Unmanned aerial vehicle group indoor cooperative positioning method based on inertial navigation and visual assistance

Abstract

The invention discloses an unmanned aerial vehicle group indoor cooperative positioning method based on inertial navigation and visual assistance, and relates to the technical field of unmanned aerial vehicle navigation, and the method comprises the steps: obtaining the photon chip IMU initial angular velocity and multiple groups of three-dimensional acceleration data of each unmanned aerial vehicle, carrying out the fusion processing of a Kalman filtering algorithm, obtaining an attitude quaternion and a corrected linear acceleration, and carrying out the calculation of the attitude quaternion. The method comprises the following steps: calculating a relative distance matrix between unmanned aerial vehicles according to a quantum entanglement distance measurement principle, constructing a spatial topology constraint, synchronously collecting neuromorphic visual event flow data, performing space-time alignment by combining attitude quaternion and the relative distance matrix, generating visual feature point cloud of space-time calibration, and constructing an indoor three-dimensional geomagnetic feature map before inputting a biologically inspired SLAM model. According to the method, real-time geomagnetic data of the unmanned aerial vehicles are collected and matched, position constraint optimization is carried out on visual feature point clouds, a biological inspiration SLAM model is subjected to special training reasoning, information is fused to output a cooperative positioning result, and the method improves the indoor positioning precision and collaboration of the unmanned aerial vehicle group.

Description

Technical Field

[0001] The present invention relates to the field of drone navigation technology, and more specifically, to an indoor collaborative positioning method for a drone swarm based on inertial navigation and vision assistance. Background Art

[0002] In the field of modern science and technology, the demand for indoor collaborative operations of drone swarms continues to rise, involving multiple scenarios such as indoor rescue, material distribution, and building inspection. The indoor environment is complex and GPS signals are blocked. Traditional positioning technology faces challenges. The inertial navigation system uses the IMU sensor carried by the drone to measure angular velocity and acceleration, and obtains position and speed information through integral calculations. However, the error accumulation is significant over long-term use. Visual positioning technology uses cameras to capture images and build environmental maps to achieve positioning. However, when the indoor lighting is unstable and the features are insufficient, the positioning accuracy is greatly reduced. The fusion of the two can make up for each other's shortcomings in the short term, but it has not completely solved the positioning problem.

[0003] For drone swarm operations, relying solely on individual positioning information is far from sufficient; the relative position relationship between groups must also be accurately grasped. Traditional relative ranging technologies, such as ultrasonic and infrared ranging, are significantly susceptible to interference in indoor environments and have limited measurement accuracy, making them unable to meet the needs of high-precision collaborative positioning of large-scale drone swarms. Existing multi-source data fusion algorithms lack real-time performance, making it difficult to efficiently process the massive amount of data generated by drone swarms. Current technologies lack effective modeling of the spatial topological constraints of drone swarms and fail to fully utilize the relative position information between groups to optimize positioning results. Meanwhile, the unique stability of indoor geomagnetic characteristics can serve as auxiliary positioning information. However, existing geomagnetic positioning technologies are limited by magnetic field distortion caused by indoor metal objects, making it difficult to accurately match geomagnetic characteristics. Quantum entanglement ranging, as a cutting-edge technology, brings new ideas for high-precision ranging. However, its application in indoor collaborative positioning of drone swarms is still immature and urgently needs further exploration. This paper aims to propose a method for indoor collaborative positioning of drone swarms that integrates inertial navigation, visual assistance, quantum entanglement ranging, and bio-inspired algorithms. This method overcomes the limitations of traditional technologies and achieves high-precision and real-time collaborative positioning of drone swarms in indoor environments.

[0004] Therefore, the existing technology has problems such as poor real-time data fusion, insufficient ranging anti-interference, weak adaptability to geomagnetic distortion, and underutilization of group topology constraints. Summary of the Invention

[0005] In order to overcome the problems of poor real-time data fusion, insufficient ranging anti-interference, weak adaptability to geomagnetic distortion and underutilization of group topology constraints in the existing technology, the present invention discloses a method for indoor collaborative positioning of drone swarms based on inertial navigation and vision assistance, which can effectively solve the above technical problems.

