution. The application provides various implementation manners, including a prediction method based on linear extrapolation of historical data frames, a nonlinear prediction method based on polynomial function fitting, a method of directly predicting by taking measurement values as states and using a standard Kalman filter, and a hybrid prediction method of inversely solving theoretical measurement values by a forward model after predicting future poses by taking sensor poses as states and using an extended Kalman filter. The application covers various prediction schemes by a unified framework, and significantly improves the accuracy and robustness of the electromagnetic positioning system in a dynamic scene.

Description

Technical Field

[0001] This invention relates to the technical field of electromagnetic positioning systems, and in particular to a method for dynamic error prediction and correction of electromagnetic positioning systems, with a focus on a real-time signal processing method for electromagnetic positioning systems operating in time-division multiplexing mode. Background Technology

[0002] The magnetic field generators in common electromagnetic positioning systems typically operate in a time-division multiplexing mode. They sequentially excite multiple signal sources according to a preset order and simultaneously measure the responses of one or more sensors to each signal source to calculate the precise pose of the sensors in real time. These systems are widely used due to their relatively simple hardware implementation and the absence of crosstalk between channels.

[0003] A typical time-division multiplexing electromagnetic positioning system operates with the following data acquisition process: The system controller sequentially excites transmitting coils one through N. Within each excitation window, the system measures the induced voltage corresponding to the sensor. to These N voltage values ​​together constitute a complete data frame, which is uploaded to the host computer for subsequent pose calculation.

[0004] However, this time-division multiplexing operating mode presents an inherent physical challenge that is often overlooked in existing technologies. A data frame is not an instantaneous snapshot, but a sequence with an inherent time span. Assuming that it takes time Δt to excite each coil and complete the measurement, the total duration of an eight-channel data frame is... This is eight times Δt. This means that the first voltage value within the frame... At the start of the frame Measured, and the last voltage value That is in This is in addition to the measurement taken at a time seven times Δt. There is a significant intra-frame temporal discrepancy between these two measurements.

[0005] When the sensor is static or moving at extremely slow speeds, the effect of this time deviation can be ignored. This is because the sensor's pose can be approximated as unchanged throughout the entire measurement time window, so the entire data frame can be considered as a sampling of the magnetic field response at the same instant.

[0006] However, in many practical applications such as medical navigation, sensors often undergo high-speed or high-acceleration movements. In such dynamic scenarios, the assumption that the pose remains constant throughout the measurement period no longer holds. Position and measurement at time The pose may have changed significantly. At this point, traditional positioning algorithms still treat this data frame, pieced together from measurements taken at different poses, as a single instantaneous value. This approach is physically incorrect and inevitably leads to severe dynamic positioning errors. These errors manifest in several ways: first, hysteresis error, meaning the positioning result always lags behind the sensor's true position; second, when the sensor starts, stops, or turns rapidly, the positioning trajectory exhibits overshoot and oscillations beyond the true path; and finally, during smooth curve motion, the calculated trajectory is distorted and does not conform to the true path.

[0007] The aforementioned dynamic errors severely impact the system's real-time performance and reliability, degrade user experience, and may even pose safety risks in demanding clinical applications. Existing technologies for handling system errors largely focus on compensating for static magnetic field distortion caused by metallic environments, or smoothing the trajectory through post-processing filtering after the positioning results are output. These methods are passive error handling approaches. Static distortion compensation methods cannot address dynamic problems caused by the sensor's motion itself; while post-processing filtering can make the trajectory visually smoother, it comes at the cost of introducing additional latency and cannot correct physical information distortion caused by intra-frame time deviations, potentially even erasing true motion details. None of these existing technologies address the intra-frame time deviation problem caused by time-division multiplexing mechanisms and sensor motion coupling at the root of signal generation.

[0008] Therefore, there is an urgent need in this field for a real-time processing method that can actively predict and correct such intra-frame time deviations, so as to fundamentally improve the positioning accuracy and robustness of time-division multiplexing positioning systems in dynamic scenarios. Summary of the Invention

[0009] The purpose of this invention is to overcome the shortcomings and deficiencies of the prior art and provide a method for dynamic error prediction and correction of electromagnetic positioning systems. This method no longer passively accepts raw measurement data with time deviation, but actively corrects all voltage values ​​in a measurement data frame to the same target timestamp by establishing a dynamic prediction model, thereby eliminating the source of dynamic error at the signal level and significantly improving the positioning performance of electromagnetic positioning systems in high-speed motion scenarios.

