A method and system for automatic tracking of personnel trajectories in dangerous areas

Abstract

This invention discloses an automatic personnel trajectory tracking method and system in hazardous areas, relating to the field of navigation technology. It involves synchronously acquiring satellite data and inertial measurement data, and using a Kalman filter for state prediction based on the inertial data. The availability of satellite data is evaluated; if available, multi-system fusion calculations are performed to calculate the observation error; if unavailable, pseudo-observation errors are generated using historical inertial data and a pre-trained error prediction model. The observation error or pseudo-observation error is then used to correct the state prediction using a Kalman filter, outputting optimal position information, and the generated trajectory is visualized in real-time on an offline map. This method can continuously correct inertial navigation drift even when satellites are unavailable, achieving continuous and reliable personnel trajectory tracking.

Description

Technical Field

[0001] This invention relates to the field of navigation technology, and specifically to a method and system for automatically tracking the trajectory of people in dangerous areas. Background Technology

[0002] In hazardous areas (such as complex environments like wilderness exploration, emergency rescue, and mining operations), real-time monitoring of personnel location and trajectory information is crucial for ensuring safety. However, existing personnel positioning technologies still face numerous challenges in complex environments. Traditional satellite navigation systems rely on satellite signals for positioning. When personnel are in areas with severe signal blockage, such as underground, indoors, or canyons, satellite signals are prone to attenuation, multipath effects, or complete interruption, leading to a sharp decline in positioning accuracy or even the inability to provide location information. Systems relying solely on satellite navigation cannot guarantee continuous and stable trajectory tracking. To compensate for the intermittent failures of satellite positioning, inertial navigation systems have been introduced as an auxiliary solution. Although inertial navigation can provide continuous attitude and position changes through autonomous calculation without relying on external signals, its working principle determines its inherent limitations: zero-point drift, calibration errors, and integration calculations from sensors accumulate over time, causing position errors to continuously amplify. After a satellite signal interruption, inertial navigation alone can produce positioning deviations of hundreds of meters in a short period, completely distorting the tracking trajectory and rendering it unable to provide effective position information.

[0003] Currently, most methods for fusing satellite navigation and inertial navigation employ Kalman filtering for information fusion, using satellite data to correct inertial drift. However, these fusion algorithms typically rely on continuously available satellite observation data. Once satellite signals become unavailable for an extended period, the system degenerates into a pure inertial navigation system, unable to suppress the rapid increase in accumulated errors, leading to a precipitous drop in the reliability of the tracking system. Existing technologies lack effective mechanisms to compensate for inertial navigation errors in scenarios with prolonged satellite signal loss, causing personnel trajectory tracking to fail during critical periods, thus failing to meet the continuity and reliability requirements for personnel safety in hazardous areas. Summary of the Invention

[0004] The purpose of this invention is to solve the problem mentioned in the background art that existing integrated navigation methods cannot suppress the cumulative error of inertial navigation in scenarios where satellite signals are unavailable for a long time, and to propose an automatic tracking method and system for personnel trajectories in dangerous areas.

[0005] A first aspect of this invention provides a method for automatically tracking the trajectory of personnel in a dangerous area, the method comprising:

[0006] Simultaneous acquisition of satellite data and inertial measurement data; the satellite data includes BDS satellite data and GPS satellite data;

[0007] Based on the inertial measurement data, state prediction is performed through a Kalman filter prediction step to obtain the state prediction data of the target person; the state prediction data includes position, velocity, and attitude.

[0008] The availability of the satellite data was evaluated, and the evaluation results were obtained.

[0009] If the evaluation result is usable, the satellite data is subjected to multi-system fusion calculation to obtain the position calculation result, and the position calculation result is compared with the state prediction data to obtain the observation error data.

[0010] If the evaluation result is unusable, historical inertial data within a preset time window is extracted to construct a time-series feature matrix, which is then input into a pre-trained error prediction model to obtain pseudo-observation error data.

[0011] The observation error data or the pseudo observation error data are used as observations, and the state prediction data is corrected through the Kalman filter update step to output the optimal estimated current position information.

[0012] The current location information is mapped and visualized on an offline map to generate and display the movement trajectory of the target person in real time.

[0013] Optionally, the availability evaluation of the satellite data, to obtain the evaluation results, includes:

[0014] The number of currently visible BDS and GPS satellites is counted. If the number of visible satellites is less than 4, the satellite data is deemed unavailable.