[0006] In order to solve the above technical problems, the technical solutions of the present invention are as follows:

[0007] A method for indoor collaborative positioning of a swarm of drones based on inertial navigation and vision assistance includes the following steps:

[0008] Obtain the initial angular velocity and multiple sets of three-dimensional acceleration data of the photonic chip IMU of each drone in the drone swarm;

[0009] The initial angular velocity and multiple sets of three-dimensional acceleration data are fused based on the Kalman filter algorithm to obtain the attitude quaternion and the corrected linear acceleration;

[0010] The relative distance matrix between each drone is calculated based on the principle of quantum entanglement ranging. The relative distance matrix is ​​used to construct the spatial topology constraints of the drone swarm;

[0011] Acquire neuromorphic visual event stream data synchronized with the acquisition time of the multiple sets of three-dimensional acceleration data, and perform spatiotemporal alignment on the visual event stream data according to the posture quaternion and the relative distance matrix to obtain a spatiotemporal calibrated visual feature point cloud;

[0012] The spatiotemporally calibrated visual feature point cloud is input into a bio-inspired SLAM model, and combined with the corrected linear acceleration and relative distance matrix, a collaborative positioning result of a swarm of drones in an indoor environment is output.

[0013] Preferably, before inputting the spatiotemporally calibrated visual feature point cloud into a bio-inspired SLAM model, the method further comprises:

[0014] Constructing a three-dimensional geomagnetic characteristic map of the indoor environment, wherein the three-dimensional geomagnetic characteristic map includes geomagnetic vectors and spatial coordinates of key points;

[0015] The real-time geomagnetic data of each drone is collected through the photonic chip IMU, and the matching degree between the real-time geomagnetic data and the three-dimensional geomagnetic feature map is calculated;

[0016] Position constraints are applied to the spatiotemporally calibrated visual feature point cloud according to the geomagnetic feature point with the highest matching degree, so as to generate a visual feature point cloud with optimized constraints.

[0017] Preferably, the training process of the bio-inspired SLAM model includes:

[0018] Constructing a neural network architecture based on insect navigation mechanisms, the architecture comprising a visual feature extraction layer, an inertial information fusion layer, a quantum confinement layer, and a path integration layer;

[0019] Using a reinforcement learning algorithm to train the neural network architecture, the objective function is to minimize the error between the predicted position and the true position;

[0020] By simulating the navigation behavior of insect colonies and introducing a social learning mechanism to optimize model parameters, the bio-inspired SLAM model is obtained.

[0021] Preferably, the reasoning process of the bio-inspired SLAM model includes:

[0022] The visual feature extraction layer processes the visual feature point cloud after the constraint optimization to extract the semantic features of the environment;

[0023] The inertial information fusion layer converts the corrected linear acceleration into displacement increments and performs spatiotemporal alignment with the environmental semantic features;

[0024] The quantum confinement layer constructs spatial constraint conditions according to the relative distance matrix and performs constrained optimization on the displacement increment;

[0025] The path integration layer is based on the insect path integration principle and combines the displacement increment after constraint optimization and the semantic features of the environment to output the absolute position and attitude of the UAV.

[0026] Preferably, the fusing process of the initial angular velocity and multiple sets of three-dimensional acceleration data based on the Kalman filter algorithm includes:

[0027] Establishing a state transfer equation, wherein the state transfer equation uses attitude quaternion and angular velocity as state variables;

[0028] Establishing an observation equation, wherein the observation equation uses three-dimensional acceleration data and geomagnetic data as observation variables;

[0029] The optimal estimated attitude quaternion and the corrected linear acceleration are obtained through iterative calculation of the prediction and update steps.

