Unmanned aerial vehicle trajectory tracking system based on multi-sensor fusion

Abstract

The invention discloses an unmanned aerial vehicle trajectory tracking system based on multi-sensor fusion, which relates to the technical field of unmanned system autonomous navigation and control, and comprises an unmanned aerial vehicle data acquisition module, an unmanned aerial vehicle feature extraction module, a trajectory prediction deviation scoring module, a trajectory dynamic correction module and a trajectory monitoring module. The unmanned aerial vehicle data acquisition module acquires and preprocesses unmanned aerial vehicle and GPS data, the unmanned aerial vehicle feature extraction module constructs a trajectory tracking feature sequence table, the trajectory prediction deviation scoring module outputs three-dimensional trajectory prediction coordinates and deviation scores, and the trajectory dynamic correction module generates correction parameters. Based on the judgment logic, the judgment logic is converted into an unmanned aerial vehicle correction instruction and attitude feedback quantity, and the track monitoring module constructs a visual interface. According to the method, the trajectory tracking precision of the unmanned aerial vehicle in a complex environment is remarkably improved, real-time detection and dynamic correction of abnormal deviation are realized, the state monitoring capability is enhanced by a three-dimensional visual interface, the trajectory tracking error is reduced, the misjudgment rate is reduced, and the flight safety and the control efficiency are guaranteed.

Description

Technical Field

[0001] The present invention relates to the technical field of autonomous navigation and control of unmanned systems, and in particular to a UAV trajectory tracking system based on multi-sensor fusion. Background Art

[0002] Drone trajectory tracking refers to the process of using advanced technical means to monitor, record and analyze the flight path of a drone in three-dimensional space in real time. This process usually combines global positioning systems, inertial navigation systems, radar and visual recognition technologies to obtain the drone's precise position, speed, altitude and attitude information. By processing and analyzing this data, the drone's flight trajectory can be constructed, thereby achieving comprehensive monitoring and management of the drone's flight status. Trajectory tracking not only helps ensure that the drone flies safely according to the predetermined route, but also can respond quickly in emergency situations, such as obstacle avoidance and return home, providing important guarantees for the drone's safe flight.

[0003] To address the issues of insufficient trajectory prediction accuracy and high misjudgment rate of abnormal deviation detection in complex drone tracking environments, existing technologies use a single sensor data-driven approach with fixed threshold rule detection. However, this approach can lead to sudden increases in trajectory prediction errors when GPS signals are blocked or subject to electromagnetic interference, and misjudgments triggered by high-frequency noise in inertial data. This can lead to inaccurate trajectory tracking, long delays in responding to abnormal deviations, and frequent mis-triggering of correction commands, causing flight oscillations. To address these issues, a drone trajectory tracking system based on multi-sensor fusion was proposed. Summary of the Invention

[0004] The purpose of the present invention is to provide a UAV trajectory tracking system based on multi-sensor fusion to solve the problems raised in the above background technology.

[0005] To solve the above technical problems, the technical solution adopted by the present invention is: a drone trajectory tracking system based on multi-sensor fusion, including a drone data acquisition module, a drone feature extraction module, a trajectory prediction deviation scoring module, a trajectory dynamic correction module and a trajectory monitoring module; The UAV data acquisition module collects and pre-processes UAV data and GPS data; The UAV feature extraction module extracts features from the pre-processed UAV data and GPS data to construct a trajectory tracking feature sequence table; The trajectory prediction deviation scoring module, combined with the trajectory tracking feature sequence table, constructs a trajectory prediction model and anomaly detection model, and outputs the three-dimensional trajectory prediction coordinates and trajectory deviation score; The trajectory dynamic correction module dynamically weights the trajectory prediction coordinates and the trajectory deviation score to obtain the UAV's trajectory correction parameters, builds a decision logic based on the trajectory deviation score, converts the trajectory correction parameters into UAV correction instructions, and generates attitude adjustment feedback; The trajectory monitoring module adjusts the feedback amount according to the posture and constructs a visual interface including the predicted trajectory line, the actual trajectory line and the safety envelope surface.