[0015] Based on meeting the satellite quantity requirement, the signal strength of each BDS satellite and GPS satellite is obtained. Satellites with signal strength greater than the preset strength threshold are recorded as usable satellites. If the number of usable satellites is less than 4, the satellite data is determined to be unusable.

[0016] Calculate the geometric precision factor (PDOP) value of available satellites. If the PDOP value exceeds a preset precision threshold, the satellite data is determined to be unusable.

[0017] Optionally, the method further includes:

[0018] When satellite data is available, the observation error data at the current moment is used as a label, and the inertial data sequence within a preset time window before the current moment is used as a sample feature to form a supervised learning sample for online incremental learning of the error prediction model.

[0019] Optionally, when satellite data changes from unavailable to available, incremental learning judgment is performed, specifically including:

[0020] Based on the starting position information P0 of the most recent satellite data unavailability phase, and the first position information P1 and the second position information P2 at the current moment, the positioning error index value Ed is calculated:

[0021] ;

[0022] Wherein, the first position information is the current position information obtained by inertial navigation using pseudo-observation error data; the second position information is the current position information obtained by fusing and solving current satellite data.

[0023] If the positioning error index value is greater than the preset positioning error threshold, it is determined that incremental learning is required and a model update request is generated.

[0024] The average error of the error prediction model is calculated based on the navigation data of a subsequent short-term window.

[0025] If the average error is greater than the preset prediction error threshold, it is determined that incremental learning is required and a model update request is generated.

[0026] When a valid model update request exists, if the remaining battery power of the navigation device is greater than a preset battery power threshold and the number of supervised learning samples reaches a preset number threshold, then online incremental learning of the model is executed.

[0027] Optionally, the error prediction model is an LSTM model, whose input sequence time window length and hidden layer dimension are determined during offline training; specifically, during offline training, an improved whale optimization algorithm is used to optimize the model's hyperparameters; the specific optimization process includes:

[0028] Step 1: Randomly generate N initial whale individuals to obtain the initial population; each whale individual represents a hyperparameter combination;

[0029] Step two: For each individual whale in the initial population, generate a corresponding reverse individual to obtain the reverse population.

[0030] ;

[0031] in, It is the value of the initial whale individual i in the j-th dimension; UB j and LB j These are the upper and lower bounds of dimension j, respectively; It is the value of the reverse individual corresponding to the initial whale individual i in the j-th dimension;

[0032] Step 3: Calculate the fitness of each whale individual in the initial population and the reverse population, select the N whale individuals with the best fitness to form a new whale population, and determine the current global optimal solution;

[0033] Step 4: Update the positions of all individual whales according to the encirclement, predation and random search mechanisms of the whale optimization algorithm, and perform boundary constraint processing on the updated positions;

[0034] Step 5: Evaluate the fitness of the updated individual whales and update the global optimal solution;

[0035] Step 6: Determine if the termination condition has been met; if the termination condition has been met, output the hyperparameter combination corresponding to the global optimal solution; otherwise, return to step 4.

[0036] A second aspect of this invention provides an automatic personnel trajectory tracking system for dangerous areas, the system comprising:

[0037] The data acquisition module is used to simultaneously acquire satellite data and inertial measurement data; the satellite data includes BDS satellite data and GPS satellite data.

[0038] The state prediction module is used to predict the state of the target person based on the inertial measurement data and through a Kalman filter prediction step to obtain the state prediction data of the target person; the state prediction data includes position, velocity and attitude.

[0039] The availability evaluation module is used to evaluate the availability of the satellite data and obtain the evaluation results.

[0040] The actual observation module is used to perform multi-system fusion calculation on the satellite data if the evaluation result is usable, to obtain the position calculation result, and to compare and calculate the observation error data based on the position calculation result with the state prediction data.

[0041] The blind observation module is used to extract historical inertial data within a preset time window to construct a time-series feature matrix if the evaluation result is unavailable, and input it into a pre-trained error prediction model to obtain pseudo-observation error data.

[0042] The update and correction module is used to take the observation error data or the pseudo observation error data as observations, correct the state prediction data through the Kalman filter update step, and output the optimal estimated current position information.

[0043] The visualization module is used to map and visualize the current location information on an offline map, and generate and display the movement trajectory of the target person in real time.

[0044] Optionally, the usability evaluation module includes:

[0045] The visibility condition judgment module is used to count the number of currently visible BDS satellites and GPS satellites. If the number of visible satellites is less than 4, the satellite data is determined to be unavailable.