[0030] Preferably, the calculation of the relative distance matrix between the drones based on the quantum entanglement ranging principle includes:

[0031] Deploy entangled photon pair transmitters and receivers on each drone;

[0032] By measuring the time difference of quantum state collapse of entangled photon pairs, the relative distance between drones is calculated;

[0033] A relative distance matrix is ​​constructed, wherein the matrix elements represent the measured distance between any two UAVs.

[0034] Preferably, the acquiring of neuromorphic visual event stream data synchronized with the acquisition of the multiple sets of three-dimensional acceleration data comprises:

[0035] Using a neuromorphic camera to capture an asynchronous stream of visual events from the environment;

[0036] Time-synchronizing the visual event stream with the multiple sets of three-dimensional acceleration data based on event triggering timestamps;

[0037] The synchronized visual event stream is converted into a visual feature point cloud in the world coordinate system through spatial transformation.

[0038] Preferably, an electronic device includes a processor, a memory, a user interface and a network interface, the memory is used to store instructions, the user interface and the network interface are used to communicate with other devices, and the processor is used to execute the instructions stored in the memory so that the electronic device performs the positioning method as described above.

[0039] Preferably, a computer-readable storage medium is provided, wherein the computer-readable storage medium stores instructions, and when the instructions are executed, the positioning method as described above is executed.

[0040] Preferably, a computer program product is provided, comprising instructions, and when the instructions are executed, the steps of the positioning method described above are performed.

[0041] Compared with the existing technology, the beneficial effects of the present invention are: the present invention efficiently fuses IMU data through the Kalman filter algorithm, introduces the state transfer equation and the observation equation, and realizes the real-time correction of the attitude quaternion and the linear acceleration. This improvement improves the data processing efficiency, ensures the positioning update frequency of large-scale drone groups during high-speed movement, and solves the problem of insufficient real-time performance of traditional fusion algorithms when processing high-dimensional data; based on the principle of quantum entanglement ranging, an entangled photon pair device is deployed, and the relative distance between drones is accurately calculated using the time difference of quantum state collapse. Quantum entanglement ranging is not affected by electromagnetic interference, and can achieve picosecond time resolution, and the measurement accuracy can reach millimeter level, overcoming the problem of reduced accuracy of traditional ranging technology in complex indoor electromagnetic environments, and providing reliable relative position information for drone groups; constructing an indoor three-dimensional geomagnetic feature map, combining the photonic chip IMU to collect geomagnetic data in real time, and matching the visual features with the matching algorithm. The point cloud is used for position constraints, and the long-term stability of geomagnetic features is utilized. Even in areas of magnetic field distortion, adaptive correction can be achieved through feature matching, which improves the robustness of the positioning system; the spatial topological constraints of the drone swarm are constructed through the relative distance matrix, and are deeply integrated into the bio-inspired SLAM model. The quantum constraint layer optimizes the displacement increment according to the topological relationship to ensure the consistency of the group positioning results, fully utilizes the group characteristics of the drone swarm, and effectively suppresses the drift problem of the traditional SLAM algorithm in dense scenes; the bio-inspired SLAM model draws on the navigation mechanism of insects, introduces a social learning mechanism, and realizes the deep fusion of vision, inertia, and quantum ranging information through a multi-layer neural network architecture. The reinforcement learning training process takes minimizing the position error as the objective function to ensure the generalization ability of the model in complex environments. The final output of the collaborative positioning result achieves sub-meter positioning accuracy, which is significantly better than the existing technology. BRIEF DESCRIPTION OF THE DRAWINGS

[0042] In order to more clearly illustrate the implementation methods of the present invention or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the implementation methods or the description of the prior art. Obviously, the drawings described below are merely exemplary. For ordinary technicians in this field, other implementation drawings can be derived based on the provided drawings without any creative work.

