A target tracking method for mobile sensor networks based on force-guided localization

Abstract

The present invention discloses a method for tracking targets in a mobile sensor network based on force-guided positioning, comprising establishing a mobile sensor network structure model; optimizing the positioning of mobile nodes according to a force-guided algorithm and updating the mobile sensor network structure model; sampling and processing the monitored target according to the sensors equipped on each mobile node to obtain the location information of the monitored target; the mobile node transmitting the acquired location information of the monitored target to other mobile nodes other than itself; fusing the location information of the monitored target acquired by each mobile node using a hierarchical clustering method; and each mobile node in the updated mobile sensor network structure model tracking the target according to the fused location information of the monitored target. This method solves the problems of low node positioning accuracy, poor target tracking stability, and low data fusion speed that exist in traditional wireless sensor networks during application.

Description

Technical Field

[0001] The present invention relates to the field of swarm intelligent robot control, and in particular to a mobile sensor network target tracking method based on force-guided positioning. Background Art

[0002] In recent years, the study of swarm intelligence and collaborative control has garnered widespread attention and research from researchers across diverse fields, including computer science, mathematics, artificial intelligence, military affairs, and control engineering, making it a major research hotspot. Currently, in the military, firefighting, police, and rescue sectors, robots are urgently needed to replace humans in dangerous or challenging tasks, possessing capabilities such as detection and tracking within a specific area. For example, in the military, robots can assist humans in tracking and locating targets in complex terrain, achieving maximum efficiency with limited human resources. Similarly, robots can also be widely used in search and rescue, performing search and rescue missions in dangerous areas. Unlike the single, simple functions of commonly used individual robots, swarm robot systems possess collaborative capabilities, resulting in increased efficiency, greater flexibility, and greater precision, making them more suitable for complex mission scenarios. Swarm robots can cover larger areas in a shorter time and operate continuously, enabling them to more effectively track and locate targets. Furthermore, swarm robots can enhance the performance of their individual counterparts through collaborative work, thereby improving the efficiency and accuracy of the entire system. Finally, the use of swarm robots can reduce operating costs because they can complete tasks without human intervention.

[0003] Haoxu et al. proposed Omni-Swarm, a system combining an omnidirectional perception front-end and a graph-based optimization back-end, to achieve high accuracy and robustness in collaborative target tracking. YulunTian et al. proposed Kimera-Multi, a distributed multi-robot system that improves the perception capabilities of multiple robots by estimating 3D mesh models. Pierre-Yves Lajoie et al. proposed DOOR-SLAM, a fully distributed robot team SLAM system that operates with fewer conservative parameters, rejects false measurements, and computes accurate trajectory estimates. BoyuZhou proposed RACER, a rapid collaborative exploration method using a decentralized drone swarm, to enable a decentralized fleet of quadrotors to quickly explore complex environments. The method optimizes coverage paths in unknown spaces. Lusk et al. proposed a formation flight pipeline that can be run on vehicles and unifies distributed formation control and task allocation solutions. It uses VIO for localization and addresses the scalability issue of general solvers. KNMcGuire et al. proposed a swarm bug algorithm or minimum navigation scheme: SGBA for the navigation strategy of micro-robots exploring unknown indoor environments. The algorithm maximizes the range coverage by letting clusters of micro-UAVs navigate in different directions from the starting point.

[0004] However, in practical applications, traditional target tracking technologies present numerous challenges. For example, the positioning accuracy of GPS-based technology is limited by environmental factors, resulting in large positioning errors. While inertial navigation systems offer high accuracy, they require significant computing resources and physical sensors, resulting in high costs. While computer vision-based methods can track and locate targets through image processing, complex scenes and lighting conditions can severely impact their stability and accuracy. Therefore, improving sensor accuracy and reliability, reducing sensor energy consumption, achieving better coordinated control, and realizing efficient target tracking in large-scale scenarios are pressing challenges in future research. Summary of the Invention

[0005] The present invention provides a mobile sensor network target tracking method based on force-guided positioning to overcome the above technical problems.

[0006] In order to achieve the above object, the technical solution of the present invention is:

[0007] A method for target tracking in a mobile sensor network based on force-guided positioning comprises the following steps:

[0008] Step S1: Set up several robots as mobile nodes and establish a mobile sensor network structure model;

[0009] Step S2: Optimizing the positioning of the mobile node according to the force-directed algorithm and updating the mobile sensor network structure model;

[0010] Step S3: sampling the monitored target according to the sensor equipped by each mobile node to obtain the location information of the monitored target;

[0011] Step S4: the mobile node transmits the acquired location information of the monitoring target to other mobile nodes except itself;

[0012] Performing information fusion on the location information of the monitored target obtained by each of the mobile nodes through a hierarchical clustering method;

[0013] Step S5: each mobile node in the updated mobile sensor network structure model realizes target tracking according to the position information of the monitored target after information fusion.