[0010] To achieve the above objectives, the technical solution provided by this invention is as follows: a method for dynamic error prediction and correction of an electromagnetic positioning system, wherein the electromagnetic positioning system includes a magnetic field generator and a sensor; the magnetic field generator operates in a time-division multiplexing mode, sequentially exciting eight signal sources according to a predetermined timing sequence, and the response of each signal source constitutes an independent measurement channel; the sensor senses the voltage value corresponding to each measurement channel, and these voltage values ​​are collected and processed for pose calculation; the method includes the following steps:

[0011] S1. Receive timing data frame: Receive a current data frame in real time. The current data frame consists of eight voltage values ​​measured sequentially according to the predetermined timing sequence. Each voltage value in the current data frame corresponds to a different measurement time.

[0012] S2. Establish a dynamic prediction model: Based on the current data frame and at least one historical data frame stored in the memory, establish a dynamic prediction model to describe the evolution of voltage values ​​over time.

[0013] S3. Generate a correction data frame: Take the timestamp corresponding to the most recently updated voltage value in the current data frame as the target timestamp, use the dynamic prediction model to calculate the predicted values ​​of all measurement channels at the same target timestamp within the measurement period of the current data frame, and combine these predicted values ​​into a correction data frame with perfectly aligned timestamps.

[0014] S4. Perform pose calculation: Input the correction data frame into a positioning algorithm to calculate the pose of the sensor.

[0015] Furthermore, the first implementation of the dynamic prediction model is linear extrapolation, as detailed below:

[0016] For each measurement channel, a first-order rate of change is calculated using the current voltage value in the current data frame and the corresponding historical voltage value in the historical data frame. This first-order rate of change is mathematically expressed as... ,in, The first rate of change, The current voltage value, The historical voltage value, The time interval between two data frames;

[0017] For each voltage value within the current data frame, a linear extrapolation is performed using the first-order rate of change based on the time difference between its measurement time and the target timestamp to obtain its predicted value at the target timestamp. This predicted value is mathematically expressed as... ,in, The predicted value, The time difference between the voltage value and the target timestamp.

[0018] Furthermore, the second implementation of the dynamic prediction model is nonlinear polynomial extrapolation, as detailed below:

[0019] For each measurement channel, using the voltage value in the current data frame and the corresponding voltage values ​​in at least two historical data frames, a second-order or higher-order polynomial function describing the change of voltage value over time is fitted using Newton's difference method or the least squares method.

[0020] The target timestamp is used as an input variable and substituted into the polynomial function to calculate the function value at the target timestamp, and the function value is used as the predicted value.

[0021] Furthermore, the third implementation of the dynamic prediction model is a Kalman filter, as detailed below:

[0022] Define a signal state vector that contains at least the current voltage values ​​of all measurement channels and their first-order rates of change.

[0023] Upon receiving each new data frame, the data frame is treated as a measurement observation, and the Kalman filter update step is executed to correct the estimated value of the signal state vector; wherein, the update step fuses the difference between the measurement observation and a prior state estimate through a Kalman gain to obtain a posterior state estimate;

[0024] The prediction step of the Kalman filter is performed to obtain the predicted value of the signal state vector at the target timestamp; wherein the prediction step utilizes a state transition matrix to propagate the posterior state estimate from its corresponding time to the target timestamp, and the state transition matrix describes the physical or kinematic laws of the evolution of the signal state vector over time.

[0025] The values ​​of all measurement channels are extracted from the predicted signal state vector and combined into the correction data frame.

[0026] Furthermore, the fourth implementation of the dynamic prediction model is a pose-state-based extended Kalman filter. The extended Kalman filter uses the pose of the sensor and its rate of change as its internal pose state vector, and the process of generating the correction data frame is as follows:

[0027] a. Pose prediction: Perform the prediction step of the extended Kalman filter, that is, use a kinematic model that describes the kinematic laws of the sensor to predict a predicted pose at the target timestamp;

[0028] b. Inverse Theoretical Value Solution: Input the predicted pose into a known forward model that describes the mapping relationship from sensor pose to voltage value, and calculate a complete time-aligned theoretical data frame;

[0029] c. Generate correction data frame: Use the theoretical data frame as the correction data frame.