[0043] Figure 1 It is a step diagram of the method of the present invention. DETAILED DESCRIPTION

[0044] The accompanying drawings are for illustrative purposes only and are not to be construed as limiting this patent;

[0045] In order to better illustrate this embodiment, some parts in the drawings may be omitted, enlarged, or reduced, and do not represent the actual product size;

[0046] It is understandable to those skilled in the art that some well-known structures and descriptions thereof may be omitted in the drawings.

[0047] The technical solution of the present invention is further described below with reference to the accompanying drawings and embodiments.

[0048] Example

[0049] A method for indoor collaborative positioning of a swarm of drones based on inertial navigation and vision assistance includes the following steps:

[0050] Obtain the initial angular velocity and multiple sets of three-dimensional acceleration data of the photonic chip IMU of each drone in the drone swarm;

[0051] The initial angular velocity and multiple sets of three-dimensional acceleration data are fused based on the Kalman filter algorithm to obtain the attitude quaternion and the corrected linear acceleration;

[0052] The relative distance matrix between each drone is calculated based on the principle of quantum entanglement ranging. The relative distance matrix is ​​used to construct the spatial topology constraints of the drone swarm;

[0053] Acquire neuromorphic visual event stream data synchronized with the acquisition time of the multiple sets of three-dimensional acceleration data, and perform spatiotemporal alignment on the visual event stream data according to the posture quaternion and the relative distance matrix to obtain a spatiotemporal calibrated visual feature point cloud;

[0054] The spatiotemporally calibrated visual feature point cloud is input into a bio-inspired SLAM model, and combined with the corrected linear acceleration and relative distance matrix, a collaborative positioning result of a swarm of drones in an indoor environment is output.

[0055] Before inputting the spatiotemporally calibrated visual feature point cloud into the bio-inspired SLAM model, the method further includes:

[0056] Constructing a three-dimensional geomagnetic characteristic map of the indoor environment, wherein the three-dimensional geomagnetic characteristic map includes geomagnetic vectors and spatial coordinates of key points;

[0057] The real-time geomagnetic data of each drone is collected through the photonic chip IMU, and the matching degree between the real-time geomagnetic data and the three-dimensional geomagnetic feature map is calculated;

[0058] Position constraints are applied to the spatiotemporally calibrated visual feature point cloud according to the geomagnetic feature point with the highest matching degree, so as to generate a visual feature point cloud with optimized constraints.

[0059] The training process of the bio-inspired SLAM model includes:

[0060] Constructing a neural network architecture based on insect navigation mechanisms, the architecture comprising a visual feature extraction layer, an inertial information fusion layer, a quantum confinement layer, and a path integration layer;

[0061] Using a reinforcement learning algorithm to train the neural network architecture, the objective function is to minimize the error between the predicted position and the true position;

[0062] By simulating the navigation behavior of insect colonies and introducing a social learning mechanism to optimize model parameters, the bio-inspired SLAM model is obtained.

[0063] The reasoning process of the bio-inspired SLAM model includes:

[0064] The visual feature extraction layer processes the visual feature point cloud after the constraint optimization to extract the semantic features of the environment;

[0065] The inertial information fusion layer converts the corrected linear acceleration into displacement increments and performs spatiotemporal alignment with the environmental semantic features;

[0066] The quantum confinement layer constructs spatial constraint conditions according to the relative distance matrix and performs constrained optimization on the displacement increment;

[0067] The path integration layer is based on the insect path integration principle and combines the displacement increment after constraint optimization and the semantic features of the environment to output the absolute position and attitude of the UAV.

[0068] The fusing process of the initial angular velocity and multiple sets of three-dimensional acceleration data based on the Kalman filter algorithm includes:

[0069] Establishing a state transfer equation, wherein the state transfer equation uses attitude quaternion and angular velocity as state variables;

[0070] Establishing an observation equation, wherein the observation equation uses three-dimensional acceleration data and geomagnetic data as observation variables;

[0071] The optimal estimated attitude quaternion and the corrected linear acceleration are obtained through iterative calculation of the prediction and update steps.