[0006] A further improvement of the technical solution of the present invention is that in the UAV data acquisition module, the acquisition and preprocessing process of UAV data and GPS data includes: The drone data includes the drone's three-axis acceleration and angular velocity data, and the GPS data includes GPS latitude and longitude coordinates and velocity scalar data; The inertial measurement unit (IMU) is fixed to the center of gravity plane of the drone, using a three-axis orthogonal layout and aligned with the body coordinate system. The IMU measures the three-axis acceleration data in the body coordinate system through a three-axis accelerometer and the angular velocity data around the body coordinate axis through a three-axis gyroscope. The collected three-axis acceleration and angular velocity data of the drone are cached by timestamp in a circular queue with a queue length of 50 groups. A GPS receiver is placed on top of the drone, with the antenna facing the sky in an unobstructed area. The GPS receiver receives satellite signals to calculate longitude and latitude coordinates and velocity scalar data. The collected longitude and latitude coordinates and velocity scalar data are parsed according to the protocol and stored in a first-in, first-out buffer with a buffer capacity of 20 groups. The inertial measurement unit is directly connected to the main control module of the drone trajectory tracking system through a four-wire serial bus. The sampling frequency is set to 200 Hz, and a shock-absorbing base is added. The global positioning system receiver uses an asynchronous serial communication protocol with a sampling frequency of 10 Hz, which is synchronized with the inertial measurement unit clock signal and has a built-in backup power supply. The high-frequency noise in the drone's three-axis acceleration and angular velocity data is suppressed using Kalman filtering technology. The state transfer matrix and observation noise covariance matrix are designed to iteratively predict and update the true values of acceleration and angular velocity. The filtered data is then timestamped and cached in a circular queue, outputting the noise-suppressed three-axis acceleration and angular velocity data. A sliding window mean filter is used to calculate the arithmetic mean of the longitude and latitude coordinates and speed scalar within the window. Data points with a direction difference of more than 30 degrees between adjacent points of the velocity vector are eliminated and replaced with the median within the window. At the same time, a buffer interpolation algorithm is enabled for the signal interruption period, and the missing values are linearly extrapolated using historical data to output the smoothed longitude and latitude coordinates and speed scalar data.

[0007] A further improvement of the technical solution of the present invention is that: in the UAV feature extraction module, the process of extracting features from the pre-processed UAV data includes: The time domain variance characteristics of the acceleration and angular velocity data of each axis in the preprocessed window are obtained respectively to obtain the three-axis acceleration variance and the three-axis angular velocity variance; The three-axis acceleration and angular velocity data of the same window are subjected to fast Fourier transform respectively to obtain the frequency domain amplitude spectrum. The three frequency points with the largest frequency domain amplitude are extracted, and the proportion of their energy to the total energy of all frequency points within the Nyquist frequency is analyzed. The frequency domain principal component energy ratio characteristics of the acceleration and angular velocity are output.

[0008] A further improvement of the technical solution of the present invention is that: in the UAV feature extraction module, the process of extracting features from the pre-processed GPS data and constructing a trajectory tracking feature sequence table includes: Based on the longitude and latitude coordinates of three consecutive points in the sliding window after preprocessing, convert their longitude and latitude coordinates into plane coordinates, analyze the inverse of the curvature radius of the arc formed by the three points, and output the trajectory curvature characteristics of the current window; Based on the pre-processed velocity scalar data, the heading angle of the sampling point is analyzed according to the difference of adjacent longitude and latitude coordinates, and the change in the modulus length of the velocity vector of adjacent sampling points is obtained, thereby obtaining the velocity change rate characteristics; The time domain variance, frequency domain principal component energy ratio, trajectory curvature and velocity change rate are aligned by timestamp interpolation and integrated into a trajectory tracking feature sequence list.

[0009] A further improvement of the technical solution of the present invention is that: in the trajectory prediction deviation scoring module, the process of building a trajectory prediction model in combination with the trajectory tracking feature sequence table and outputting the three-dimensional trajectory prediction coordinates includes: A stacked dual-memory unit layer based on the long short-term memory network architecture is used. The first and second layers each contain 32 memory units. Using 10 sets of data as time windows, the temporal variance features and velocity change rate features are combined into a 7-dimensional prediction sequence. Each dimension of the feature is then subjected to maximum-minimum normalization to construct a trajectory prediction model. The prediction sequence is input into the long short-term memory network model in sequence according to the time step. The internal forget gate and input gate control the information flow. The memory unit stores the long-term dependencies between time series features. The hidden state output by the output gate at the last time step is mapped into the three-dimensional trajectory prediction coordinates through the fully connected layer. The three-dimensional trajectory prediction coordinates include the longitude coordinate, latitude coordinate, and altitude coordinate of the drone. The mean square error is used to measure the difference between the predicted coordinates of the three-dimensional trajectory and the true coordinates of the trajectory. The partial derivatives of the loss function with respect to the weights and biases of each layer are calculated by the gradient descent method. The maximum number of iterations is set to 5000, and the training is terminated when the maximum number of iterations is reached.

