Target tracking method and tracking system applied to intelligent water cannon

Abstract

The invention relates to the technical field of ship intelligent equipment, in particular to a target tracking method and tracking system applied to an intelligent water cannon. The method comprises the following steps of: performing multi-source data acquisition on a target and a target environment by adopting a sensor, preprocessing the collected multi-source data, and fusing multi-source image data by adopting a data fusion technology; extracting target features from the multi-source image data by adopting a target extraction algorithm; a multi-target tracking model is established, target features extracted from the multi-source image data are tracked in real time, a multi-target tracking control strategy is introduced, and tracking priority ranking is performed on multiple target points; and the water cannon execution mechanism realizes steering and water flow injection control of the water cannon based on a PID control algorithm. The influence coefficient kv of the water flow jetting speed and the influence coefficient kp of the priority score on the water flow jetting speed serve as compensation items of a PID control algorithm, the influence of the water flow jetting speed on the steering precision of the water cannon is corrected, and the control precision of the water cannon is improved.

Description

Technical Field

[0001] The present invention relates to the technical field of ship intelligent equipment, and in particular to a target tracking method and a tracking system applied to an intelligent water cannon. Background Art

[0002] Smart water cannon is a precision mechanical manufacturing technology equipment that integrates modern information technology and automatic control technology. It is usually used to automatically detect, locate and extinguish fires. It has the characteristics of fast response speed, accurate positioning and high fire extinguishing efficiency. Especially during navigation at sea, if a fire occurs, the ship can quickly spray a large amount of water to extinguish the fire through the equipped smart water cannon to ensure the safety of the ship and personnel.

[0003] However, in the multi-target tracking scenario at sea, each target has its own motion characteristics, and targets may block or collide with each other, which makes the target movement unpredictable and increases the difficulty of tracking the target by the intelligent water cannon.

[0004] At the same time, when water is ejected from a water cannon, a reaction force is generated, which will produce a torque on the water cannon itself. If the water jet velocity is high, the reaction force generated will also be large, thereby affecting the stability and steering accuracy of the water cannon. Therefore, a target tracking method and tracking system for an intelligent water cannon are provided. Summary of the invention

[0005] The purpose of the present invention is to provide a target tracking method and a tracking system for an intelligent water cannon, so as to solve the problem raised in the above background technology that in a multi-target tracking scenario, each target has its own motion characteristics, and the targets may block or collide with each other, which makes the movement of the target unpredictable and increases the difficulty of tracking the target by the intelligent water cannon.

[0006] To achieve the above object, the present invention provides a target tracking method for an intelligent water cannon, comprising the following steps:

[0007] S1. Use sensors to collect multi-source data of the target and target environment, pre-process the collected multi-source data, and use data fusion technology to fuse the multi-source image data;

[0008] S2, using target extraction algorithm to extract target features from multi-source image data;

[0009] S3. Establish a multi-target tracking model to track the target features extracted from multi-source image data in real time, and introduce a multi-target tracking control strategy to sort the tracking priorities for multiple target points;

[0010] S4, sending the position information of the tracked target point to the water cannon actuator, which controls the steering and water jetting of the water cannon based on the PID control algorithm;

[0011] Among them, the influence of water flow velocity on the steering accuracy of the water cannon is introduced into the PID control algorithm to achieve the optimal control of the steering and water flow injection of the water cannon.

[0012] As a further improvement of the technical solution, the sensor includes at least a visual sensor and an infrared sensor;

[0013] The visual sensor is used to collect video streams containing visual features of the target and the target environment. The video streams are used to provide information about the contour of the surrounding environment, the contour of the target, the movement trajectory and speed of the target, the surface texture features of the target, and the distance between the target and the sensor.

[0014] Infrared sensors are used to detect the infrared characteristics of the target environment, which specifically include heat source location information and thermal radiation intensity.

[0015] As a further improvement of the technical solution, in S1, the multi-source image data is fused using data fusion technology, and the specific steps involved in the data fusion technology are:

[0016] S1.1, using ICP algorithm to align the preprocessed visual features and infrared features in the same spatial frame; align the target contour information captured by the visual sensor and the temperature distribution information captured by the infrared sensor in the same coordinate system, so that the information of both can be combined for more accurate target tracking and positioning;

[0017] The specific alignment steps are as follows:

[0018] Convert edge features of visible light images into point cloud form;

[0019] For infrared images, extract temperature distribution features and convert them into corresponding point clouds;

[0020] Based on the initial alignment, the ICP algorithm is used to find the best rotation and translation to minimize the distance between the point cloud of the visible light image and the point cloud of the infrared image.