[0014] Furthermore, the mobile node is optimized and positioned according to the force-directed algorithm in step S2, specifically:

[0015] Step S2.1: multiple UWB ranging modules are set on the mobile node, and the relative distance information between any two mobile nodes can be measured by the UWB ranging modules; and an actual distance matrix D is obtained based on the relative distance information;

[0016] Randomly initialize the ranging module to obtain the estimated position of all ranging modules in the mobile sensor network structure model

[0017] Step S2.2: Based on the estimated position of the ranging module Calculate the estimated distance matrix

[0018] Obtaining the estimated distance matrix The calculation formula is

[0019]

[0020] in, Represents the estimated distance matrix The Euclidean distance between the i-th ranging module and the j-th ranging module in ; represents the estimated position of the i-th ranging module; represents the estimated position of the jth ranging module;

[0021] Step S2.3: Estimated position of each ranging module based on the force guidance algorithm Establish a mobile sensor network structure model, select a set of estimated positions of ranging modules that optimize the system energy of the mobile sensor network structure model

[0022] Step S2.4: Estimated position based on the selected optimized ranging module Finally, the actual position x of the ranging module corresponding to the actual distance matrix D is obtained;

[0023] The calculation formula for the actual position x of the ranging module is:

[0024]

[0025] Where: Q represents an orthogonal K×K matrix; T represents an n×K translation matrix;

[0026] The posture of the mobile node is calculated according to the actual position x, so as to update the mobile sensor network structure model.

[0027] Furthermore, the estimated positions of a group of ranging modules that optimize the system energy are selected in step S2.3. Specifically

[0028] Step S2.3.1: Assume that when ranging module i and ranging module j move away from or towards each other under the action of the guiding force, the potential energy H generated by the guiding force ij for

[0029]

[0030] Where: d ij Indicates the actual value obtained by UWB ranging; It's about x i ,x j The distance function of

[0031] Step S2.3.2: According to the guiding force potential energy H ij Obtain the total system energy of the mobile sensor network structure model;

[0032] The total energy H of the system is a monotonically decreasing function, and the calculation formula is

[0033]

[0034] Where: Represents a collection of ranging modules; represents the Frobenius norm of the matrix; i and j represent the ranging module;

[0035] Step S2.3.3: Definition and Iterative control equations for ranging modules in mobile sensor networks;

[0036]

[0037] Where: Represents the estimated distance matrix Elements in It means exist Component in the direction of the vector; It means exist The component of the vector direction;

[0038] Step S2.3.4: Based on formula (4) and formula (6), the total energy H of the current system is compared with the preset energy threshold ∈ D Make comparisons;

[0039] If the total energy H of the system is greater than the preset energy threshold ∈ D , then perform iterative calculation according to formula (6);

[0040] If the total energy H of the system is less than the preset energy threshold ∈ D , then obtain the estimated position of the corresponding ranging module according to the current system total energy H

[0041] Furthermore, in step S2.4, the posture of the mobile node is calculated according to the actual position x, and the posture calculation formula is:

[0042]

[0043] Where: P i represents the posture information of node i; P 0i represents the location information of node i; θ i represents the direction of node i; x i A,x i B,x i C∈x i They represent the position of the ranging module installed on node i; and the direction of node i is x i C is the endpoint pointing to x i A and x i The vector of the midpoint of the line B.

[0044] Furthermore, the acquisition of the location information of the monitored target in step S3 includes:

[0045] Step S3.1: Obtain a 360-degree panoramic image within the detection range through a fisheye camera installed on the top of the mobile node;

[0046] Step S3.2: radially expanding the panoramic image through a radial transformation to remove image distortion;

[0047] Step S3.3: Obtaining the position of the target detection box in the radially expanded panoramic image based on the deep convolutional neural network;

[0048] Step S3.4: Calculate the coordinates of the target location based on the location of the target detection frame in the panoramic image.

[0049] Furthermore, the formula for calculating the target location coordinates in step S3.4 is:

[0050]

[0051] Where: (x t ,y t ) represents the coordinates of the global image where the target is located; d represents the function between the target detection box area and the target distance; a represents the area of ​​the target detection box; W represents the width of the panoramic image; x C Represents the horizontal coordinate of the center point of the target detection frame; θ t Indicates the position of the target relative to the fisheye camera.

[0052] Furthermore, in step S4, the location information of the monitored target obtained by each mobile node is fused by a hierarchical clustering method; specifically,

[0053] Step S4.1: Each mobile node broadcasts the location information of one or more monitored targets acquired in its own coordinate system to other mobile nodes in real time via wireless communication of the UWB ranging module;

[0054] Step S4.2: The current mobile node integrates the acquired location information of all monitored targets;

[0055] The information fusion strategy is to convert the location information of the monitored target obtained by the other nodes into the location information in the current node coordinate system;

[0056] Compare the difference between the converted position information of the monitored target and the position information of the monitored target detected by the current node with a preset distance error threshold;

[0057] If the position difference is greater than a preset distance error threshold, the converted position information of the monitored target is not retained;

[0058] If the position difference is less than a preset distance error threshold, clustering the converted position information of the monitored target with the position information of the monitored target detected by the current node to obtain one or more target position data classes;

[0059] Step S4.3: averaging the acquired position information of each target position data class to obtain optimized target position information;

[0060] The optimized target position information closest to the center of the mobile sensor network structure model is used as target tracking information.

[0061] Furthermore, in step S4.2, the position information of the monitored target obtained by other nodes is converted into the position information in the current node coordinate system. The conversion formula is:

[0062] p ti =p tj R(θ i -θ j )+(p i -p j ) (10)

[0063] Where: p ti represents the location information of the target in the coordinate system of mobile node i; p tj represents the target location information of node j received by node i; θ i and θ j represents the direction of mobile node i and mobile node j; p i With p j Represents the current position coordinates of the mobile node; R(θ i -θ j ) represents the rotation matrix in K-dimensional space.