[0030] Furthermore, in the process of generating the correction data frame, after step b, there is also a fusion step, which specifically involves: obtaining one or more real measured voltage values ​​at the current time near the target timestamp, and replacing the corresponding components in the theoretical data frame with them to generate a correction data frame that incorporates the real measured values.

[0031] Compared with the prior art, the present invention has the following beneficial effects:

[0032] 1. Eliminating dynamic errors at their source. Most existing technologies smooth positioning results through post-processing filtering, a passive error suppression method that cannot eliminate physical information distortion caused by intra-frame time deviations. The method of this invention establishes a dynamic prediction model, proactively correcting measurement data from a time series to a single timestamp before pose calculation. This eliminates the source of dynamic errors at the root of signal generation, achieving a more realistic reflection of the sensor's motion state.

[0033] 2. Significantly improves positioning accuracy and smoothness under high-speed motion. By compensating for intra-frame time deviations, the method of this invention can effectively suppress positioning lag, overshoot, and oscillation problems caused by the sensor during high-speed motion, sharp turns, or start-stop states. The final calculated motion trajectory is closer to the real physical path, and the trajectory smoothness and continuity are significantly improved, thus providing users with more stable and reliable real-time positioning feedback.

[0034] 3. Providing a Multi-Mode, Scalable Technical Framework: This invention provides a unified prediction and correction framework, and elaborates on various specific implementation methods, including linear extrapolation, nonlinear polynomial fitting, standard Kalman filtering, and pose-based extended Kalman filtering. This framework offers high flexibility and scalability, allowing users to select the most suitable prediction model based on their application scenario's requirements for real-time performance, accuracy, and computational resources. For example, a computationally inexpensive linear extrapolation model can be chosen for resource-constrained embedded systems, while a high-performance extended Kalman filter model can be selected for applications requiring extreme accuracy. Attached Figure Description

[0035] Figure 1 This is a schematic diagram of the internal structure of the magnetic field generator in an electromagnetic positioning system.

[0036] Figure 2 This is a flowchart of the pose calculation process for an electromagnetic positioning system using the method of this invention. Detailed Implementation

[0037] The present invention will now be described in detail with reference to the accompanying drawings and specific embodiments.

[0038] The implementation of this invention relies on an electromagnetic positioning system. (See reference...) Figure 1 As shown in the figure, this diagram illustrates the internal structure of a magnetic field generator in an electromagnetic positioning system. The magnetic field generator includes a housing and an internal support structure on which multiple transmitting coils arranged in a specific spatial layout are fixed. In this embodiment, eight transmitting coils are used (see Figure 1). Figure 1 (As shown in Figures 1, 2, 3, 4, 5, 6, 7, and 8). During system operation, a magnetic field generator, as illustrated, sequentially excites these transmitting coils according to a preset timing sequence, causing them to generate a time-varying magnetic field. A sensor located within this magnetic field will induce voltage signals, which are then collected and processed for pose calculation.

[0039] This invention proposes a method for dynamic error prediction and correction in electromagnetic positioning systems, the core process of which is as follows: Figure 2 As shown below, different embodiments will be used to explain in detail the specific implementation of each step in this process.

[0040] The initial step of this method is to receive a timing data frame. The data acquisition unit of the electromagnetic positioning system obtains a current data frame from the sensor. This data frame consists of eight voltage values, each corresponding to a specific voltage value. Figure 1 The response generated by one of the eight transmitting coils at a specific moment during excitation. Since the excitation is time-division multiplexed, the measurement times of these eight voltage values ​​are sequentially increased, resulting in intra-frame time deviation.

[0041] Next, the electromagnetic positioning system establishes a dynamic prediction model using time-series data frames. This step builds an independent dynamic prediction model for each measurement channel based on the received current data frame and at least one historical data frame stored in memory.

[0042] In the step of generating a correction data frame, the timestamp corresponding to the most recently updated voltage value in the current data frame is used as the target timestamp. The system uses the established dynamic prediction model to predict or interpolate all eight voltage values ​​in the current data frame to a unified target timestamp, such as the measurement time of the last voltage value in the current data frame. These time-aligned prediction values ​​together constitute a correction data frame.

[0043] Finally, the corrected data frame is used to calculate the sensor's final accurate pose through the pose calculation of the positioning algorithm.

[0044] The core of this invention lies in establishing multiple implementation methods for dynamic prediction models, which are illustrated below through several embodiments.