[0046] The availability condition judgment module is used to obtain the signal strength of each BDS satellite and GPS satellite based on the requirement of the number of satellites. Satellites with signal strength greater than a preset strength threshold are recorded as available satellites. If the number of available satellites is less than 4, the satellite data is determined to be unavailable.

[0047] The accuracy condition judgment module is used to calculate the geometric precision factor (PDOP) value of available satellites. If the PDOP value exceeds a preset accuracy threshold, the satellite data is determined to be unusable.

[0048] Optionally, the system further includes a sample management module, which, when satellite data is available, uses the observation error data at the current moment as a label and the inertial data sequence within a preset time window before the current moment as sample features to form a supervised learning sample for online incremental learning of the error prediction model.

[0049] Optionally, the system further includes a learning condition judgment module, used to perform incremental learning judgment when satellite data changes from unavailable to available; the learning condition judgment module includes:

[0050] The offset error degree judgment module is used to calculate the positioning error index value Ed based on the starting position information P0 of the most recent satellite data unavailability period, and the first position information P1 and the second position information P2 at the current moment.

[0051] ;

[0052] Wherein, the first position information is the current position information obtained by inertial navigation using pseudo-observation error data; the second position information is the current position information obtained by fusing and solving current satellite data.

[0053] If the positioning error index value is greater than the preset positioning error threshold, it is determined that incremental learning is required and a model update request is generated.

[0054] The prediction error level judgment module is used to calculate the average error of the error prediction model based on the navigation data of the next short window.

[0055] If the average error is greater than the preset prediction error threshold, it is determined that incremental learning is required and a model update request is generated.

[0056] The resource condition judgment module is used to perform online incremental learning of the model when there is a valid model update request, if the remaining power of the navigation device is greater than a preset power threshold and the number of supervised learning samples reaches a preset number threshold.

[0057] Optionally, the error prediction model is an LSTM model, whose input sequence time window length and hidden layer dimension are determined during offline training; specifically, the system further includes a hyperparameter optimization module, used to optimize the model's hyperparameters during offline training using an improved whale optimization algorithm; the hyperparameter optimization module includes:

[0058] The initial population generation module is used to randomly generate N initial whale individuals to obtain the initial population; each whale individual represents a combination of hyperparameters.

[0059] The reverse population generation module is used to generate a corresponding reverse individual for each whale individual in the initial population, thus obtaining a reverse population:

[0060] ;

[0061] in, It is the value of the initial whale individual i in the j-th dimension; UB j and LB j These are the upper and lower bounds of dimension j, respectively; It is the value of the reverse individual corresponding to the initial whale individual i in the j-th dimension;

[0062] The elite population determination module is used to calculate the fitness of each whale individual in the initial population and the reverse population, select the N whale individuals with the best fitness to form a new whale population, and determine the current global optimal solution.

[0063] The position update module is used to update the position of all individual whales according to the encirclement, predation and random search mechanisms of the whale optimization algorithm, and to perform boundary constraint processing on the updated position;

[0064] The fitness assessment module evaluates the fitness of the updated individual whales and updates the global optimal solution.

[0065] The iteration control module is used to determine whether the termination condition has been met; if the termination condition has been met, it outputs the hyperparameter combination corresponding to the global optimal solution; otherwise, it returns to the position update module.

[0066] The beneficial effects of this invention are:

[0067] 1. By employing a Kalman filter prediction-update mechanism that combines satellite positioning with inertial navigation, measurement noise and transient anomalies can be effectively suppressed, thereby improving the accuracy of position estimation.

[0068] 2. Simultaneously utilize BDS and GPS for multi-system fusion calculation, increase the redundancy of positioning sources, reduce the limitations of a single satellite system, and improve positioning reliability.

[0069] 3. The system evaluates the availability of satellite data and dynamically selects the most suitable data source and processing strategy based on whether the data is available or unavailable. When satellite data is unavailable, existing technologies, lacking observation input, can only execute the prediction step of the Kalman filter, causing the update step to completely stall and the system to degenerate into pure inertial navigation with continuously accumulating errors. This invention uses pseudo-observations generated by a pre-trained error prediction model as input for filtering and updating, thereby significantly mitigating the long-term drift problem of pure inertial navigation. Attached Figure Description

[0070] Figure 1 A flowchart illustrating an automatic personnel trajectory tracking method in a dangerous area, provided by an embodiment of the present invention;

[0071] Figure 2 This is an architecture diagram of an automatic personnel trajectory tracking system in a dangerous area, provided as an embodiment of the present invention. Detailed Implementation

[0072] To further illustrate the technical means and effects of the present invention in achieving its intended purpose, the following detailed description of the specific implementation methods, structures, features and effects of the present invention is provided in conjunction with the accompanying drawings and preferred embodiments.