[0072] The method of calculating the relative distance matrix between drones based on the quantum entanglement ranging principle includes:

[0073] Deploy entangled photon pair transmitters and receivers on each drone;

[0074] By measuring the time difference of quantum state collapse of entangled photon pairs, the relative distance between drones is calculated;

[0075] A relative distance matrix is ​​constructed, wherein the matrix elements represent the measured distance between any two UAVs.

[0076] The acquiring of neuromorphic visual event stream data synchronized with the acquisition of the multiple sets of three-dimensional acceleration data comprises:

[0077] Using a neuromorphic camera to capture an asynchronous stream of visual events from the environment;

[0078] Time-synchronizing the visual event stream with the multiple sets of three-dimensional acceleration data based on event triggering timestamps;

[0079] The synchronized visual event stream is converted into a visual feature point cloud in the world coordinate system through spatial transformation.

[0080] An electronic device includes a processor, a memory, a user interface, and a network interface, wherein the memory is used to store instructions, the user interface and the network interface are used to communicate with other devices, and the processor is used to execute the instructions stored in the memory so that the electronic device performs the positioning method described above.

[0081] A computer-readable storage medium stores instructions, and when the instructions are executed, the positioning method as described above is executed.

[0082] A computer program product comprises instructions, and when the instructions are executed, the steps of the positioning method described above are performed.

[0083] The test was conducted in an indoor experimental field measuring 20m×15m×5m. The field was filled with obstacles of various shapes and colors, such as tables, chairs, and cabinets, to simulate a real and complex indoor environment. The experiment used a swarm of six drones of the same model, each equipped with the following equipment:

[0084] Photonic chip IMU: A high-precision photonic chip inertial measurement unit is selected, with an angular velocity measurement range of ±2000° / s, an acceleration measurement range of ±16g, and a sampling frequency of 100Hz. It is used to obtain the initial angular velocity and three-dimensional acceleration data of the drone.

[0085] Entangled photon pair transmitting and receiving device: This device uses special equipment designed based on the principle of quantum entanglement, with measurement accuracy reaching the centimeter level, and is used to achieve relative distance measurement between drones.

[0086] Neuromorphic camera: The DVS128 neuromorphic camera is selected. It has asynchronous event-driven characteristics and can collect visual event stream data of the environment in real time with a time resolution of microseconds.

[0087] Geomagnetic sensor: A high-precision geomagnetic sensor integrated in the photonic chip IMU with a measurement accuracy of ±0.1μT is used to collect real-time geomagnetic data of the drone.

[0088] At the same time, a high-performance server was prepared as the data processing center. The server was configured with an Intel Xeon Gold 6248R processor, 128GB of memory, and an NVIDIA Tesla V100 graphics card. The Ubuntu 20.04 operating system was installed, and a software development environment based on Python 3.8 was built. Necessary software packages such as the PyTorch 1.12 deep learning framework and the NumPy 1.22 scientific computing library were installed.

[0089] In the specific implementation, please refer to Figure 1 After each drone is started, the photonic chip IMU begins to collect initial angular velocity and three-dimensional acceleration data at a frequency of 100 Hz. For example, within the first second, drone 1 collects 100 sets of data, each set of data contains three-axis angular velocity (ωx, ωy, ωz) and three-axis acceleration (ax, ay, az).

[0090] The collected initial angular velocity and three-dimensional acceleration data are fused based on the Kalman filter algorithm. First, the state transfer equation is established, with the attitude quaternion q = [q0, q1, q2, q3] and angular velocity ω = [ωx, ωy, ωz] as state variables. The state transfer equation is expressed as:

[0091]

[0092] in, is the predicted state at time k based on time k-1, F k-1 is the state transition matrix, B k-1 is the control matrix, u k-1 is the control input, ω k-1 is the process noise.