[0045] Example 1: Prediction based on a linear extrapolation model

[0046] In this embodiment, the dynamic prediction model employs a linear extrapolation model. This model assumes that the voltage value of each measurement channel changes linearly between two consecutive data frames. The specific steps for generating the correction data frame are as follows:

[0047] First, for each measurement channel, the current voltage value in the current data frame is used. Historical voltage values ​​corresponding to historical data frames Calculate the first rate of change, i.e., the speed of voltage change. The first-order rate of change is mathematically expressed as... ;in, This is the time interval between two data frames.

[0048] Then, for each voltage value within the current data frame, there is a time difference between its measurement time and the target timestamp. Using the calculated first rate of change Perform linear extrapolation to obtain its predicted value at the target timestamp. The predicted value is mathematically expressed as .

[0049] Example 2: Prediction based on a nonlinear polynomial extrapolation model

[0050] In this embodiment, the dynamic prediction model employs a nonlinear polynomial extrapolation model to better fit the sensor's acceleration or deceleration motion. The specific steps for generating the correction data frame are as follows:

[0051] For each measurement channel, the system uses the current voltage value in the current data frame and the corresponding historical voltage values ​​in at least two historical data frames, i.e., voltage values ​​in at least three time series, to fit a second-order or higher-order polynomial function describing the change of voltage value over time using the Newton difference quotient method or the least squares method.

[0052] Then, the target timestamp is used as an input variable and substituted into the polynomial function to calculate the function value at the target timestamp, and the function value is used as the predicted value for the measurement channel.

[0053] Example 3: Prediction based on Kalman filter model

[0054] In this embodiment, the dynamic prediction model employs a Kalman filter. This method treats the signals from the eight measurement channels as a unified dynamic system. The implementation steps are as follows:

[0055] First, define a signal state vector that contains at least the current voltage values ​​of all eight measurement channels and their first-order rates of change, forming a sixteen-dimensional state vector.

[0056] Secondly, upon receiving each new data frame, the data frame is treated as a measurement observation, and a Kalman filter update step is performed to correct the estimated signal state vector. This update step uses a Kalman gain to fuse the difference between the measurement observation and a prior state estimate to obtain a more accurate posterior state estimate.

[0057] Then, the prediction step of the Kalman filter is performed to obtain the predicted value of the state vector at the target timestamp. This prediction step uses a state transition matrix to propagate the posterior state estimate from its corresponding time point to the target timestamp.

[0058] Finally, the voltage values ​​of all eight measurement channels are extracted from the predicted signal state vector and combined into a correction data frame.

[0059] Example 4: Prediction using an extended Kalman filter model based on pose state

[0060] In this embodiment, the dynamic prediction model employs an extended Kalman filter based on pose state. This model performs predictions directly in the physical pose space, theoretically resulting in higher accuracy. The specific steps for generating the correction data frame are as follows:

[0061] First, the extended Kalman filter model uses the sensor's position, velocity, attitude, and angular velocity as its internal state vectors.

[0062] Next, the prediction step of the extended Kalman filter is executed, using a kinematic model that describes the kinematic laws of the sensor to predict a predicted pose at the target timestamp.

[0063] Then, the theoretical value is inversely solved. The system inputs the predicted pose into a known forward model that describes the mapping relationship from sensor pose to induced voltage, such as a pre-calibrated magnetic dipole model, to calculate a complete, time-aligned theoretical voltage data frame.

[0064] Finally, this theoretical voltage data frame is used as the correction data frame.

[0065] In the scheme of Embodiment 4 above, a fusion step can be added. This step is performed after the theoretical value inverse solution, and specifically involves: obtaining one or more real measured voltage values ​​near the target timestamp, and replacing the corresponding components in the theoretical voltage data frame with them to generate a final corrected data frame that incorporates the latest real measurement information.

[0066] The above embodiments are preferred embodiments of the present invention, but the embodiments of the present invention are not limited to the above embodiments. Any changes, modifications, substitutions, combinations, or simplifications made without departing from the spirit and principle of the present invention shall be considered equivalent substitutions and shall be included within the protection scope of the present invention.