[0021] The transformation is iteratively updated until the predetermined convergence condition is reached. The ICP algorithm gradually reduces the difference between the two sets of point clouds by iteratively finding the nearest neighbor point pairs and estimating the optimal transformation, thereby completing the alignment of visual features and infrared features in the same spatial frame, ensuring that data from different sensors can be compared and fused in the same spatial coordinate system.

[0022] S1.2, merging the visual features and infrared features aligned in the same spatial frame to construct a complete feature vector F;

[0023] S1.3, using a recursive feature elimination algorithm to filter out a feature subset M in the feature vector F;

[0024] S1.4. Different weights are assigned to each feature according to its importance in the feature subset M, and a weighted feature vector F is obtained that combines visual features and infrared features. w ;

[0025] S1.5. Use the Dempster-Shafer evidence theory algorithm to integrate the information of visual and infrared sensors and calculate the values ​​of the trust function Bel and the likelihood function Pl.

[0026] As a further improvement of the technical solution, in S2, the target extraction algorithm is used to extract target features from multi-source image data, and the specific steps involved are:

[0027] Prepare a training data set, including target samples and non-target samples. The target samples contain the feature vector of the heat source, and the non-target samples do not contain the feature vector of the heat source. For each target sample, there is a positive class label y i = +1, each non-target sample has a negative class label y i = -1;

[0028] Select the kernel function K and regularization parameter C;

[0029] Known weighted eigenvector F w , and each feature vector has a corresponding label y, using the weighted feature vector F w And the corresponding target label y trains the SVM model, where the target label y represents the category to which the feature vector belongs;

[0030] Use the trained SVM model to classify multi-source image data to detect and separate heat sources:

[0031]

[0032] In the formula, y(F w ) represents the predicted weighted feature vector F including the data points w The output label of; sign(.) represents the sign function, which returns the sign of a data point. If Then sign(.) returns +1, indicating that the data point is classified as a target; if Then sign(.) returns -1, indicating that the data point is classified as non-target; N represents the total number of support vectors; a irepresents the Lagrange multiplier; y i represents the label of the i-th support vector in the training data set; K(F w，i , F w ) represents the kernel function, which is used to calculate the new data point F w With the support vector F w，i The similarity or distance between them; i represents the index variable; F w，i represents the weighted feature vector of the i-th support vector in the training data set; F w represents the weighted feature vector of the new input data point, that is, the data point that needs to be classified; b represents the bias term.

[0033] As a further improvement of the technical solution, the multi-target tracking model is based on a Kalman filter and is used to track multiple target features extracted from multi-source image data in real time. The specific steps involved are:

[0034] S3.1. For each target detected for the first time, use the SVM classifier to confirm the target and initialize the basic information of the target's position, size, and shape characteristics;

[0035] S3.2. In subsequent frames, extract the corresponding weighted feature vector F for each target t , calculate the weighted feature vector F of the current frame t and the weighted feature vector F of the previous frame t―1 The similarity between two feature vectors is quantified by Euclidean distance.

[0036] S3.3, construct a cost matrix, and include the deviation between the target's motion model prediction and the actual observation, and the similarity between the target and the observation into the cost matrix, and obtain the final cost matrix C′ ij , and apply the Hungarian algorithm to find matching solutions for multiple target features;

[0037] S3.4, use the state transfer equation to predict the target state at the next moment;

[0038] X t|t―1 =AX t―1 +Bu t―1 +W t―1 ;

[0039] Where, X t|t―1 represents the predicted value of the state vector at time t given time t-1; X t―1 represents the system state vector at time t-1; A represents the state transfer matrix, which is used to describe the change of system state over time; B represents the control input matrix, which is used to describe the influence of external control input on the state; u t―1 represents the control vector at time t-1; W t―1represents process noise;

[0040] P t|t―1 =AP t―1 A T +Q;

[0041] Where P t|t―1 represents the predicted state covariance matrix, that is, the state covariance prediction matrix when the information at time t-1 is given at time t; Q represents the process noise covariance matrix; A T represents the transposed matrix of the state transfer matrix A; P t―1 represents the state covariance matrix at the previous time t-1;

[0042] S3.5, according to the weighted feature vector F of the current frame t and the weighted feature vector F of the previous frame t―1 The similarity calculation results between them are used to correct the predicted results for each target using the observation equation to obtain a new state estimate, which is used to update the target state information in real time.