[0073] This invention provides a method for automatically tracking the movement of people in hazardous areas. See also... Figure 1 , Figure 1 A flowchart illustrating an automatic personnel trajectory tracking method in a hazardous area, provided as an embodiment of the present invention. The method includes the following steps:

[0074] S101 simultaneously acquires satellite data and inertial measurement data.

[0075] S102, based on inertial measurement data, state prediction is performed through a Kalman filter prediction step to obtain the state prediction data of the target personnel.

[0076] S103, conduct an availability assessment of the satellite data and obtain the assessment results.

[0077] S104. If the evaluation result is usable, then observation error data is generated based on satellite data; otherwise, a pre-trained error prediction model is called to generate pseudo observation error data. Specifically, if the evaluation result is usable, multi-system fusion calculation is performed on the satellite data to obtain the position calculation result, and this result is compared with the state prediction data to calculate the observation error data; if the evaluation result is unusable, historical inertial data within a preset time window is extracted to construct a time-series feature matrix, which is then input into the pre-trained error prediction model to obtain pseudo observation error data.

[0078] S105 uses the observation error data or pseudo-observation error data as the observation, corrects the state prediction data through the Kalman filter update step, and outputs the optimal estimated current position information.

[0079] S106 maps and visualizes the current location information on an offline map, generating and displaying the movement trajectory of the target personnel in real time.

[0080] The satellite data includes BDS (BeiDou Navigation Satellite System) satellite data and GPS (Global Positioning System) satellite data; the state prediction data includes position, velocity and attitude information.

[0081] This invention provides an automatic personnel trajectory tracking method for hazardous areas. By analyzing historical inertial data using a pre-trained error prediction model, and generating pseudo-observation errors to replace actual satellite observations when satellite signals are unavailable, the method enables the Kalman filter to continuously correct inertial navigation drift, suppressing sharp drops in trajectory tracking accuracy. This achieves continuous and reliable automatic personnel trajectory tracking in complex environments, significantly enhancing safety assurance capabilities.

[0082] In one implementation, the navigation device is equipped with a GNSS receiving module, including a BeiDou satellite receiving module and a GPS satellite receiving module, as well as an inertial measurement unit (IMU). The IMU includes a three-axis accelerometer and a three-axis angular velocity sensor, etc., for synchronously acquiring satellite data and inertial measurement data, providing input for subsequent state prediction and observation error calculation.

[0083] In one implementation, step S103 involves evaluating the availability of satellite data, and the evaluation results include:

[0084] Step 1: Count the number of currently visible BDS and GPS satellites. If the number of visible satellites is less than 4, the satellite data is deemed unavailable.

[0085] Step two: Based on meeting the required number of satellites, acquire the signal strength of each BDS and GPS satellite. Satellites with signal strength greater than a preset strength threshold are recorded as usable satellites. If the number of usable satellites is less than four, the satellite data is deemed unusable. The preset strength threshold can be set to 35 dB-Hz.

[0086] Step 3: Calculate the geometric precision factor (PDOP) value of available satellites. If the PDOP value exceeds the preset precision threshold, the satellite data is deemed unusable. The preset precision threshold can be set to 6.

[0087] This implementation method sequentially determines the number of visible satellites, signal strength, and geometric precision factor (PDOP) to ensure that only satellite data with a sufficient number of reliable signals and reasonable spatial distribution participates in positioning. This significantly improves positioning accuracy and stability, enhances the system's robustness in environments with satellite signal obstruction or interference, and provides a reliable basis for inertial navigation correction or spurious observation generation.

[0088] In one implementation, the satellite data includes pseudorange observations, carrier phase observations, and satellite ephemeris information. In step S104, multi-system fusion calculation is performed on the satellite data to obtain the position calculation results, including:

[0089] Step 1: Perform time synchronization and coordinate unification processing on satellite data from different systems to establish a unified navigation reference coordinate system.

[0090] Step 2: Calculate the spatial position of each satellite at the observation time based on the satellite ephemeris. Combine pseudorange and carrier phase observations, and use weighted least squares method to perform multi-system fusion calculation. By integrating the observation weights and error characteristics of different systems, the three-dimensional position calculation result of the target personnel is obtained.