Claims

1. A method for predicting and correcting dynamic errors in an electromagnetic positioning system, wherein the electromagnetic positioning system includes a magnetic field generator and a sensor; the magnetic field generator operates in a time-division multiplexing mode, sequentially exciting eight signal sources according to a predetermined time sequence, and the response of each signal source constitutes an independent measurement channel; The sensor detects the voltage value corresponding to each measurement channel, and these voltage values ​​are collected and processed for pose calculation; characterized in that, The method includes the following steps: S1. Receive timing data frame: Receive a current data frame in real time. The current data frame consists of eight voltage values ​​measured sequentially according to the predetermined timing sequence. Each voltage value in the current data frame corresponds to a different measurement time. S2. Establish a dynamic prediction model: Based on the current data frame and at least one historical data frame stored in the memory, establish a dynamic prediction model to describe the evolution of voltage values ​​over time. S3. Generate a correction data frame: Take the timestamp corresponding to the most recently updated voltage value in the current data frame as the target timestamp, use the dynamic prediction model to calculate the predicted values ​​of all measurement channels at the same target timestamp within the measurement period of the current data frame, and combine these predicted values ​​into a correction data frame with perfectly aligned timestamps. S4. Perform pose calculation: Input the correction data frame into a positioning algorithm to calculate the pose of the sensor.

2. The method for dynamic error prediction and correction of an electromagnetic positioning system according to claim 1, characterized in that, The first implementation of the dynamic prediction model is linear extrapolation, as detailed below: For each measurement channel, a first-order rate of change is calculated using the current voltage value in the current data frame and the corresponding historical voltage value in the historical data frame. This first-order rate of change is mathematically expressed as... ,in, The first rate of change, The current voltage value, The historical voltage value, The time interval between two data frames; For each voltage value within the current data frame, a linear extrapolation is performed using the first-order rate of change based on the time difference between its measurement time and the target timestamp to obtain its predicted value at the target timestamp. This predicted value is mathematically expressed as... ,in, The predicted value, The time difference between the voltage value and the target timestamp.

3. The method for dynamic error prediction and correction of an electromagnetic positioning system according to claim 1, characterized in that, The second implementation of the dynamic prediction model is nonlinear polynomial extrapolation, as detailed below: For each measurement channel, using the voltage value in the current data frame and the corresponding voltage values ​​in at least two historical data frames, a second-order or higher-order polynomial function describing the change of voltage value over time is fitted using Newton's difference method or the least squares method. The target timestamp is used as an input variable and substituted into the polynomial function to calculate the function value at the target timestamp, and the function value is used as the predicted value.

4. The method for dynamic error prediction and correction of an electromagnetic positioning system according to claim 1, characterized in that, The third implementation of the dynamic prediction model is a Kalman filter, as detailed below: Define a signal state vector that contains at least the current voltage values ​​of all measurement channels and their first-order rates of change. Upon receiving each new data frame, the data frame is treated as a measurement observation, and the Kalman filter update step is executed to correct the estimated value of the signal state vector; wherein, the update step fuses the difference between the measurement observation and a prior state estimate through a Kalman gain to obtain a posterior state estimate; The prediction step of the Kalman filter is performed to obtain the predicted value of the signal state vector at the target timestamp; wherein the prediction step utilizes a state transition matrix to propagate the posterior state estimate from its corresponding time to the target timestamp, and the state transition matrix describes the physical or kinematic laws of the evolution of the signal state vector over time. The values ​​of all measurement channels are extracted from the predicted signal state vector and combined into the correction data frame.

5. The method for dynamic error prediction and correction of an electromagnetic positioning system according to claim 1, characterized in that, The fourth implementation of the dynamic prediction model is a pose-state-based extended Kalman filter. The extended Kalman filter uses the pose of the sensor and its rate of change as its internal pose state vector. The process of generating the correction data frame is as follows: a. Pose prediction: Perform the prediction step of the extended Kalman filter, that is, use a kinematic model that describes the kinematic laws of the sensor to predict a predicted pose at the target timestamp; b. Inverse Theoretical Value Solution: Input the predicted pose into a known forward model that describes the mapping relationship from sensor pose to voltage value, and calculate a complete time-aligned theoretical data frame; c. Generate correction data frame: Use the theoretical data frame as the correction data frame.

6. The method for dynamic error prediction and correction of an electromagnetic positioning system according to claim 5, characterized in that, In the process of generating the correction data frame, after step b, there is also a fusion step, which specifically involves: obtaining one or more real measured voltage values ​​at the current time near the target timestamp, and replacing the corresponding components in the theoretical data frame with them to generate a correction data frame that incorporates the real measured values.

Images