[0093] Then, the observation equation is established, with three-dimensional acceleration data and geomagnetic data as observation variables. The observation equation is expressed as:

[0094]

[0095] Among them, z k is the observed value, H k is the observation matrix, v k is the observation noise.

[0096] Through iterative calculations in the prediction and update steps, after processing 100 sets of data, the optimal estimated attitude quaternion q and the corrected linear acceleration a' are obtained. For example, the attitude quaternion q1 = [0.707, 0, 0, 0.707] and the corrected linear acceleration a'1 = [0.2, 0.1, -9.8] m / s of UAV 1 at the end of the first second are obtained. 2 .

[0097] After deploying the entangled photon pair transmitter and receiver on each drone, the quantum entanglement ranging program is started. Each drone transmits an entangled photon pair and measures the quantum state collapse time difference Δt of the entangled photon pair. The relative distance between the drones is calculated according to the formula d = cΔt / 2 (where c is the speed of light). For example, the time difference Δt measured between drone 1 and drone 2 is 12 =100ns, then the relative distance d between them 12 =30×10 8 ×100×10 -9 / 2=15m.

[0098] Repeat the above process to measure the relative distance between any two drones in the drone group and construct the relative distance matrix D, which is a 6×6 square matrix with element D ij Represents the measured distance between UAV i and UAV j. This relative distance matrix is ​​used to construct the spatial topological constraints of the UAV swarm.

[0099] A neuromorphic camera is used to collect the asynchronous visual event stream of the environment. Based on the event trigger timestamp, the visual event stream is time-synchronized with multiple sets of three-dimensional acceleration data. For example, when the photonic chip IMU collects a set of acceleration data at t = 0.5s, the visual event triggered by the neuromorphic camera near that time is found, and precise synchronization is achieved through methods such as time interpolation. Then, the synchronized visual event stream is converted into a visual feature point cloud in the world coordinate system through spatial transformation. Specifically, according to the attitude quaternion and initial position information of the drone, the coordinate transformation formula is used to convert the pixel coordinates of the visual event into three-dimensional space coordinates to obtain the visual feature point cloud P, where each point contains three-dimensional coordinate (x, y, z) information.

[0100] In the experimental site, high-precision geomagnetic measurement equipment is used to pre-collect the geomagnetic vectors and spatial coordinates of key points to construct a three-dimensional geomagnetic feature map M of the indoor environment. During the collection process, 100 key points are evenly selected in the site, and geomagnetic data are collected 10 times for each key point. The average value is taken as the geomagnetic vector of the point. For example, the geomagnetic vector of key point (1, 2, 3) is [0.3, 0.2, -0.1] μT. These data are stored in the three-dimensional geomagnetic feature map M.

[0101] Each UAV collects real-time geomagnetic data through the photonic chip IMU, calculates the matching degree between the real-time geomagnetic data and the geomagnetic vectors of key points in the three-dimensional geomagnetic feature map M, and uses the cosine similarity calculation method to calculate the cosine similarity between the geomagnetic data d1 = [0.31, 0.21, -0.09] μT collected by UAV 1 at a certain moment and the geomagnetic vectors of each key point in the three-dimensional geomagnetic feature map, and finds the geomagnetic feature point with the highest matching degree.

[0102] The spatial-temporally calibrated visual feature point cloud P is constrained according to the geomagnetic feature points with the highest matching degree. The coordinates of the visual feature point cloud are adjusted to match the spatial positions of the geomagnetic feature points to generate the constraint-optimized visual feature point cloud P'.

[0103] A neural network architecture based on the insect navigation mechanism is constructed, which includes a visual feature extraction layer, an inertial information fusion layer, a quantum constraint layer and a path integration layer. The visual feature extraction layer adopts a convolutional neural network (CNN) structure, which contains 3 convolution layers and 2 pooling layers to extract the semantic features of the visual feature point cloud; the inertial information fusion layer converts the corrected linear acceleration into displacement increments and aligns them with the visual semantic features in space and time; the quantum constraint layer constructs spatial constraints according to the relative distance matrix; and the path integration layer is designed based on the insect path integration principle.