[0064] Beneficial Effects: The present invention discloses a method for target tracking in a mobile sensor network based on force-guided positioning. This method uses UWB ranging technology to achieve node positioning and network structure adjustment in a mobile sensor network without relying on global positioning systems such as GPS. A force-guided algorithm is used to accurately locate mobile nodes without the need for any global information. Distributed collaborative positioning between nodes and node position and posture estimation can be achieved using only the UWB ranging module. Deep learning methods are used for visual target recognition in individual mobile nodes, and hierarchical clustering methods are used to achieve visual information fusion and target position determination between nodes. High-precision and high-efficiency sensor network target recognition and collaborative tracking control are achieved without a central control system or global positioning system. BRIEF DESCRIPTION OF THE DRAWINGS

[0065] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the following is a brief introduction to the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative labor.

[0066] Figure 1 This is a flow chart of a target tracking method for a mobile sensor network based on force-guided positioning according to the present invention;

[0067] Figure 2 This is a simulation diagram of the force-guided positioning algorithm in this embodiment;

[0068] Figure 3 This is a diagram showing changes in total system energy in this embodiment;

[0069] Figure 4 This is a performance comparison analysis chart of the NLP, MDS, and FD algorithms in this embodiment;

[0070] Figure 5 This is a simulation experiment diagram of the group control algorithm in this embodiment;

[0071] Figure 6 This is a simulation diagram of the obstacle avoidance algorithm in this embodiment;

[0072] Figure 7 are the performance parameters of the swarm control simulation experiment in this embodiment, including positioning error, angle estimation error, minimum distance between nodes, and distance from the swarm center to the target;

[0073] Figure 8 This is the robot hardware experiment configuration in this embodiment;

[0074] Figure 9 This is a diagram showing the calculation process of the tracking target position coordinates in this embodiment;

[0075] Figure 10 This is a demonstration of visual target information sharing in a mobile sensor network in this embodiment;

[0076] Figure 11 This is a diagram demonstrating the hierarchical clustering results in this embodiment;

[0077] Figure 12 This is a diagram demonstrating the test operation process in this embodiment. DETAILED DESCRIPTION

[0078] 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.

[0079] This embodiment provides a target tracking method for a mobile sensor network based on force-guided positioning. Figure 1 As shown, the following steps are included:

[0080] Step S1: Set up several robots as mobile nodes and establish a mobile sensor network structure model; obtain the distance between two robots through the UWB ranging module. The swarm robot system is regarded as a wireless sensor network model, and the robots are regarded as nodes. The model contains m nodes;

[0081] Step S2: Optimizing the positioning of the mobile node according to the force-directed algorithm and updating the mobile sensor network structure model;

[0082] Step S3: sampling the monitored target according to the sensor equipped by each mobile node to obtain the location information of the monitored target;

[0083] Step S4: the mobile node transmits the acquired location information of the monitoring target to other mobile nodes except itself;

[0084] Performing information fusion on the location information of the monitored target obtained by each of the mobile nodes through a hierarchical clustering method;

[0085] Step S5: each mobile node in the updated mobile sensor network structure model realizes target tracking according to the position information of the monitored target after information fusion.

[0086] By adopting the UWB ranging module, the node positioning and structural adjustment of the sensor network are realized without relying on global positioning systems such as GPS. Based on the hardware of the single-node multi-ranging module, the force guidance model is combined to determine the distance and posture between nodes to achieve group control; and by using deep learning methods for target recognition on a single mobile node, the hierarchical clustering method is used to realize the visual information fusion and target position determination between nodes; thus, high-precision and high-efficiency sensor network target recognition and collaborative tracking control are achieved.

[0087] In a specific embodiment, the step S2 is to optimize the positioning of the mobile node according to the force-directed algorithm, specifically:

[0088] Step S2.1: Three UWB ranging modules are installed on the mobile node. The three UWB ranging modules are arranged in an equilateral triangle, that is, any two UWB ranging modules are arranged side by side at the front end of the robot, and the third UWB ranging module is arranged at the center of the rear end of the robot. The UWB ranging modules can measure the relative distance information between any two mobile nodes, and obtain the actual distance matrix D based on the relative distance information.

[0089] Wireless sensor (UWB) provides ranging module usually includes the following components:

[0090] Transmitter: transmits high-frequency radio signals;

[0091] Receiver: Receives and amplifies the radio signal from the transmitter;

[0092] Filter: used to remove noise and other interfering signals;

[0093] Pulse counter: used to measure the time it takes for a signal to reflect, scatter, or transmit;

[0094] Calculator: used to calculate distance;

[0095] When the UWB ranging module receives the reflected signal from the target, it can calculate the distance from the target to the sensor by measuring the time it takes to receive the signal;

[0096] Randomly initialize the ranging module to obtain the estimated position of all ranging modules in the mobile sensor network structure model

[0097] Step S2.2: Based on the estimated position of the ranging module Calculate the estimated distance matrix

[0098] Obtaining the estimated distance matrix The calculation formula is

[0099]

[0100] in, Represents the estimated distance matrix The Euclidean distance between the i-th ranging module and the j-th ranging module in ; represents the estimated position of the i-th ranging module; represents the estimated position of the jth ranging module;