[0091] This approach fully utilizes redundant information from multiple satellite systems, improving the accuracy and reliability of three-dimensional position calculation.

[0092] In one embodiment, the method further includes:

[0093] When satellite data is available, the observation error data at the current moment is used as a label, and the inertial data sequence within a preset time window before the current moment is used as a sample feature to form a supervised learning sample for online incremental learning of the error prediction model.

[0094] Specifically, when satellite data becomes available again, a decision is made on whether to perform incremental learning, which includes:

[0095] Step 1: Based on the starting position information P0 of the most recent satellite data unavailability phase, and the first position information P1 and the second position information P2 at the current moment, calculate the positioning error index value Ed:

[0096] ;

[0097] The first position information is the current position information obtained by inertial navigation using pseudo-observation error data; the second position information is the current position information obtained by fusing and solving current satellite data. Let L2 norm be used to calculate the Euclidean distance between two points.

[0098] If the positioning error index value exceeds the preset positioning error threshold, incremental learning is required, and a model update request is generated. The positioning error threshold can be set between 0.1 and 0.3.

[0099] Step two: Calculate the average error of the error prediction model based on navigation data from a subsequent short window; for example, collect actual data and predicted data within two minutes to calculate the average error.

[0100] If the average error exceeds the preset prediction error threshold, incremental learning is deemed necessary, and a model update request is generated. The prediction error threshold can be set to 1 meter and can be dynamically adjusted according to the task accuracy.

[0101] Step 3: When a valid model update request exists, if the remaining battery power of the navigation device is greater than a preset battery threshold and the number of supervised learning samples reaches a preset quantity threshold, then online incremental learning of the model is executed. The preset battery threshold can be set to 50%. The preset quantity threshold can be set to 500.

[0102] This embodiment optimizes the error prediction model online in a timely manner, which more effectively corrects the drift of inertial navigation during the satellite unavailability phase. Furthermore, incremental learning is performed only when necessary and resources permit, effectively balancing algorithm performance and device energy consumption.

[0103] In one embodiment, the error prediction model can be an LSTM (Long Short-Term Memory) model, whose input sequence time window length and hidden layer dimension are determined during offline training. Specifically, during offline training, an improved whale optimization algorithm is used to optimize the model's hyperparameters; the optimization process includes:

[0104] Step 1: Randomly generate N initial whale individuals to obtain the initial population. Each whale individual represents a hyperparameter combination. N represents the population size, which can be between 30 and 50.

[0105] Step two: For each individual whale in the initial population, generate a corresponding reverse individual to obtain the reverse population.

[0106] ;

[0107] in, It is the value of the initial whale individual i in the j-th dimension; UB j and LB j These are the upper and lower bounds of dimension j, respectively; It is the value of the reverse individual corresponding to the initial whale individual i in the j-th dimension.

[0108] Step 3: Calculate the fitness of each whale individual in the initial and reverse populations, select the N whale individuals with the best fitness to form a new whale population, and determine the current global optimum. Specifically, train the model using the hyperparameter combination corresponding to each whale individual, and use the mean absolute error of the trained model on the validation set as the fitness. A lower fitness is better.

[0109] Step four: Update the positions of all individual whales according to the encirclement, predation and random search mechanisms of the whale optimization algorithm, and apply boundary constraints to the updated positions.

[0110] Step 5: Evaluate the fitness of the updated individual whales and update the global optimal solution.

[0111] Step 6: Determine if the termination condition has been met; if the termination condition has been met, output the hyperparameter combination corresponding to the global optimal solution; otherwise, return to step 4.

[0112] This embodiment utilizes the whale optimization algorithm system to search for hyperparameters such as the time window length and hidden layer dimension of the LSTM, eliminating the need for manual parameter tuning, avoiding reliance on experience, and improving the reliability of model performance.

[0113] In one implementation, the traditional whale optimization algorithm generates the initial population randomly, resulting in significant differences in individual quality, with some individuals potentially far from the optimal solution region. This invention introduces a reverse learning strategy, simultaneously constructing an initial population and a reverse population. The best N individuals are selected from these two complementary populations as the initial population. This increases the probability that the initial individuals are distributed near the optimal solution, reducing the interference of inferior initial individuals on the optimization process and decreasing the time cost of eliminating inferior individuals and exploring effective regions in the early iterations.