[0104] A reinforcement learning algorithm is used to train the neural network architecture, with the objective function of minimizing the error between the predicted position and the actual position. During the training process, the motion trajectory of the drone in an indoor environment is simulated, and a reward mechanism is set up. When the error between the predicted position and the actual position is small, a positive reward is given, otherwise a negative reward is given.

[0105] By simulating the navigation behavior of insect groups, a social learning mechanism is introduced to optimize model parameters. For example, different neural network models are allowed to learn from each other and exchange some parameters during the training process. After 1,000 training cycles, a trained bio-inspired SLAM model is obtained.

[0106] The visual feature extraction layer processes the visual feature point cloud P' after constraint optimization, and extracts the environmental semantic features S through convolution and pooling operations, such as identifying the features of objects such as walls, tables and chairs.

[0107] The inertial information fusion layer converts the corrected linear acceleration a' into a displacement increment Δs, calculates it based on the integral relationship between acceleration and displacement, and combines it with the time interval, and then aligns the displacement increment Δs with the environmental semantic feature S in time and space.

[0108] The quantum constraint layer constructs spatial constraints based on the relative distance matrix D and performs constrained optimization on the displacement increment Δs to ensure that the position update of the drone conforms to the spatial topological relationship.

[0109] The path integration layer is based on the insect path integration principle, combined with the displacement increment after constraint optimization and the semantic features of the environment, and outputs the absolute position X = [x, y, z] and posture Q = [q0, q1, q2, q3] of the drone. For example, the absolute position of drone 1 is (5, 3, 2) m and the posture is [0.70, 0, 0, 0.707].

[0110] The high-performance server used in the experiment is an electronic device. Its processor executes instructions stored in the memory to implement the above-mentioned positioning method. The functions of each step are implemented by calling relevant library functions through the program. The user interface and network interface are used for data communication between the drone and the server. The collected data is transmitted to the server for processing and the positioning results are returned to the drone.

[0111] At the same time, the program that implements the positioning method is packaged as a computer program product and stored in a computer-readable storage medium, such as a hard disk, a USB flash drive, etc. When the instructions are run, the above positioning method can be executed to realize the collaborative positioning of the drone group in an indoor environment.

[0112] The same or similar reference numerals correspond to the same or similar components;

[0113] The terms used in the drawings to describe positional relationships are for illustrative purposes only and should not be construed as limiting this patent;

[0114] Obviously, the above embodiments of the present invention are merely examples for clearly illustrating the present invention, and are not limitations on the implementation methods of the present invention. For ordinary technicians in the field, other different forms of changes or modifications can be made based on the above description. It is not necessary and impossible to enumerate all the implementation methods here. Any modifications, equivalent substitutions and improvements made within the spirit and principles of the present invention should be included in the scope of protection of the claims of the present invention.

Claims

1. A method for indoor collaborative positioning of drone swarms based on inertial navigation and vision assistance, characterized in that: The following steps are involved: Obtain the initial angular velocity and multiple sets of three-dimensional acceleration data of the photonic chip IMU of each drone in the drone swarm; The initial angular velocity and multiple sets of three-dimensional acceleration data are fused based on the Kalman filter algorithm to obtain the attitude quaternion and the corrected linear acceleration; The relative distance matrix between each drone is calculated based on the principle of quantum entanglement ranging. The relative distance matrix is ​​used to construct the spatial topology constraints of the drone swarm; Acquire neuromorphic visual event stream data synchronized with the acquisition time of the multiple sets of three-dimensional acceleration data, and perform spatiotemporal alignment on the visual event stream data according to the posture quaternion and the relative distance matrix to obtain a spatiotemporal calibrated visual feature point cloud; The spatiotemporally calibrated visual feature point cloud is input into a bio-inspired SLAM model, and combined with the corrected linear acceleration and relative distance matrix, a collaborative positioning result of a swarm of drones in an indoor environment is output.