[0010] A further improvement of the technical solution of the present invention is that: in the trajectory prediction deviation scoring module, the process of building an anomaly detection model by combining the trajectory tracking feature sequence table and outputting the trajectory deviation score includes: Align the frequency domain principal component energy ratio features and trajectory curvature features by timestamp, construct a single sample input vector, and perform zero-mean normalization on each dimension of the feature; The isolation forest algorithm is used to build an anomaly detection model. 80% of the historical data is randomly selected as the training set and 20% as the validation set. The isolation trees are trained independently. 128 groups of samples are randomly selected from each isolation tree as the training subset. The maximum depth is set to 8. A split point is selected for each feature dimension. The data is recursively divided until the sample is isolated and the depth limit is reached. 100 isolation trees are generated to form an isolation forest. Based on a single sample input vector, its path length in each isolation tree is obtained. The path length of the isolation tree and the path length benchmark value constructed are comprehensively calculated. , calculate the trajectory deviation score , the calculation process is as follows: ; ; ; in, For isolated trees The mean of is the total number of training samples, is the harmonic number, ; Based on the statistical deviation score distribution of the validation set, calculate the mean and standard deviation, and set the abnormal threshold It is the sum of the mean and three times the standard deviation. When the trajectory deviation score is greater than the abnormal threshold, it is determined to be trajectory deviation. Every 24 hours, 10 isolated trees are retrained with new data to update the path length benchmark value.

[0011] A further improvement of the technical solution of the present invention is that in the trajectory dynamic correction module, the process of dynamically weighting the trajectory prediction coordinates and the trajectory deviation score to obtain the trajectory correction parameters of the drone includes: Dynamically set the predicted coordinate weight based on the ratio of the trajectory deviation score to the anomaly threshold and deviation compensation weights , and the sum of the predicted coordinate weight and the deviation compensation weight is 1. The calculation process is as follows: ; ; Among them, is the regulating factor; when , based on the trajectory curvature characteristics and speed change rate characteristics Calculate direction compensation and height compensation , thereby obtaining the trajectory correction parameters of the UAV, which include the direction correction amount , speed correction and altitude correction , the calculation process is as follows: ; ; ; ; ; in, Curvature-velocity coupling coefficient, is the safety envelope height, is the height tolerance factor, To predict the height, To predict the heading angle, To predict the speed, is the speed correction gain.

[0012] A further improvement of the technical solution of the present invention is that: in the trajectory dynamic correction module, the decision logic is constructed based on the trajectory deviation score, and the process of converting the trajectory correction parameters into the UAV correction instructions includes: When three times in a row , it is judged as continuous deviation, triggering the correction instruction, when it is no more than three times in a row , only records the log, does not trigger the control instruction, the correction instruction includes the direction correction instruction , speed correction command and altitude correction commands , the calculation process is as follows: ; ; ; ; in, is the proportionality coefficient, is the integration coefficient, is the heading angle deviation for the last three times, is the speed deviation, is the current actual speed, is the speed tracking coefficient, is the height tracking coefficient.

[0013] A further improvement of the technical solution of the present invention is that: in the trajectory dynamic correction module, the process of generating the posture adjustment feedback includes: Through the proportional-integral-derivative controller, based on the correction 、 and The difference between the corresponding actual measurement value is used to obtain the heading angle error, speed error and altitude error in real time. The control weight is dynamically adjusted by combining the proportional term, integral term and differential term. The differential term suppresses high-frequency oscillation, and the integral term eliminates steady-state deviation. The attitude adjustment feedback is output. The attitude adjustment feedback includes the heading servo adjustment amount, throttle adjustment amount and elevator adjustment amount.

[0014] A further improvement of the technical solution of the present invention is that in the trajectory monitoring module, the process of constructing a visual interface including a predicted trajectory line, an actual trajectory line, and a safety envelope surface according to the posture adjustment feedback amount includes: The system integrates the trajectory prediction coordinates output by the trajectory prediction model, the smooth trajectory of the global positioning system, and the safety envelope parameters. It uses plane projection transformation to map the latitude and longitude coordinates into three-dimensional space coordinates, constructing a red predicted trajectory line, a blue actual trajectory line, and a translucent green cylindrical safety envelope. The interface refreshes at a 0.1-second cycle, synchronously displays the real-time attitude adjustment feedback value, and supports zooming, rotation, and trajectory feature parameter access. When the actual trajectory point continues to exceed the safety envelope, the secondary warning mechanism is triggered. The primary warning marks the deviation point and pops up a prompt. The secondary warning forces the control mode to switch. The operator can manually adjust the envelope tolerance parameters and click on the trajectory line to view the time domain variance and the frequency domain principal component energy ratio.

[0015] Due to the adoption of the above technical solution, the present invention has the following technical advancements compared to the prior art: 1. The present invention provides a UAV trajectory tracking system based on multi-sensor fusion. By fusing the spatiotemporal alignment data of an inertial measurement unit and a global positioning system, combined with the temporal modeling capability of a long short-term memory network, it significantly improves the trajectory prediction accuracy in complex electromagnetic interference and GPS signal shielding scenarios, and reduces the measured prediction error compared to traditional single-sensor solutions.