[0101] Step S2.3: The force guidance algorithm optimizes the mobile node positioning problem through the principle of interaction between the two ends of the spring, and estimates the position of each ranging module based on the force guidance algorithm. Establish a mobile sensor network structure model, select a set of estimated positions of ranging modules that optimize the system energy of the mobile sensor network structure model

[0102] Step S2.4: Estimated position based on the selected optimized ranging module Finally, the actual position x of the ranging module corresponding to the actual distance matrix D is obtained;

[0103] The calculation formula for the actual position x of the ranging module is:

[0104]

[0105] Where: Q represents an orthogonal K×K matrix; T represents an n×K translation matrix;

[0106] Calculating the posture of the mobile node according to the actual position x to update the mobile sensor network structure model;

[0107] The attitude information of each node is calculated by the position information of its three ranging modules. The attitude calculation formula is:

[0108]

[0109] Where: P i represents the posture information of node i; P 0i represents the location information of node i; θ i represents the direction of node i; x i A,x i B,x i C∈x i They represent the position of the ranging module installed on node i; and the direction of node i is x i C is the endpoint pointing to x i A and x i The vector of the midpoint of the line B.

[0110] The mobile wireless network structure obtained by the force-guided positioning algorithm may not only rotate and translate, but also flip. That is, the optimization result may be a mirror image of the real sensor network structure (e.g. Figure 1 (as shown on the right). When a flip occurs, the arrangement of the ranging modules on each mobile node will mirror the actual location. Therefore, by simply checking whether the estimated location of the ranging module on each node mirrors the actual installation location, we can determine whether a flip has occurred. If a flip has occurred, we can use the calculation method described in formula (1) to perform a reverse operation to restore the actual network structure.

[0111] In a specific embodiment, the estimated positions of a group of ranging modules that optimize the system energy are selected in step S2.3. Specifically

[0112] Step S2.3.1: Obtain the distance between the two robots through the UWB ranging module. Use the distance information obtained by the UWB ranging module as the default length of the spring between the connecting nodes. During the operation of the force guidance algorithm, the repulsive and attractive forces on the spring in the force guidance algorithm are analogous to the repulsive and attractive forces on each robot (mobile node) by other mobile nodes. Based on time conditions, acceleration and displacement are generated to achieve mobile node positioning. Assume that when ranging module i and ranging module j move away from or approach each other under the action of the guiding force, the potential energy H generated by the guiding force is ij for

[0113]

[0114] Where: d ij Indicates the actual value obtained by UWB ranging; It's about The distance function, and

[0115] Step S2.3.2: According to the guiding force potential energy H ij Obtain the total system energy of the mobile sensor network structure model;

[0116] The calculation formula of the total energy H of the system is:

[0117]

[0118] Where: Represents a collection of ranging modules; represents the Frobenius norm of the matrix; i and j represent the ranging module;

[0119] According to the time derivative of the total energy H of the system, it can be seen from formula (5) that the total energy H of the system is a monotonically decreasing function;

[0120]

[0121] Where: It means exist Component in the direction of the vector; It means exist The component of the vector direction; x i Indicates that the ranging module i is in the K-dimensional space Position on x jIndicates that the ranging module j is in the K-dimensional space Position on

[0122] Step S2.3.3: Definition and Iterative control equations for ranging modules in mobile sensor networks;

[0123]

[0124] Where: Represents the estimated distance matrix Elements in It means exist Component in the direction of the vector; It means exist The above formula represents the control equation of the positioning algorithm of the ranging module in the sensor network. As can be seen from the description of this formula, ranging module i will be affected by the traction or stretching force from all ranging modules in the mobile wireless network. The magnitude of the force from ranging module j is The direction of the force is from x i Point to x j ;if Then the force is positive, x i Received from x j The traction in the direction of x j Move; if Then the force is negative, x i Received from x j The tensile force affects the direction of x j Move in the opposite direction of

[0125] From formula (4) and formula (5), we can see that the total energy equation of the system is It's about If the distance measurement module in the sensor network estimates the position using the iterative control equation of formula (6), the total energy of the system is a non-positive function with respect to the time derivative. That is, the system energy decreases monotonically until When the system reaches a stable state, there will be two stable states:

[0126] The first stable state: X1=X2=X3=…=X n ,(i∈V)

[0127] In this stable state, all estimated positions will converge to one point. Obviously, this stable state is not a desirable result unless all ranging modules in the sensor network converge to one point. According to formula (4), when the system approaches this stable state, the energy of the system will infinitely approach Therefore, in order to avoid the occurrence of such erroneous optimization results, the final optimization results To accurately describe the real sensor network structure, it is necessary to ensure The initial value of satisfies the following constraints:

[0128]

[0129] According to formula (5), the total energy of the system will decrease monotonically over time. When the initial value of the system energy meets the constraint conditions, the system energy cannot reach Therefore, the optimization result cannot reach the state where all nodes converge to one point.