[0043] Among them, the specific expression of the observation equation is:

[0044] Z t =HX t +V t ;

[0045] In the formula, Z t represents the observation vector at time t; H represents the observation matrix, which maps the system state vector to the observation space; X t represents the system state vector at time t; V t Represents observation noise, which is used to reflect the uncertainty of sensor measurement;

[0046] Then calculate the Kalman gain K t , and update the state estimate:

[0047] K t =P t|t―1 H T (HP t|t―1 H T +R) ―1 ;

[0048] X′ t =X t|t―1 +K t (Z t ―HX t|t―1 );

[0049] Finally, update the state covariance matrix:

[0050] P′ t =(I-Kt H)P t|t―1 ;

[0051] In the formula, H T represents the transpose of the observation matrix H; X′ t represents the updated state estimation vector; R represents the observation noise covariance matrix; P′ t represents the updated state covariance matrix; I represents the identity matrix.

[0052] As a further improvement of this technical solution, in S3.3, the final cost matrix C′ ij The specific expression is:

[0053] C′ ij =C ij +a·p ij ;

[0054] in:

[0055] C ij = distance(z i , x i );

[0056]

[0057] Where a represents the prediction deviation p ij The weight factor of ij represents the basic cost; C′ ij represents the final cost matrix; p ij represents the prediction deviation; distance represents the Euclidean distance; represents the predicted state of target j at the current time t; z i represents the observation value of target i, and represents the observation vector Z at time t t The i-th element of i represents the state vector of target i predicted by the model, and represents the state vector X at time t t The ith element of .

[0058] As a further improvement of the technical solution, in S3, a multi-target tracking control strategy is introduced to sort the tracking priorities of multiple target points. The specific steps involved are:

[0059] S3.6. Obtaining the initial thermal radiation intensity T of the target point through multi-source image data i (0), target point area A i (0), target point heat source combustion efficiency E i (0) and the temperature distribution D of the target point heat source i (0), and calculate the priority score according to the quantitative scoring algorithm;

[0060] S3.7, use the quick sort algorithm to sort all target points from high to low priority scores, and sort multiple target points in descending order;

[0061] S3.8, select the target with the highest priority from the sorted list for tracking;

[0062] S3.9. When new temperature data arrives, repeat steps S3.6 to S3.8 to keep the priority sorting up to date.

[0063] As a further improvement of the technical solution, in S3.6, there are n target points, and at the initial moment, each target point has a thermal radiation intensity T i (0), target point area A i (0), target point heat source combustion efficiency E i (0) and the temperature distribution D of the target point heat source i (0), since the state of the target point changes with time, the time variable t is introduced into the quantitative scoring algorithm, and the priority score is calculated. The priority score is used to determine the threat level of the target. The specific expression involved is:

[0064] P(T i (t),A i (t),E i (t),D i (t),t)=w T (t)·T i (t)+w A (t)·A i (t)+w E (t)·E i (t)+w D (t)·D i (t);

[0065] Where, T i (t) represents the thermal radiation intensity at time t; A i (t) represents the target point area at time t; E i (t) represents the combustion efficiency at time t; D i (t) represents the temperature distribution at time t; w T (t) represents the weight parameter of thermal radiation intensity at time t; w A (t) represents the weight parameter of the target point area at time t; w E (t) represents the weight parameter of combustion efficiency at time t; w D (t) represents the weight parameter of temperature distribution at time t; t represents time;

[0066] The temperature distribution D(t) of the target point heat source is obtained by analyzing the flame image. In the infrared image, the temperature distribution of different parts of the heat source can be extracted, and a comprehensive temperature distribution score can be given based on this;

[0067] in:

[0068] T i (t) = T i (0)+ΔT i t;

[0069] A i (t) = A i (0)+ΔA i t;

[0070] E i (t) = E i (0)+ΔE i t;

[0071] D i (t) = D i (0)+ΔD i t;

[0072] In the formula, ΔT i Indicates the change in thermal radiation intensity over time t; ΔA i Indicates the change in the target point area over time t; ΔE i Indicates the change in combustion efficiency over time t; ΔD i Represents the change of temperature distribution over time t;

[0073] And the weight parameters also change over time, as follows:

[0074] x T (t) = γ + δ·t;

[0075] w A (t) = ∈ + μ·t;

[0076] w E (t) = α + β·t;

[0077] w D (t) = η + θ·t;

[0078] Among them, γ, δ,∈, μ, α, β, η, and θ are constants with initial values ​​set according to expert opinions, which are used to describe how different factors change over time and the importance of these factors for priority assessment.