[0016] 2. The present invention provides a UAV trajectory tracking system based on multi-sensor fusion. It adopts the isolation forest algorithm to analyze the cross-modal characteristics of frequency domain energy proportion and trajectory curvature, and combines the deviation score continuous over-threshold judgment logic. The anomaly detection misjudgment rate is lower than that of the traditional fixed threshold method, which improves the response speed and effectively avoids flight oscillations caused by erroneous triggering of correction instructions.

[0017] 3. The present invention provides a UAV trajectory tracking system based on multi-sensor fusion, which generates direction, speed and altitude correction parameters through a dynamic weighted fusion layer, and constructs a three-dimensional geographic information visualization interface to map the predicted trajectory, actual trajectory and safety envelope in real time, thereby realizing full-dimensional monitoring of trajectory status and human intervention channels, thereby improving the operator's decision-making efficiency and extending the system's safety fault tolerance time. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments. Obviously, the drawings described below are only some embodiments described in the present invention. For ordinary technicians in this field, other drawings can also be obtained based on these drawings.

[0019] Figure 1 A block diagram of the present invention. DETAILED DESCRIPTION

[0020] To make the objectives, technical solutions, and advantages of the embodiments of the present invention more clear, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. All other embodiments obtained by ordinary technicians in this field based on the embodiments of the present invention without making any creative efforts shall fall within the scope of protection of the present invention.