[0130] The second stable state:

[0131] In this stable state, all elements in the estimated distance matrix are consistent with all elements in the true distance matrix, that is, Estimated position at this time The real space position x of the ranging module has the same rigid body structure. However, in the real world, due to factors such as ranging error and communication delay, the optimization result of the positioning algorithm cannot reach stability;

[0132] Step S2.3.4: Therefore, in practical application, we set a distance error threshold ∈ D , since the total energy H of the system is a monotonically decreasing function; therefore, in the iterative calculation process, based on formula (4) and formula (6), the current total energy H of the system is compared with the preset energy threshold ∈ D Make comparisons;

[0133] If the total energy H of the system is greater than the preset energy threshold ∈ D , then perform iterative calculation according to formula (6);

[0134] If the total energy H of the system is less than the preset energy threshold ∈ D , then obtain the estimated position of the corresponding ranging module according to the current system total energy H

[0135] In a specific embodiment, in a robot hardware experiment, such as Figure 8As shown, the visual sensor is used to detect the target, wherein the experimental hardware configuration and the installation position of the ranging module are described. The fisheye camera is combined with a deep convolutional neural network to achieve detection. The position information of the monitored target is obtained in step S3, including

[0136] Step S3.1: Obtain a 360-degree panoramic image within the detection range through a fisheye camera installed on the top of the mobile node;

[0137] Step S3.2: radially expanding the panoramic image through a radial transformation to remove image distortion;

[0138] Step S3.3: Obtaining the position of the target detection box in the radially expanded panoramic image based on the deep convolutional neural network;

[0139] Step S3.4: Calculate the coordinates of the target location based on the location of the target detection frame in the panoramic image.

[0140] The formula for calculating the target location coordinates is:

[0141]

[0142] Where: (x t ,y t ) represents the coordinates of the global image where the target is located; d represents the function between the target detection box area and the target distance; a represents the area of ​​the target detection box; W represents the width of the panoramic image; x C Represents the horizontal coordinate of the center point of the target detection frame; θ t Indicates the position of the target relative to the fisheye camera.

[0143] That is, assuming that the image width is W, the x-axis coordinate of the center point of the target detection frame is x C , then the orientation of the target relative to the camera can be expressed as θ t =x C / W*360-180. If the target is in the center of the image, the target azimuth is 0 degrees. If the target is at the left edge of the image, the azimuth is -180 degrees. If the target is at the right edge of the image, the azimuth is +180 degrees. The distance calculation of the target can use nonlinear fitting to obtain the functional relationship between the target box area and the target distance. According to a large amount of data statistics, the calculation function of the target distance can be expressed as d=-1.1*a+16.53, where a is the area of ​​the target detection box; according to the upper description, the coordinates of the target can be obtained as

[0144] In a specific embodiment, the hierarchical clustering method is used in step S4 to fuse the location information of the monitored target obtained by each mobile node, specifically:

[0145] Step S4.1: Each mobile node broadcasts the location information of one or more monitored targets acquired in its own coordinate system to other mobile nodes in real time via wireless communication of the UWB ranging module;

[0146] Step S4.2: The current mobile node integrates the acquired location information of all monitored targets;

[0147] The information fusion strategy is to convert the location information of the monitored target obtained by the other nodes into the location information in the current node coordinate system;

[0148] Compare the difference between the converted position information of the monitored target and the position information of the monitored target detected by the current node with a preset distance error threshold;

[0149] If the position difference is greater than a preset distance error threshold, the converted position information of the monitored target is not retained;

[0150] If the position difference is less than a preset distance error threshold, clustering the converted position information of the monitored target with the position information of the monitored target detected by the current node to obtain one or more target position data classes;

[0151] Step S4.3: averaging the acquired position information of each target position data class to obtain optimized target position information;

[0152] The optimized target position information closest to the center of the mobile sensor network structure model is used as target tracking information.

[0153] Figure 11 This paper demonstrates the data fusion process using hierarchical clustering. Assume that a sensor network contains three nodes, each detecting three targets. x1, x2, and x3 represent the location information of three targets from different nodes, and the distance between these three data points is less than a distance threshold. Therefore, these three data points can be grouped together. Similarly, the final clustering result is three targets. Based on this result, the location information of these three targets can be obtained through mean calculation. The experimental results below demonstrate that this data fusion method based on hierarchical clustering can provide accurate target location information for group tracking.

[0154] In a specific embodiment, Figure 10 As shown, in step S4.2, the position information of the monitored target obtained by other nodes is converted into the position information in the current node coordinate system. The conversion formula is:

[0155] p ti =p tjR(θ i -θ j )+(p i -p j ) (10)

[0156] Where: p ti represents the location information of the target in the coordinate system of mobile node i; p tj represents the target location information of node j received by node i; θ i and θ j represents the direction of mobile node i and mobile node j; p i With p j Represents the current position coordinates of the mobile node; R(θ i -θ j ) represents the rotation matrix in K-dimensional space.

[0157] UWB technology utilizes wireless carrier communications with frequency bandwidths exceeding 1 GHz, transmitting data using narrow, non-sinusoidal pulses in the nanosecond to picosecond range. Due to its extremely short pulse duration, UWB achieves ultra-wideband spectrum: bandwidths exceeding 500 MHz. UWB offers significant advantages, including strong penetration, low power consumption, robust multipath mitigation, high security, reduced system complexity, and precise positioning accuracy.

[0158] The force-directed algorithm is a physics-based optimization algorithm that uses the interactions between charged particles to optimize multidimensional nonlinear programming problems. In this algorithm, each variable is treated as a charged particle, assigned a random position and velocity, and each variable has a charge. During the algorithm's execution, each variable is subject to the repulsive and attractive forces of other variables, resulting in acceleration and displacement.