[0114] This invention provides an automatic personnel tracking system for hazardous areas. See also... Figure 2 , Figure 2 This is an architecture diagram of an automatic personnel trajectory tracking system in a hazardous area, provided as an embodiment of the present invention. The system includes:

[0115] The data acquisition module is used to simultaneously acquire satellite data and inertial measurement data; the satellite data includes BDS satellite data and GPS satellite data.

[0116] The state prediction module is used to predict the state of the target personnel based on inertial measurement data and through a Kalman filter prediction step.

[0117] The availability assessment module is used to assess the availability of satellite data and obtain the assessment results.

[0118] The actual observation module is used to perform multi-system fusion calculation on satellite data if the evaluation result is usable, to obtain the position calculation result, and to compare and calculate the observation error data based on the position calculation result with the state prediction data.

[0119] The blind observation module is used to extract historical inertial data within a preset time window to construct a time-series feature matrix if the evaluation result is unavailable. This matrix is ​​then input into a pre-trained error prediction model to obtain pseudo-observation error data.

[0120] The update and correction module is used to take observation error data or pseudo-observation error data as observations, correct the state prediction data through the Kalman filter update step, and output the optimal estimated current position information.

[0121] The visualization module is used to map and visualize the current location information on an offline map, and to generate and display the movement trajectory of the target personnel in real time.

[0122] The satellite data includes BDS satellite data and GPS satellite data; the state prediction data includes position, velocity and attitude information.

[0123] An automatic personnel trajectory tracking system for hazardous areas, based on embodiments of the present invention, analyzes historical inertial data using a pre-trained error prediction model. When satellite signals are unavailable, it generates pseudo-observation errors to replace actual satellite observations. This allows the Kalman filter to continuously correct inertial navigation drift, suppressing a sharp drop in trajectory tracking accuracy. Consequently, it achieves continuous and reliable automatic personnel trajectory tracking in complex environments, significantly enhancing safety assurance capabilities.

[0124] In one implementation, the usability evaluation module includes:

[0125] The visibility condition judgment module is used to count the number of currently visible BDS and GPS satellites. If the number of visible satellites is less than 4, the satellite data is determined to be unavailable.

[0126] The availability condition judgment module is used to obtain the signal strength of each BDS satellite and GPS satellite based on the requirement of the number of satellites. Satellites with signal strength greater than a preset strength threshold are recorded as available satellites. If the number of available satellites is less than 4, the satellite data is determined to be unavailable.

[0127] The accuracy condition judgment module is used to calculate the geometric precision factor (PDOP) value of available satellites. If the PDOP value exceeds the preset accuracy threshold, the satellite data is determined to be unusable.

[0128] In one embodiment, the system further includes a sample management module, which, when satellite data is available, uses the observation error data at the current moment as a label and the inertial data sequence within a preset time window before the current moment as a sample feature to form a supervised learning sample for online incremental learning of the error prediction model.

[0129] In one implementation, the system further includes a learning condition judgment module, used to perform incremental learning judgment when satellite data changes from unavailable to available. This learning condition judgment module includes:

[0130] The offset error degree judgment module is used to calculate the positioning error index value Ed based on the starting position information P0 of the most recent satellite data unavailability period, and the first position information P1 and the second position information P2 at the current moment.

[0131] ;

[0132] The first position information is the current position information obtained by inertial navigation using pseudo-observation error data; the second position information is the current position information obtained by fusing and solving current satellite data.

[0133] If the positioning error index value is greater than the preset positioning error threshold, it is determined that incremental learning is required, and a model update request is generated.

[0134] The prediction error level judgment module is used to calculate the average error of the error prediction model based on the navigation data of the next short window.

[0135] If the average error exceeds the preset prediction error threshold, incremental learning is required, and a model update request is generated.

[0136] The resource condition judgment module is used to perform online incremental learning of the model when there is a valid model update request, if the remaining power of the navigation device is greater than a preset power threshold and the number of supervised learning samples reaches a preset number threshold.

[0137] In one embodiment, the system further includes a hyperparameter optimization module for optimizing the model's hyperparameters using an improved whale optimization algorithm during offline training.

[0138] The hyperparameter optimization module includes an initial population generation module, a reverse population generation module, an elite population determination module, a position update module, a fitness evaluation module, and an iteration control module, which are used to implement steps one to six of the above hyperparameter optimization process, and will not be described in detail here.