[0021] Examples, such as Figure 1 As shown, the present invention provides a UAV trajectory tracking system based on multi-sensor fusion, including a UAV data acquisition module, a UAV feature extraction module, a trajectory prediction deviation scoring module, a trajectory dynamic correction module and a trajectory monitoring module; The UAV data acquisition module collects and pre-processes UAV data and GPS data. The UAV data includes the UAV's three-axis acceleration and angular velocity data, and the GPS data includes the GPS latitude and longitude coordinates and speed scalar data. The inertial measurement unit is fixed on the UAV's center of gravity plane, and a three-axis orthogonal layout is adopted to align with the body coordinate system. The inertial measurement unit measures the three-axis acceleration data in the body coordinate system through a three-axis accelerometer, and measures the angular velocity data around the body coordinate axis through a three-axis gyroscope. The collected UAV three-axis acceleration and angular velocity data are cached according to the timestamp into a circular queue with a queue length of 50 groups. The global positioning system receiver is placed on the top of the UAV with the antenna facing the sky in an unobstructed area. The global positioning system receiver receives satellite signals to solve the longitude and latitude coordinates and speed scalar data. The collected longitude and latitude coordinates and speed scalar data are parsed according to the protocol and stored in a first-in-first-out buffer with a buffer capacity of 20 groups. Among them, the inertial measurement unit is connected to the UAV through a four-wire serial bus The trajectory tracking system is directly connected to the main control module, with a sampling frequency set to 200 Hz and an additional shock-absorbing base. The global positioning system receiver uses an asynchronous serial communication protocol with a sampling frequency of 10 Hz, synchronized with the inertial measurement unit clock signal, and has a built-in backup power supply. The high-frequency noise of the drone's three-axis acceleration and angular velocity data is suppressed through Kalman filtering technology. The state transfer matrix and observation noise covariance matrix are designed to iteratively predict and update the true values of acceleration and angular velocity. The filtered data is synchronously aligned by timestamp and cached in a circular queue, and the noise-suppressed three-axis acceleration and angular velocity data is output. A sliding window mean filter is used to calculate the arithmetic mean of the longitude and latitude coordinates and velocity scalars within the window. Data points with a direction difference of more than 30 degrees between adjacent points of the velocity vector are eliminated and replaced with the median within the window. At the same time, the buffer interpolation algorithm is enabled during the signal interruption period, and the missing values are linearly extrapolated using historical data to output smoothed longitude and latitude coordinates and velocity scalar data. The UAV feature extraction module extracts features from the preprocessed UAV data and GPS data, constructs a trajectory tracking feature sequence table, obtains the time domain variance characteristics of the acceleration and angular velocity data of each axis in the preprocessed window, obtains the three-axis acceleration variance and three-axis angular velocity variance, performs fast Fourier transform on the three-axis acceleration and angular velocity data in the same window, obtains the frequency domain amplitude spectrum, extracts the three frequency points with the largest frequency domain amplitude, analyzes the proportion of their energy to the sum of the energy of all frequency points within the Nyquist frequency, and outputs the frequency domain principal components of acceleration and angular velocity. The energy proportion feature is based on the longitude and latitude coordinates of three consecutive points in the sliding window after preprocessing. The longitude and latitude coordinates are converted into plane coordinates, the inverse of the curvature radius of the arc formed by the three points is analyzed, and the trajectory curvature feature of the current window is output. Based on the preprocessed velocity scalar data, the heading angle of the sampling point is analyzed according to the difference of adjacent longitude and latitude coordinates, and the modulus change of the velocity vector of adjacent sampling points is obtained to obtain the velocity change rate feature. The time domain variance, frequency domain principal component energy proportion, trajectory curvature and velocity change rate are interpolated and aligned according to the timestamp and integrated into a trajectory tracking feature sequence list; The trajectory prediction deviation scoring module combines the trajectory tracking feature sequence list to build a trajectory prediction model and anomaly detection model, outputs three-dimensional trajectory prediction coordinates and trajectory deviation scores, and adopts a stacked dual memory unit layer based on the long short-term memory network architecture. The first and second layers each contain 32 memory units. With 10 groups of data as the time window, the time domain variance features and the speed change rate features are combined into a 7-dimensional prediction sequence, and each dimension feature is subjected to maximum-minimum normalization to build a trajectory prediction model. The prediction sequence is input into the long short-term memory network model in sequence according to the time step. The information flow is controlled by the internal forget gate and input gate. The memory unit stores the long-term dependency between the time series features. The hidden state output by the output gate at the last time step is mapped to the three-dimensional trajectory prediction coordinates through the fully connected layer. The three-dimensional trajectory prediction coordinates include the longitude coordinates, dimensional coordinates and altitude coordinates of the drone. The mean square error is used to measure the three-dimensional trajectory. The difference between the predicted coordinates and the true coordinates of the trajectory is calculated by the gradient descent method. The partial derivatives of the loss function with respect to the weights and biases of each layer are calculated. The maximum number of iterations is set to 5000 times. The training is terminated when the maximum number of iterations is reached. The frequency domain principal component energy ratio features and the trajectory curvature features are aligned by timestamp, a single sample input vector is constructed, and each dimension feature is zero-mean normalized. The isolation forest algorithm is used to construct an anomaly detection model. 80% of the historical data is randomly selected as the training set and 20% as the validation set. The isolated trees are trained independently. 128 groups of samples are randomly selected from each isolated tree as the training subset. The maximum depth is set to 8. The split point is selected for each feature dimension. The data is recursively divided until the sample is isolated and the depth limit is reached. 100 isolated trees are generated to form an isolation forest. Based on the single sample input vector, its path length in each isolated tree is obtained. The path length of the isolated tree and the path length benchmark value constructed are comprehensively constructed. , calculate the trajectory deviation score , the calculation process is as follows: ; ; ; in, For isolated trees The mean of is the total number of training samples, is the harmonic number, , based on the statistical deviation score distribution of the validation set, calculate the mean and standard deviation, and set the abnormal threshold It is the sum of the mean and three times the standard deviation. When the trajectory deviation score is greater than the abnormal threshold, it is determined to be trajectory deviation. Every 24 hours, 10 isolated trees are retrained with new data to update the path length benchmark value. The trajectory dynamic correction module dynamically weights the trajectory prediction coordinates and trajectory deviation scores to obtain the trajectory correction parameters of the drone. Based on the trajectory deviation score, the judgment logic is constructed to convert the trajectory correction parameters into drone correction instructions and generate attitude adjustment feedback. The predicted coordinate weight is dynamically set according to the ratio of the trajectory deviation score to the abnormal threshold. and deviation compensation weights , and the sum of the predicted coordinate weight and the deviation compensation weight is 1. The calculation process is as follows: ; ; in, is the adjustment factor, when , based on the trajectory curvature characteristics and speed change rate characteristics Calculate direction compensation and height compensation , thereby obtaining the trajectory correction parameters of the UAV, which include the direction correction amount , speed correction and altitude correction , the calculation process is as follows: ; ; ; ; ; in, Curvature-velocity coupling coefficient, is the safety envelope height, is the height tolerance factor, To predict the height, To predict the heading angle, To predict the speed, For speed correction gain, when three times in a row , it is judged as continuous deviation, triggering the correction instruction, when it is no more than three times in a row , only records the log, does not trigger the control instruction, the correction instruction includes the direction correction instruction , speed correction command and altitude correction commands , the calculation process is as follows: ; ; ; ; in, is the proportionality coefficient, is the integration coefficient, is the heading angle deviation for the last three times, is the speed deviation, is the current actual speed, is the speed tracking coefficient, is the height tracking coefficient, which is based on the correction quantity through the proportional-integral-derivative controller. 、 and The difference between the corresponding actual measurement value and the heading angle error, speed error and altitude error are obtained in real time. The control weight is dynamically adjusted by combining the proportional term, integral term and differential term. The differential term suppresses high-frequency oscillation, and the integral term eliminates steady-state deviation. The attitude adjustment feedback is output. The attitude adjustment feedback includes the heading servo adjustment amount, the throttle adjustment amount and the elevator adjustment amount; The trajectory monitoring module constructs a visual interface that includes the predicted trajectory line, actual trajectory line and safety envelope surface based on the attitude adjustment feedback. It integrates the trajectory prediction coordinates output by the trajectory prediction model, the smooth trajectory of the global positioning system and the safety envelope surface parameters, and uses plane projection transformation to map the latitude and longitude coordinates into three-dimensional space coordinates to construct a red predicted trajectory line, a blue actual trajectory line and a translucent green cylindrical safety envelope surface. The interface refreshes at a cycle of 0.1 seconds and synchronously displays the real-time attitude adjustment feedback value. It supports scaling, rotation and trajectory feature parameter access functions. When the actual trajectory point continues to exceed the safety envelope surface, the secondary warning mechanism is triggered. The primary warning marks the deviation point and pops up a prompt. The secondary warning forces the switching of the control mode. The operator can manually adjust the envelope surface tolerance parameters and click on the trajectory line to view the time domain variance and the frequency domain principal component energy ratio.