[0159] Specifically, repulsive and attractive forces are determined by both distance and charge. The closer the two variables are, the greater the repulsive force between them; the greater the attractive force between them when the charges of the two variables have opposite signs. Through continuous iterative calculations, force-guided algorithms can find optimal or locally optimal solutions to problems. Force-guided algorithms are applicable to a variety of optimization problems, such as function optimization, combinatorial optimization, and machine learning. Compared to other optimization algorithms, force-guided algorithms offer advantages such as ease of implementation, rapid convergence, and strong robustness. The principle behind force-guided localization algorithms based on wireless sensors (UWB) is to connect the ranging modules of a sensor network into a spring network using virtual springs. The distance information obtained by the UWB ranging modules serves as the default length of the springs connecting the nodes. Initially, each ranging module obtains an estimated position and a corresponding estimated distance matrix through random initialization. This estimated distance matrix inevitably deviates from the distance matrix obtained by UWB ranging. The estimated distance between the two ranging modules is compared with the actual distance. When there is a difference, a contraction force or a stretching force will be generated between the ranging modules. Under the guidance of these forces, all ranging modules interact with each other and eventually reach a state of equilibrium, that is, the swarm robot achieves precise node positioning. Specifically, by installing UWB positioning tags on objects, real-time measurement of the position and motion trajectory of the object can be achieved, and by fusing the UWB measurement results with the estimation results of the force guidance algorithm, more accurate positioning results can be obtained. At the same time, image information is the information carrier with the largest amount of information exposure, and the deep learning algorithm extracts image features to the greatest extent and uses them to identify and track targets. This patent uses a deep learning-based visual sensor to enable swarm robots to have the ability of collaborative visual tracking, and finally uses a hierarchical clustering method to achieve visual information fusion between nodes and target position determination.

[0160] This paper also uses simulation experiments to test the force-guided positioning algorithm; this paper uses Python programming language to construct a two-dimensional space and randomly deploys four nodes in the virtual two-dimensional space. Figure 2 As shown in the figure on the left, the circles represent the actual position of the ranging module in the sensor network. Every three circles form a group of equilateral triangles, and each equilateral triangle represents a node in the sensor network. The solid triangle represents the estimated position of the ranging module in the initial state. The star symbol represents the estimated position when the force-guided positioning algorithm reaches a stable state. The dotted line in the figure represents the moving trajectory of the estimated position under the action of the force-guided positioning algorithm. Figure 2 As can be seen, the estimated positions are initially randomly distributed in two-dimensional space and differ greatly from the true positions. Under the action of the force-guided localization algorithm, the estimated positions gradually approach the true positions and eventually coincide with them. Figure 3The curve describes the trend of the total energy H of the system changing with time. The x-axis is the number of iterations (time) and the y-axis is the total energy of the system. Figure 3 It can be seen that under the action of the force-guided positioning algorithm, the total energy of the system continues to decrease and eventually reaches a stable state. This experimental result is consistent with the system stability proof results described above. It further proves that force-guided positioning can achieve distributed sensor network node positioning;

[0161] The force-guided positioning algorithm proposed in this paper actually uses the Euclidean distance matrix to restore the node position information. In the field of mathematical optimization research, there are many methods to solve similar problems, among which the more classic algorithms are MDS matrix decomposition and nonlinear programming. This paper uses simulation experiments to compare the performance of MDS, nonlinear programming and force-guided positioning algorithms. The main indicators examined are calculation speed and positioning error. In the simulation experiment, the three positioning algorithms were tested on a sensor network consisting of 1 to 10 nodes under the same initial state conditions, and the results were obtained. Figure 4 The results shown.

[0162] The simulation experiment was conducted in a virtual two-dimensional space constructed using Python. Initially, 10 sensor nodes (rectangles) were randomly distributed at the center of the two-dimensional space. The tracked target (triangle) was placed 15 meters from the center of the experimental space and moved sinusoidally to the right of the space. Figure 5 The solid line in the middle represents the target's trajectory, and the ten dashed lines represent the sensor node's trajectory. The trajectories demonstrate that, under the guidance of the swarm controller, the sensor network nodes maintain formation and continuously track the target. During this motion, each sensor node utilizes a force-guided localization algorithm for collaborative positioning with other nodes, and the resulting positioning is used to implement swarm control. To simulate realistic UWB ranging errors, a 0.1-meter error is added to the ranging results of each ranging module. The simulation results demonstrate that, despite these errors, the force-guided localization algorithm can still provide accurate node position information for swarm control, ensuring the stability of both swarm control and tracking.

[0163] In order to test the obstacle avoidance ability of the group, three obstacles were randomly set up in the experimental field. Figure 5 Each sensor node uses separation control force to keep a distance from obstacles. Figure 5The diagram shows a sensor network traversing obstacle number 1. The node trajectories in the diagram demonstrate that the swarm effectively avoids obstacles, achieving swarm obstacle avoidance control. The triangles in the diagram represent tracked targets, and the black hexagons represent obstacles. The rectangles represent sensor nodes, and the three circles within each rectangle represent the ranging modules installed on the sensor node. The X marks the positioning result obtained by the sensor node using the force-guided localization algorithm. Because the force-guided localization algorithm is a distributed algorithm, the positioning result calculated by each sensor node is different. Figure 6 The figure shows the positioning result of a randomly selected node. As can be seen, ranging errors and computational delays lead to positioning errors in the force-guided localization algorithm. However, this error does not significantly impact swarm control. Under the influence of the swarm controller, force-guided localization not only prevents collisions between nodes but also enables swarm obstacle avoidance and continuous target tracking.