[0139] The above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention in any way. Although the present invention has been disclosed above with reference to preferred embodiments, it is not intended to limit the present invention. Any person skilled in the art can make some modifications or alterations to the above-disclosed technical content to create equivalent embodiments without departing from the scope of the present invention. Any simple modifications, equivalent changes and alterations made to the above embodiments based on the technical essence of the present invention without departing from the scope of the present invention are within the scope of the claims of the present invention.

Claims

1. A method of automatically tracking a trajectory of a person in a hazardous area, characterized by, The method includes: Simultaneous acquisition of satellite data and inertial measurement data; the satellite data includes BDS satellite data and GPS satellite data; Based on the inertial measurement data, state prediction is performed through a Kalman filter prediction step to obtain the state prediction data of the target person; the state prediction data includes position, velocity, and attitude. The availability of the satellite data was evaluated, and the evaluation results were obtained. If the evaluation result is usable, the satellite data is subjected to multi-system fusion calculation to obtain the position calculation result, and the position calculation result is compared with the state prediction data to obtain the observation error data. If the evaluation result is unusable, historical inertial data within a preset time window is extracted to construct a time-series feature matrix, which is then input into a pre-trained error prediction model to obtain pseudo-observation error data. The observation error data or the pseudo observation error data are used as observations, and the state prediction data is corrected through the Kalman filter update step to output the optimal estimated current position information. The current location information is mapped and visualized on an offline map to generate and display the movement trajectory of the target person in real time. When satellite data is available, the observation error data at the current moment is used as a label, and the inertial data sequence within a time window of a preset length before the current moment is used as a sample feature to form a supervised learning sample for online incremental learning of the error prediction model. When satellite data becomes available after being unavailable, incremental learning is performed to determine its suitability, specifically including: Based on the starting position information P0 of the most recent satellite data unavailability phase, and the first position information P1 and the second position information P2 at the current moment, the positioning error index value Ed is calculated: ； Wherein, the first position information is the current position information obtained by inertial navigation using pseudo-observation error data; the second position information is the current position information obtained by fusing and solving current satellite data. If the positioning error index value is greater than the preset positioning error threshold, it is determined that incremental learning is required and a model update request is generated. The average error of the error prediction model is calculated based on the navigation data of a subsequent short-term window. If the average error is greater than the preset prediction error threshold, it is determined that incremental learning is required and a model update request is generated. When a valid model update request exists, if the remaining battery power of the navigation device is greater than a preset battery power threshold and the number of supervised learning samples reaches a preset number threshold, then online incremental learning of the model is executed.

2. The method of claim 1, wherein, The availability evaluation of the satellite data yields the following results: The number of currently visible BDS and GPS satellites is counted. If the number of visible satellites is less than 4, the satellite data is deemed unavailable. Based on meeting the satellite quantity requirement, the signal strength of each BDS satellite and GPS satellite is obtained. Satellites with signal strength greater than the preset strength threshold are recorded as usable satellites. If the number of usable satellites is less than 4, the satellite data is determined to be unusable. Calculate the geometric precision factor (PDOP) value of available satellites. If the PDOP value exceeds a preset precision threshold, the satellite data is determined to be unusable.

3. The method of claim 1, wherein, The error prediction model is an LSTM model, and the time window length and hidden layer dimension of its input sequence are determined during offline training. Specifically, during offline training, an improved whale optimization algorithm is used to optimize the model's hyperparameters. The specific optimization process includes: Step 1: Randomly generate N initial whale individuals to obtain the initial population; each whale individual represents a hyperparameter combination; Step two: For each individual whale in the initial population, generate a corresponding reverse individual to obtain the reverse population. ； wherein, is the value of the initial whale individual i in the jth dimension; UB j and LB j are the upper and lower limit values of the dimension j, respectively; is the value of the corresponding inverted individual of the initial whale individual i in the jth dimension; Step 3: Calculate the fitness of each whale individual in the initial population and the reverse population, select the N whale individuals with the best fitness to form a new whale population, and determine the current global optimal solution; Step 4: Update the positions of all individual whales according to the encirclement, predation and random search mechanisms of the whale optimization algorithm, and perform boundary constraint processing on the updated positions; Step 5: Evaluate the fitness of the updated individual whales and update the global optimal solution; Step 6: Determine if the termination condition has been met; if the termination condition has been met, output the hyperparameter combination corresponding to the global optimal solution; otherwise, return to step 4.