[0022] First, the drone data acquisition module acquires and preprocesses the three-axis acceleration and angular velocity data from the inertial measurement unit (IMU) and the longitude, latitude, and velocity data from the global positioning system (GPS) in real time. Subsequently, the UAV feature extraction module extracts temporal variance, frequency-domain principal component energy ratio, trajectory curvature, and velocity change rate features to construct a time-aligned trajectory tracking feature sequence list. Next, the trajectory prediction deviation scoring module synchronously runs a long short-term memory (LSTM) trajectory prediction model and an isolation forest anomaly detection model based on the feature sequence list, outputting three-dimensional predicted coordinates and a real-time deviation score for the next 0.5 seconds. Based on this, the trajectory dynamic correction module dynamically assigns weight coefficients for the predicted coordinates and deviation compensation using a sigmoid function. When the deviation score exceeds a threshold three times consecutively, the fused heading, velocity, and altitude correction parameters are converted into proportional-integral control commands. Finally, the trajectory monitoring module maps the correction commands into attitude feedback. Combining the predicted trajectory, actual trajectory, and dynamic safety envelope, it enables full-dimensional visual monitoring of the flight status within a three-dimensional geographic information interface.

[0023] The above description is merely a specific embodiment of the present application, but the scope of protection of the present application is not limited thereto. Any changes or substitutions that can be easily conceived by a person skilled in the art within the technical scope disclosed in this application should be included in the scope of protection of this application. Therefore, the scope of protection of this application should be based on the scope of protection of the claims.

Claims

1. A UAV trajectory tracking system based on multi-sensor fusion, characterized by: It includes UAV data acquisition module, UAV feature extraction module, trajectory prediction deviation scoring module, trajectory dynamic correction module and trajectory monitoring module; The UAV data acquisition module collects and pre-processes UAV data and GPS data; The UAV feature extraction module extracts features from the pre-processed UAV data and GPS data to construct a trajectory tracking feature sequence table; The trajectory prediction deviation scoring module, combined with the trajectory tracking feature sequence table, constructs a trajectory prediction model and anomaly detection model, and outputs the three-dimensional trajectory prediction coordinates and trajectory deviation score; The trajectory dynamic correction module dynamically weights the trajectory prediction coordinates and the trajectory deviation score to obtain the UAV's trajectory correction parameters, builds a decision logic based on the trajectory deviation score, converts the trajectory correction parameters into UAV correction instructions, and generates attitude adjustment feedback; The trajectory monitoring module adjusts the feedback amount according to the posture and constructs a visual interface including the predicted trajectory line, the actual trajectory line and the safety envelope surface.