[0079] As a further improvement of the technical solution, in S4, the steering and water jetting control of the water cannon is realized by a PID control algorithm, and the specific steps involved are:

[0080] S4.1. Initialize the parameter K of the PID control algorithm p , K i , K d , and the influence coefficient k of water jet velocity v The influence coefficient k of the priority score on the water jet velocity p As a compensation item;

[0081] S4.2. Based on the feature subset M of the target point, obtain the target position r i (t) and the current position of the water monitor y i (t), and calculate the error e i (t);

[0082] S4.3, using PID control algorithm to calculate PID control output signal;

[0083]

[0084] In the formula, u i (t) represents the PID control output signal of the i-th target at time t; K p K represents the proportional gain, which is used to determine the direct impact of the error e(t) on the control signal u(t); i Represents the integral gain, which is used to eliminate steady-state errors and accumulate past errors; K d represents the differential gain, which is used to reduce overshoot and improve the response speed of the system; e i (t) represents the error signal of the i-th target at time t; e i (τ) represents the error signal of the i-th target at a time point τ in the past, reflecting the difference between the actual position and the expected position of the target; represents the error e from time 0 to time t i The integral of (τ); Indicates the error e i (t) rate of change at time t; P i Indicates the priority score of the target;

[0085] Among them, e i (t) = r i (t)―y i (t); k v k P P i Indicates the water jet velocity v i Compensation item for the influence on the steering accuracy of water cannon;

[0086] S4.4, the calculated control signal u i (t) Applied to the water cannon actuator (servo motor) to adjust the angle of the water cannon;

[0087] S4.5. Score P according to the priority of the target i , adjust the water jet velocity v i ;

[0088] S4.6. Repeat steps S4.2 to S4.5 in the next sampling cycle.

[0089] On the other hand, the present invention provides a target tracking system for an intelligent water cannon, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement any of the steps of the target tracking method for an intelligent water cannon described above.

[0090] Compared with the prior art, the present invention has the following beneficial effects:

[0091] 1. In the target tracking method and tracking system applied to intelligent water cannon, in the multi-target tracking model, the deviation between the target's motion model prediction and the actual observation and the similarity between the target and the observation are incorporated into the cost matrix to obtain the final cost matrix C′ ij , more accurately matches targets and observations, thereby improving tracking accuracy, enabling the system to handle target tracking not only in static environments, but also in dynamically changing environments, helping to distinguish different targets in complex scenes and avoid mismatching, thereby enhancing the adaptability of the system;

[0092] In the application of intelligent water cannon, the Hungarian algorithm is used to find matching targets, ensuring that the water cannon can quickly and accurately aim at the target and update the target's status information in real time.

[0093] 2. In the target tracking method and tracking system applied to intelligent water cannon, the influence of water flow velocity on the steering accuracy of water cannon is considered in the PID control algorithm, and the influence coefficient k of water flow velocity is v The influence coefficient k of the priority score on the water jet velocity p As a compensation item of the PID control algorithm, it is used to correct the influence of the water flow velocity on the steering accuracy of the water cannon, realize the optimized control of the steering and water flow injection of the water cannon, and improve the control accuracy of the water cannon. BRIEF DESCRIPTION OF THE DRAWINGS

[0094] Figure 1 The figure is a flow chart of the overall method of the present invention. DETAILED DESCRIPTION

[0095] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in 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. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0096] Embodiment 1:

[0097] See also Figure 1 As shown, this embodiment provides a target tracking method applied to an intelligent water cannon, comprising the following steps:

[0098] S1. Use sensors to collect multi-source data of the target and target environment, pre-process the collected multi-source data, and use data fusion technology to fuse the multi-source image data;

[0099] The sensor includes at least a visual sensor and an infrared sensor;

[0100] The visual sensor is used to collect video streams containing visual features of the target and the target environment. The video streams are used to provide information about the contour of the surrounding environment, the contour of the target, the movement trajectory and speed of the target, the surface texture features of the target, and the distance between the target and the sensor.