4. A personnel trajectory automatic tracking system in a dangerous area, characterized by, The system includes: The data acquisition module is used to simultaneously acquire satellite data and inertial measurement data; the satellite data includes BDS satellite data and GPS satellite data. The state prediction module is used to predict the state of the target person based on the inertial measurement data and through a Kalman filter prediction step to obtain the state prediction data of the target person; the state prediction data includes position, velocity and attitude. The availability evaluation module is used to evaluate the availability of the satellite data and obtain the evaluation results. The actual observation module is used to perform multi-system fusion calculation on the satellite data if the evaluation result is usable, to obtain the position calculation result, and to compare and calculate the observation error data based on the position calculation result with the state prediction data. The blind observation module is used to extract historical inertial data within a preset time window to construct a time-series feature matrix if the evaluation result is unavailable, and input it into a pre-trained error prediction model to obtain pseudo-observation error data. The update and correction module is used to take the observation error data or the pseudo observation error data as observations, correct the state prediction data through the Kalman filter update step, and output the optimal estimated current position information. The visualization module is used to map and visualize the current location information on an offline map, and generate and display the movement trajectory of the target person in real time. The sample management module is used to use the observation error data at the current moment as a label and the inertial data sequence within a preset time window before the current moment as sample features when satellite data is available, to form a supervised learning sample for online incremental learning of the error prediction model. The learning condition judgment module is used to perform incremental learning judgments when satellite data changes from unavailable to available; the learning condition judgment module includes: The offset error degree judgment module is used to calculate the positioning error index value Ed based on the starting position information P0 of the most recent satellite data unavailability period, and the first position information P1 and the second position information P2 at the current moment. ； Wherein, the first position information is the current position information obtained by inertial navigation using pseudo-observation error data; the second position information is the current position information obtained by fusing and solving current satellite data. If the positioning error index value is greater than the preset positioning error threshold, it is determined that incremental learning is required and a model update request is generated. The prediction error level judgment module is used to calculate the average error of the error prediction model based on the navigation data of the next short window. If the average error is greater than the preset prediction error threshold, it is determined that incremental learning is required and a model update request is generated. The resource condition judgment module is used to perform online incremental learning of the model when there is a valid model update request, if the remaining power of the navigation device is greater than a preset power threshold and the number of supervised learning samples reaches a preset number threshold.

5. The automatic tracking system of personnel trajectories in hazardous areas according to claim 4, characterized in that, The usability evaluation module includes: The visibility condition judgment module is used to count the number of currently visible BDS satellites and GPS satellites. If the number of visible satellites is less than 4, the satellite data is determined to be unavailable. The availability condition judgment module is used to obtain the signal strength of each BDS satellite and GPS satellite based on the requirement of the number of satellites. Satellites with signal strength greater than a preset strength threshold are recorded as available satellites. If the number of available satellites is less than 4, the satellite data is determined to be unavailable. The accuracy condition judgment module is used to calculate the geometric precision factor (PDOP) value of available satellites. If the PDOP value exceeds a preset accuracy threshold, the satellite data is determined to be unusable.

6. The automatic tracking system of personnel trajectories in hazardous areas according to claim 4, characterized in that, The error prediction model is an LSTM model, and the time window length and hidden layer dimension of its input sequence are determined during offline training. Specifically, the system also includes a hyperparameter optimization module, which is used to optimize the hyperparameters of the model using an improved whale optimization algorithm during offline training. The hyperparameter optimization module includes: The initial population generation module is used to randomly generate N initial whale individuals to obtain the initial population; each whale individual represents a combination of hyperparameters. The reverse population generation module is used to generate a corresponding reverse individual for each whale individual in the initial population, thus obtaining a reverse population: ； wherein, is the value of the initial whale individual i in the jth dimension; UB j and LB j are the upper and lower limit values of the dimension j, respectively; is the value of the jth dimension of the corresponding inverse individual of the initial whale individual i. The elite population determination module is used to calculate the fitness of each whale individual in the initial population and the reverse population, select the N whale individuals with the best fitness to form a new whale population, and determine the current global optimal solution. The position update module is used to update the position of all individual whales according to the encirclement, predation and random search mechanisms of the whale optimization algorithm, and to perform boundary constraint processing on the updated position; The fitness assessment module evaluates the fitness of the updated individual whales and updates the global optimal solution. The iteration control module is used to determine whether the termination condition has been met; if the termination condition has been met, it outputs the hyperparameter combination corresponding to the global optimal solution; otherwise, it returns to the position update module.

Images