[0164] Figure 7 Describes the performance parameters of the system. Figure 7 (a, b, c) are the positioning error, the minimum distance between nodes, and the distance from the group center to the target, respectively. Figure 7 In (a), the y-axis is the mean positioning error of all nodes in the sensor network, and the x-axis is time (in seconds). As shown in the figure, this round of experiment lasted 300 seconds. During the experiment, the mean positioning error remained around 0.1 meters, with a maximum error of 0.25 meters. Figure 7 (b) shows the minimum distance between nodes in the sensor network. As shown in the figure, the distance between nodes remained around 2 meters throughout the experiment, which is the same as the separation force threshold set in the simulation. Although the distance between nodes fluctuated slightly due to obstacles during the experiment, the minimum distance between individuals remained above 1 meter, and no collisions occurred. Figure 7 (c) shows the distance between the swarm center and the target. At the beginning of the experiment, the sensor network was approximately 15 meters away from the target. As the experiment progressed, the nodes began to track the target, driven by force-guided localization and the swarm control algorithm. Finally, around 100 seconds later, the front-end node of the sensor network caught up with the target. Thereafter, the distance between the sensor network and the target remained approximately 6 meters. The force-guided localization algorithm provides relatively accurate positioning information for swarm control, ensuring efficient target tracking while maintaining swarm control of the sensor network system.

[0165] This embodiment uses three omnidirectional mobile robots as sensor mobile nodes. Figure 8As shown, the robot chassis uses Maxnum wheels driven by four encoder motors. The robot's control core is a Nvidia-TX2 edge computing chip. Its primary functions are force-guided positioning, target recognition, group control calculations, and the generation of control signals for the chassis controller. The ranging module used for force-guided positioning is LinktrackS. This module can achieve precise ranging within a range of 50 meters, with an average error of ±0.1 meter. Furthermore, LinktrackS uses a distributed communication protocol, enabling high-speed distributed network communication without the need for a central controller. To enable obstacle avoidance in the sensor network, each sensor node is equipped with a Silan S2L lidar, which uses Time of Flight (ToF) technology to achieve centimeter-level obstacle detection. It has a scanning frequency of 15Hz, a detection range of 18 meters, and a maximum measurement resolution of 0.2 degrees.

[0166] This embodiment uses a fisheye panoramic camera combined with a YOLO deep convolutional neural network. The fisheye camera used in this experiment has a viewing angle of 210 degrees and a resolution of 5 million pixels. To achieve 360-degree all-round target detection, the camera is placed horizontally on top of the mobile platform; Figure 9 This article describes the specific process of using a fisheye camera to obtain target location information.

[0167] In this specific embodiment, the robot experiment was conducted at the Dalian Maritime University open-air roller skating rink. The test site was 20 meters by 20 meters and the experiment lasted approximately 20 minutes. For the complete experimental process, please refer to the video link in the appendix. During the experiment, three robots were used as mobile sensor network nodes, and one experimenter was the tracked target, participating in the sensor network tracking test. The main purpose was to test the target tracking capability of the sensor network. Due to the performance limitations of the hardware encoder motor, the maximum movement speed of the sensor network node is 1 meter per second, so the movement speed of the target is also controlled within 1 meter per second. Figure 12 is the moving trajectory of the target and sensor network in the experiment. Figure 12 The solid line in the middle represents the target's trajectory, and the dashed lines represent the trajectories of the three nodes in the sensor network. Experimental results show that the sensor network's trajectory is largely consistent with the trajectory of the tracked target. The figure also uses three groups of hollow circles to depict the sensor network's state at t = 5s, t = 18s, and t = 35s. As can be seen in the figure, the three mobile nodes maintain a stable grouping during target tracking, constantly changing their formation to adapt to environmental changes.

[0168] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the above embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the above embodiments, or replace some or all of the technical features therein with equivalents. However, these modifications or replacements do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention.

Claims

1. A target tracking method for mobile sensor networks based on force-guided positioning, characterized in that: The following steps are involved: Step S1: Set up several robots as mobile nodes and establish a mobile sensor network structure model; Step S2: Optimizing the positioning of the mobile node according to the force-directed algorithm and updating the mobile sensor network structure model; In S2, the mobile node is optimized and positioned according to the force-directed algorithm, specifically: Step S2.1: multiple UWB ranging modules are set on the mobile nodes, and relative distance information between any two mobile nodes is measured by the UWB ranging modules; and an actual distance matrix D is obtained based on the relative distance information; Randomly initialize the ranging module to obtain the estimated position of all ranging modules in the mobile sensor network structure model Step S2.2: Based on the estimated position of the ranging module Calculate the estimated distance matrix Obtaining the estimated distance matrix The calculation formula is in, Represents the estimated distance matrix The Euclidean distance between the i-th ranging module and the j-th ranging module in ; represents the estimated position of the i-th ranging module; represents the estimated position of the jth ranging module; Step S2.3: Estimated position of each ranging module based on the force guidance algorithm Establish a mobile sensor network structure model, select a set of estimated positions of ranging modules that optimize the system energy of the mobile sensor network structure model Step S2.4: Estimated position based on the selected optimized ranging module Finally, the actual position x of the ranging module corresponding to the actual distance matrix D is obtained; The calculation formula for the actual position x of the ranging module is: Where: Q represents an orthogonal K×K matrix; T represents an n×K translation matrix; Calculating the posture of the mobile node according to the actual position x to update the mobile sensor network structure model; Step S3: sampling the monitored target according to the sensor equipped by each mobile node to obtain the location information of the monitored target; Step S4: the mobile node transmits the acquired location information of the monitored target to other mobile nodes except itself; Performing information fusion on the location information of the monitored target obtained by each of the mobile nodes through a hierarchical clustering method; Step S5: each mobile node in the updated mobile sensor network structure model realizes target tracking according to the position information of the monitored target after information fusion.