[0101] Use the Canny edge detection algorithm to extract the contour information of the surrounding environment and the contour information of the target, use the texture analysis algorithm (such as the gray-level co-occurrence matrix GLCM) to extract the texture features of the target, and use the structured light or stereo vision algorithm to obtain the distance information between the target and the sensor;

[0102] Infrared sensors are used to detect infrared characteristics of the target environment. The infrared characteristics specifically include the location information of the heat source and the intensity of thermal radiation. The heat source is not limited to the fire source, but also includes other human bodies or organisms that can emit heat.

[0103] Infrared sensors can detect infrared radiation emitted by objects and convert it into electrical signals, which are then used to generate images through processing systems. Infrared sensors can detect temperature changes in real time and quickly locate the location of heat sources.

[0104] The position of the heat source in the image is determined by the threshold segmentation method, and the numerical representation of the thermal radiation intensity is obtained by quantifying the temperature distribution of the heat source.

[0105] In this embodiment, the specific method for preprocessing multi-source data is: using Gaussian filtering algorithm to reduce image noise, using black body calibration method to perform temperature calibration on thermal image, ensuring temperature reading, correcting errors caused by factors such as ambient temperature change and sensor drift, so that the measured temperature value is as close to the true value as possible, wherein the black body is an idealized complete absorber that can be used to generate radiation energy of known temperature. By placing the black body in the field of view of the infrared camera, the measurement value of the camera can be calibrated, and by comparing the actual temperature of the black body standard source with the temperature measured by the infrared camera, the gain and bias of the camera are adjusted so that the measured value is consistent with the true value;

[0106] Use bilateral filtering algorithm to remove noise from infrared images and improve the accuracy of heat source detection;

[0107] In this embodiment, the multi-source image data is fused using data fusion technology, and the specific steps involved in the data fusion technology are:

[0108] S1.1, using ICP algorithm to align the preprocessed visual features and infrared features in the same spatial frame; align the target contour information captured by the visual sensor and the temperature distribution information captured by the infrared sensor in the same coordinate system, so that the information of both can be combined for more accurate target tracking and positioning;

[0109] The specific alignment steps are as follows:

[0110] Convert edge features of visible light images into point cloud form;

[0111] For infrared images, extract temperature distribution features and convert them into corresponding point clouds;

[0112] Based on the initial alignment, the ICP algorithm is used to find the best rotation and translation to minimize the distance between the point cloud of the visible light image and the point cloud of the infrared image.

[0113] The transformation is iteratively updated until the predetermined convergence condition is reached. The ICP algorithm gradually reduces the difference between the two sets of point clouds by iteratively finding the nearest neighbor point pairs and estimating the optimal transformation, thereby completing the alignment of visual features and infrared features in the same spatial frame, ensuring that data from different sensors can be compared and fused in the same spatial coordinate system.

[0114] S1.2, merge the visual features (visual features include contour, texture, distance and other information) and infrared features (infrared features include heat source location, thermal radiation intensity and other information) aligned in the same spatial frame to construct a complete feature vector F, where the feature vector F contains all the key information that helps target tracking and positioning;

[0115] S1.3, using a recursive feature elimination algorithm to filter out a feature subset M in the feature vector F, and filter the feature subset M to reduce the dimension and improve the generalization ability of the model;

[0116] M=[C,T,D,H p , H i ], where C represents the target contour feature vector; T represents the target texture feature vector; D represents the target and sensor distance feature vector; H p Represents the heat source position vector; H i represents the thermal radiation intensity vector;

[0117] The specific steps involved in screening the feature subset M using the recursive feature elimination algorithm are:

[0118] Set the SVM model and the number of features to be retained n, use all the features to train an SVM model, score the features according to the weights corresponding to the support vectors of the SVM model, remove multiple features with low scores, repeat the above steps until the number of remaining features is equal to n, and output the final selected feature subset M;

[0119] If w is the weight vector of the SVM model and X is the feature matrix, the importance score of the feature is expressed as:

[0120] Importance(X j )=|w j |;

[0121] In the formula, Importance(X j ) means that for feature X j Importance rating; X j represents the jth feature; w j Represents the weight corresponding to the jth feature;