2. The UAV trajectory tracking system based on multi-sensor fusion according to claim 1, characterized in that: In the UAV data acquisition module, the acquisition and preprocessing process of UAV data and GPS data includes: The drone data includes the drone's three-axis acceleration and angular velocity data, and the GPS data includes GPS latitude and longitude coordinates and velocity scalar data; The inertial measurement unit (IMU) is fixed to the center of gravity plane of the drone, using a three-axis orthogonal layout and aligned with the body coordinate system. The IMU measures the three-axis acceleration data in the body coordinate system through a three-axis accelerometer and the angular velocity data around the body coordinate axis through a three-axis gyroscope. The collected three-axis acceleration and angular velocity data of the drone are cached by timestamp in a circular queue with a queue length of 50 groups. A GPS receiver is placed on top of the drone, with the antenna facing the sky in an unobstructed area. The GPS receiver receives satellite signals to calculate longitude and latitude coordinates and velocity scalar data. The collected longitude and latitude coordinates and velocity scalar data are parsed according to the protocol and stored in a first-in, first-out buffer with a buffer capacity of 20 groups. The inertial measurement unit is directly connected to the main control module of the drone trajectory tracking system through a four-wire serial bus. The sampling frequency is set to 200 Hz, and a shock-absorbing base is added. The global positioning system receiver uses an asynchronous serial communication protocol with a sampling frequency of 10 Hz, which is synchronized with the inertial measurement unit clock signal and has a built-in backup power supply. The high-frequency noise in the drone's three-axis acceleration and angular velocity data is suppressed using Kalman filtering technology. The state transfer matrix and observation noise covariance matrix are designed to iteratively predict and update the true values of acceleration and angular velocity. The filtered data is then timestamped and cached in a circular queue, outputting the noise-suppressed three-axis acceleration and angular velocity data. A sliding window mean filter is used to calculate the arithmetic mean of the longitude and latitude coordinates and speed scalar within the window. Data points with a direction difference of more than 30 degrees between adjacent points of the velocity vector are eliminated and replaced with the median within the window. At the same time, a buffer interpolation algorithm is enabled for the signal interruption period, and the missing values are linearly extrapolated using historical data to output the smoothed longitude and latitude coordinates and speed scalar data.

3. The UAV trajectory tracking system based on multi-sensor fusion according to claim 2, characterized in that: In the UAV feature extraction module, the process of extracting features from the pre-processed UAV data includes: The time domain variance characteristics of the acceleration and angular velocity data of each axis in the preprocessed window are obtained respectively to obtain the three-axis acceleration variance and the three-axis angular velocity variance; The three-axis acceleration and angular velocity data of the same window are subjected to fast Fourier transform respectively to obtain the frequency domain amplitude spectrum. The three frequency points with the largest frequency domain amplitude are extracted, and the proportion of their energy to the total energy of all frequency points within the Nyquist frequency is analyzed. The frequency domain principal component energy ratio characteristics of the acceleration and angular velocity are output.

4. The UAV trajectory tracking system based on multi-sensor fusion according to claim 3 is characterized by: In the UAV feature extraction module, the process of extracting features from the pre-processed GPS data and constructing a trajectory tracking feature sequence table includes: Based on the longitude and latitude coordinates of three consecutive points in the sliding window after preprocessing, convert their longitude and latitude coordinates into plane coordinates, analyze the inverse of the curvature radius of the arc formed by the three points, and output the trajectory curvature characteristics of the current window; Based on the pre-processed velocity scalar data, the heading angle of the sampling point is analyzed according to the difference of adjacent longitude and latitude coordinates, and the change in the modulus length of the velocity vector of adjacent sampling points is obtained, thereby obtaining the velocity change rate characteristics; The time domain variance, frequency domain principal component energy ratio, trajectory curvature and velocity change rate are aligned by timestamp interpolation and integrated into a trajectory tracking feature sequence list.

5. The UAV trajectory tracking system based on multi-sensor fusion according to claim 4 is characterized by: In the trajectory prediction deviation scoring module, the process of building a trajectory prediction model and outputting three-dimensional trajectory prediction coordinates in combination with the trajectory tracking feature sequence table includes: A stacked dual-memory unit layer based on the long short-term memory network architecture is used. The first and second layers each contain 32 memory units. Using 10 sets of data as time windows, the temporal variance features and velocity change rate features are combined into a 7-dimensional prediction sequence. Each dimension of the feature is then subjected to maximum-minimum normalization to construct a trajectory prediction model. The prediction sequence is input into the long short-term memory network model in sequence according to the time step. The internal forget gate and input gate control the information flow. The memory unit stores the long-term dependencies between time series features. The hidden state output by the output gate at the last time step is mapped into the three-dimensional trajectory prediction coordinates through the fully connected layer. The three-dimensional trajectory prediction coordinates include the longitude coordinate, latitude coordinate, and altitude coordinate of the drone. The mean square error is used to measure the difference between the predicted coordinates of the three-dimensional trajectory and the true coordinates of the trajectory. The partial derivatives of the loss function with respect to the weights and biases of each layer are calculated by the gradient descent method. The maximum number of iterations is set to 5000, and the training is terminated when the maximum number of iterations is reached.