2. The method for target tracking in a mobile sensor network based on force-guided positioning according to claim 1, characterized in that: The estimated positions of a group of ranging modules selected in step S2.3 that optimize the system energy of the mobile sensor network structure model Specifically Step S2.3.1: Assume that when ranging module i and ranging module j move away from or towards each other under the action of the guiding force, the potential energy H generated by the guiding force ij for Where: d ij Indicates the actual value obtained by UWB ranging; x i 、x j ∈D and d ij =||x i -x j ||; It's about x i ,x j The distance function, and Step S2.3.2: The potential energy H generated by the guiding force ij Obtain the total system energy of the mobile sensor network structure model; The total energy H of the system is a monotonically decreasing function, and the calculation formula is Where: Represents a collection of ranging modules; represents the Frobenius norm of the matrix; i and j represent the ranging module; Step S2.3.3: Definition and Iterative control equations for ranging modules in mobile sensor networks; Where: Represents the estimated distance matrix Elements in It means exist Component in the direction of the vector; It means exist The component of the vector direction; Step S2.3.4: Based on formula (4) and formula (6), the total energy H of the current system is compared with the preset energy threshold ∈ D Make comparisons; If the total energy H of the system is greater than the preset energy threshold ε D , then perform iterative calculation according to formula (6); If the total energy H of the system is less than the preset energy threshold ε D , then obtain the estimated position of the corresponding ranging module according to the current system total energy H 3. The method for target tracking in a mobile sensor network based on force-guided positioning according to claim 1, characterized in that: In step S2.4, the posture of the mobile node is calculated according to the actual position x. The posture calculation formula is: Where: P i represents the posture information of node i; P 0i Represents the location information of node i; θ i represents the direction of node i; x i A,x i B,x i C∈x i They represent the position of the ranging module installed on node i; and the direction of node i is x i C is the endpoint pointing to x i A and x i The vector of the midpoint of the line B.

4. The method for target tracking in a mobile sensor network based on force-guided positioning according to claim 1, characterized in that: The step S3 of obtaining the location information of the monitored target includes: Step S3.1: Obtain a 360-degree panoramic image within the detection range through a fisheye camera installed on the top of the mobile node; Step S3.2: radially expanding the panoramic image through a radial transformation to remove image distortion; Step S3.3: Obtaining the position of the target detection box in the radially expanded panoramic image based on the deep convolutional neural network; Step S3.4: Calculate the coordinates of the target location based on the location of the target detection frame in the panoramic image.

5. The method for target tracking in a mobile sensor network based on force-guided positioning according to claim 4, characterized in that: The formula for calculating the target location coordinates in step S3.4 is: Where: (x t ,y t ) represents the coordinates of the global image where the target is located; d represents the function between the target detection box area and the target distance; a represents the area of ​​the target detection box; W represents the width of the panoramic image; x C Represents the horizontal coordinate of the center point of the target detection box; θ t Indicates the position of the target relative to the fisheye camera.

6. The method for target tracking in a mobile sensor network based on force-guided positioning according to claim 1, characterized in that: In step S4, the location information of the monitored target obtained by each mobile node is fused by the hierarchical clustering method; specifically, Step S4.1: Each mobile node broadcasts the location information of one or more monitored targets acquired in its own coordinate system to other mobile nodes in real time via wireless communication of the UWB ranging module; Step S4.2: The current mobile node integrates the acquired location information of all monitored targets; The information fusion strategy is to convert the location information of the monitored target obtained by the other mobile nodes into the location information in the current node coordinate system; Compare the difference between the converted position information of the monitored target and the position information of the monitored target detected by the current node with a preset distance error threshold; If the position difference is greater than a preset distance error threshold, the converted position information of the monitored target is not retained; If the position difference is less than a preset distance error threshold, clustering the converted position information of the monitored target with the position information of the monitored target detected by the current node to obtain one or more target position data classes; Step S4.3: averaging the acquired position information of each target position data class to obtain optimized target position information; The optimized target position information closest to the center of the mobile sensor network structure model is used as target tracking information.

7. The method for target tracking in a mobile sensor network based on force-guided positioning according to claim 6, characterized in that: In step S4.2, the location information of the monitored target obtained by other mobile nodes is converted into the location information in the current node coordinate system. The conversion formula is: p ti (p tj R(θ i -θ j )+(p i -p j ) (10) Where: p ti represents the location information of the target in the coordinate system of mobile node i; p tj represents the target location information of node j received by node i; θ i and θ j represents the direction of mobile node i and mobile node j; p i With p j Represents the current position coordinates of the mobile node; R(θ i -θ j ) represents the rotation matrix in K-dimensional space.

Images