2. The positioning method according to claim 1, wherein: Before inputting the spatiotemporally calibrated visual feature point cloud into the bio-inspired SLAM model, the method further includes: Constructing a three-dimensional geomagnetic characteristic map of the indoor environment, wherein the three-dimensional geomagnetic characteristic map includes geomagnetic vectors and spatial coordinates of key points; The real-time geomagnetic data of each drone is collected through the photonic chip IMU, and the matching degree between the real-time geomagnetic data and the three-dimensional geomagnetic feature map is calculated; Position constraints are applied to the spatiotemporally calibrated visual feature point cloud according to the geomagnetic feature point with the highest matching degree, so as to generate a visual feature point cloud with optimized constraints.

3. The positioning method according to claim 2, characterized in that: The training process of the bio-inspired SLAM model includes: Constructing a neural network architecture based on insect navigation mechanisms, the architecture comprising a visual feature extraction layer, an inertial information fusion layer, a quantum confinement layer, and a path integration layer; Using a reinforcement learning algorithm to train the neural network architecture, the objective function is to minimize the error between the predicted position and the true position; By simulating the navigation behavior of insect colonies and introducing a social learning mechanism to optimize model parameters, the bio-inspired SLAM model is obtained.

4. The positioning method according to claim 3, characterized in that: The reasoning process of the bio-inspired SLAM model includes: The visual feature extraction layer processes the visual feature point cloud after the constraint optimization to extract the semantic features of the environment; The inertial information fusion layer converts the corrected linear acceleration into displacement increments and performs spatiotemporal alignment with the environmental semantic features; The quantum confinement layer constructs spatial constraint conditions according to the relative distance matrix and performs constrained optimization on the displacement increment; The path integration layer is based on the insect path integration principle and combines the displacement increment after constraint optimization and the semantic features of the environment to output the absolute position and attitude of the UAV.

5. The positioning method according to claim 1, wherein: The fusing process of the initial angular velocity and multiple sets of three-dimensional acceleration data based on the Kalman filter algorithm includes: Establishing a state transfer equation, wherein the state transfer equation uses attitude quaternion and angular velocity as state variables; Establishing an observation equation, wherein the observation equation uses three-dimensional acceleration data and geomagnetic data as observation variables; The optimal estimated attitude quaternion and the corrected linear acceleration are obtained through iterative calculation of the prediction and update steps.

6. The positioning method according to claim 1, characterized in that: The method of calculating the relative distance matrix between drones based on the quantum entanglement ranging principle includes: Deploy entangled photon pair transmitters and receivers on each drone; By measuring the time difference of quantum state collapse of entangled photon pairs, the relative distance between drones is calculated; A relative distance matrix is ​​constructed, wherein the matrix elements represent the measured distance between any two UAVs.

7. The positioning method according to claim 1, characterized in that: The acquiring of neuromorphic visual event stream data synchronized with the acquisition of the multiple sets of three-dimensional acceleration data comprises: Using a neuromorphic camera to capture an asynchronous stream of visual events from the environment; Time-synchronizing the visual event stream with the multiple sets of three-dimensional acceleration data based on event triggering timestamps; The synchronized visual event stream is converted into a visual feature point cloud in the world coordinate system through spatial transformation.

8. An electronic device, characterized in that: It includes a processor, a memory, a user interface and a network interface, the memory is used to store instructions, the user interface and the network interface are used to communicate with other devices, and the processor is used to execute the instructions stored in the memory so that the electronic device performs the positioning method according to any one of claims 1 to 7.

9. A computer-readable storage medium, characterized in that The computer-readable storage medium stores instructions, and when the instructions are executed, the positioning method according to any one of claims 1 to 7 is executed.

10. A computer program product, characterized in that The computer program product comprises instructions, and when the instructions are executed, the steps of the positioning method according to any one of claims 1 to 7 are performed.

Images