[0122] In the recursive feature elimination (RFE) algorithm, each feature X j The corresponding weight w j It reflects the contribution of this feature to the target variable;

[0123] S1.4. Different weights are assigned to each feature according to its importance in the feature subset M, and a weighted feature vector F is obtained that combines visual features and infrared features. w ;

[0124] In the feature subset M = [C, T, D, H p , H i ], given the feature importance I(C), I(T), I(D), I(H p ), I(H i ), define a weighted eigenvector F w as follows:

[0125]

[0126] In the formula, w C represents the weight of the target object contour feature vector; w T represents the weight of the target object’s texture feature vector; w D Represents the weight of the feature vector of the distance between the target and the sensor; represents the weight of the heat source position vector; Represents the weight of the thermal radiation intensity vector;

[0127] By integrating visual features (such as contour, texture, distance) and infrared features (such as heat source location and thermal radiation intensity), the weighted feature vector F w It can better capture the diversity and dynamic characteristics of the target. Visual features provide rich appearance information, while infrared features provide temperature and heat distribution information, which is especially important for detecting heat sources. Combining the above two types of information can improve the accuracy and robustness of target tracking.

[0128] Weighted eigenvector F w As a feature vector that integrates multiple information, it can be directly used in subsequent classification, recognition and tracking algorithms. w The target features can be extracted and the target can be classified more accurately, thus improving the overall performance of target tracking;

[0129] S1.5, using the Dempster-Shafer evidence theory algorithm to integrate the information of the visual and infrared sensors and calculate the values ​​of the trust function Bel and the likelihood function Pl;

[0130] The specific steps involved in the Dempster-Shafer evidence theory algorithm are:

[0131] Define a recognition frame Θ, where Θ represents the possible location of the target;

[0132] Assign a basic probability assignment (BPA) to each piece of evidence provided by a sensor. BPA is a basic probability assignment function m:2 Θ →[0, 1], BPA is given a hypothesis set The probability mass, for each sensor, BPA reflects the degree of trust of the sensor in different hypotheses;

[0133] Dempster's combination rule is used to merge the BPAs from different sensors. The BPA of the visual sensor is m1 and the BPA of the infrared sensor is m2. Their combination m = m1 ⊕ m2 is defined as follows:

[0134]

[0135] Among them, K is the conflict coefficient, which indicates the degree of conflict between different BPAs and is used to measure the degree of contradiction between the evidence provided by different sensors. It is defined as:

[0136]

[0137] Wherein, m1(B) represents the basic probability distribution of the visual sensor to the hypothesis set B; m2(C) represents the basic probability distribution of the infrared sensor to the hypothesis set C; m(A) represents the distribution value of the basic probability distribution function m to the hypothesis set A, the hypothesis set A is a subset of the recognition framework Θ, in the intelligent water cannon target tracking problem, the hypothesis set A is a set of target positions; B represents the hypothesis set of the visual sensor; C represents the hypothesis set of the infrared sensor; and the sets A, B and C are all subsets on the recognition framework Θ;

[0138] From the combined BPA, calculate the trust function Bel and the likelihood function Pl:

[0139]

[0140]

[0141] Where Bel(A) represents the trust function of hypothesis set A; Pl(A) represents the likelihood function of hypothesis set A; V represents the set of all hypotheses that support hypothesis set A. In the trust function Bel, hypothesis set V belongs to set A. In the likelihood function Pl, hypothesis set V and set A have an intersection; m(V) represents the trust assigned by the basic probability distribution function m to hypothesis set V.

[0142] The Dempster-Shafer evidence theory algorithm (also known as the belief function theory) is used to deal with decision-making problems under incomplete information. It provides a framework to combine evidence from multiple sources to form a degree of confidence in a proposition;

[0143] In the application of smart water cannon, data from visual sensors and infrared sensors are converted into corresponding BPAs; visual sensors provide information about target shape, texture and other features, while infrared sensors provide information about the location and intensity of heat sources; the data of each sensor is converted into a BPA, and then these BPAs are fused together to form a more accurate target position estimate. The Dempster-Shafer evidence theory algorithm can effectively combine data from different sensors to improve the accuracy of target tracking.