6. The UAV trajectory tracking system based on multi-sensor fusion according to claim 5, characterized in that: In the trajectory prediction deviation scoring module, the process of building an anomaly detection model and outputting a trajectory deviation score by combining the trajectory tracking feature sequence table includes: Align the frequency domain principal component energy ratio features and trajectory curvature features by timestamp, construct a single sample input vector, and perform zero-mean normalization on each dimension of the feature; The isolation forest algorithm is used to build an anomaly detection model. 80% of the historical data is randomly selected as the training set and 20% as the validation set. The isolation trees are trained independently. 128 groups of samples are randomly selected from each isolation tree as the training subset. The maximum depth is set to 8. A split point is selected for each feature dimension. The data is recursively divided until the sample is isolated and the depth limit is reached. 100 isolation trees are generated to form an isolation forest. Based on a single sample input vector, its path length in each isolation tree is obtained. The path length of the isolation tree and the path length benchmark value constructed are comprehensively calculated. , calculate the trajectory deviation score , the calculation process is as follows: ； ； ； in, For isolated trees The mean of is the total number of training samples, is the harmonic number, ; Based on the statistical deviation score distribution of the validation set, calculate the mean and standard deviation, and set the abnormal threshold It is the sum of the mean and three times the standard deviation. When the trajectory deviation score is greater than the abnormal threshold, it is determined to be trajectory deviation. Every 24 hours, 10 isolated trees are retrained with new data to update the path length benchmark value.

7. The UAV trajectory tracking system based on multi-sensor fusion according to claim 6, characterized in that: In the trajectory dynamic correction module, the process of dynamically weighting the trajectory prediction coordinates and the trajectory deviation score to obtain the trajectory correction parameters of the drone includes: Dynamically set the predicted coordinate weight based on the ratio of the trajectory deviation score to the anomaly threshold and deviation compensation weights , and the sum of the predicted coordinate weight and the deviation compensation weight is 1. The calculation process is as follows: ； ； in, is the regulating factor; when , based on the trajectory curvature characteristics and speed change rate characteristics Calculate direction compensation and height compensation , thereby obtaining the trajectory correction parameters of the UAV, which include the direction correction amount , speed correction and altitude correction , the calculation process is as follows: ； ； ； ； ； in, Curvature-velocity coupling coefficient, is the safety envelope height, is the height tolerance factor, To predict the height, To predict the heading angle, To predict the speed, is the speed correction gain.

8. The UAV trajectory tracking system based on multi-sensor fusion according to claim 7, characterized in that: In the trajectory dynamic correction module, the process of building a decision logic based on the trajectory deviation score and converting the trajectory correction parameters into UAV correction instructions includes: When three times in a row , it is judged as continuous deviation, triggering the correction instruction, when it is no more than three times in a row , only records the log, does not trigger the control instruction, the correction instruction includes the direction correction instruction , speed correction command and altitude correction commands , the calculation process is as follows: ； ； ； ； in, is the proportionality coefficient, is the integration coefficient, is the heading angle deviation for the last three times, is the speed deviation, is the current actual speed, is the speed tracking coefficient, is the height tracking coefficient.

9. The UAV trajectory tracking system based on multi-sensor fusion according to claim 8, characterized in that: In the trajectory dynamic correction module, the process of generating the posture adjustment feedback includes: Through the proportional-integral-derivative controller, based on the correction 、 and The difference between the corresponding actual measurement value is used to obtain the heading angle error, speed error and altitude error in real time. The control weight is dynamically adjusted by combining the proportional term, integral term and differential term. The differential term suppresses high-frequency oscillation, and the integral term eliminates steady-state deviation. The attitude adjustment feedback is output. The attitude adjustment feedback includes the heading servo adjustment amount, throttle adjustment amount and elevator adjustment amount.

10. The UAV trajectory tracking system based on multi-sensor fusion according to claim 9, characterized in that: In the trajectory monitoring module, the process of constructing a visual interface including a predicted trajectory line, an actual trajectory line, and a safety envelope surface according to the posture adjustment feedback amount includes: The system integrates the trajectory prediction coordinates output by the trajectory prediction model, the smooth trajectory of the global positioning system, and the safety envelope parameters. It uses plane projection transformation to map the latitude and longitude coordinates into three-dimensional space coordinates, constructing a red predicted trajectory line, a blue actual trajectory line, and a translucent green cylindrical safety envelope. The interface refreshes at a 0.1-second cycle, synchronously displays the real-time attitude adjustment feedback value, and supports zooming, rotation, and trajectory feature parameter access. When the actual trajectory point continues to exceed the safety envelope, the secondary warning mechanism is triggered. The primary warning marks the deviation point and pops up a prompt. The secondary warning forces the control mode to switch. The operator can manually adjust the envelope tolerance parameters and click on the trajectory line to view the time domain variance and the frequency domain principal component energy ratio.

Images