[0144] S2, using target extraction algorithm to extract target features from multi-source image data;

[0145] In this embodiment, the target extraction algorithm is used to extract target features from multi-source image data, and the specific steps involved are:

[0146] Prepare a training data set, including target samples and non-target samples. The target samples contain the feature vector of the heat source, and the non-target samples do not contain the feature vector of the heat source. For each target sample, there is a positive class label y i = +1, each non-target sample has a negative class label y i = -1;

[0147] Select the kernel function K and regularization parameter C;

[0148] Known weighted eigenvector F w , and each feature vector has a corresponding label y, using the weighted feature vector F w And the corresponding target label y trains the SVM model, where the target label y represents the category to which the feature vector belongs;

[0149] Use the trained SVM model to classify multi-source image data to detect and separate heat sources:

[0150]

[0151] In the formula, y(F w ) represents the predicted weighted feature vector F including the data points w The output label of; sign(.) represents the sign function, which returns the sign of a data point. If Then sign(.) returns +1, indicating that the data point is classified as a target; if Then sign(.) returns -1, indicating that the data point is classified as non-target; N represents the total number of support vectors; a i represents the Lagrange multiplier; y i represents the label of the i-th support vector in the training data set; K(F w，i , F w ) represents the kernel function, which is used to calculate the new data point F w With the support vector F w，i The similarity or distance between them; i represents the index variable; F w，i represents the weighted feature vector of the i-th support vector in the training data set; F w represents the weighted feature vector of the new input data point, that is, the data point that needs to be classified; b represents the bias term.

[0152] The specific steps involved in training the SVM model are:

[0153] Set SVM parameters, including regularization parameter C and kernel function K;

[0154] Use the weighted feature vector F w and the corresponding target label y to train the SVM model;

[0155]

[0156] Constraint conditions:

[0157]

[0158] Map the weighted feature vector F w to a high-dimensional space;

[0159]

[0160] In the formula, w represents the weight vector; b represents the bias term; ξ i represents the slack variable; C represents the regularization parameter; y i represents the label of the i-th training sample; N represents the total number of training samples; represents the mapping function Map the weighted feature vector F w to the vector in the high-dimensional space; w T represents the transpose matrix of the weight vector w;

[0161] By introducing the Lagrange multiplier a i solve the dual problem to find the optimal a i , and solve to obtain the weight vector w and the bias term b;

[0162]

[0163]

[0164] where k is the support vector index that satisfies 0 < a k < C; a i represents the Lagrange multiplier; y k represents the label of the k-th training sample;

[0165] Save the trained SVM model and use the trained SVM model to perform classification prediction on new data.

[0166] S3. Establish a multi-object tracking model to perform real-time tracking on the target features extracted from multi-source image data, and introduce a multi-object tracking control strategy to perform tracking priority sorting for multiple target points;

[0167] Introducing a multi-object tracking control strategy can more accurately distinguish and identify targets, reduce the possibility of false alarms and missed detections, and improve the overall reliability;

[0168] In this embodiment, the multi-target tracking model is based on a Kalman filter and is used to track multiple target features extracted from multi-source image data in real time. The specific steps involved are:

[0169] S3.1. For each target detected for the first time, use the SVM classifier to confirm the target and initialize the basic information of the target's position, size, and shape characteristics;

[0170] Initialize template state vector X0 = [x0, y0, w0, h0] T , where (x0, y0) represents the center coordinates of the target, w0 represents the width of the target, and h0 represents the height of the target;

[0171] S3.2. In subsequent frames, extract the corresponding weighted feature vector F for each target t , calculate the weighted feature vector F of the current frame t and the weighted feature vector F of the previous frame t―1 The similarity between two feature vectors is quantified by Euclidean distance.

[0172] S3.3, construct a cost matrix, and include the deviation between the target's motion model prediction and the actual observation, and the similarity between the target and the observation into the cost matrix, and obtain the final cost matrix C′ ij , and apply the Hungarian algorithm to find matching solutions for multiple target features;

[0173] Including the deviation between the target's motion model prediction and the actual observation into the cost matrix can better match the target and the observation value, and maintain high tracking accuracy even when the target moves quickly or the environment is complex; Including the similarity between the target and the observation into the cost matrix can enhance the robustness of the model in the face of challenges such as occlusion, lighting changes, and perspective changes. Similarity measurement helps to distinguish different targets in complex scenes and avoid mismatching. Incorrect data association will lead to confusion of target trajectories and even produce false targets;

[0174] Construct the final cost matrix C′ ij Finally, the Hungarian algorithm is applied to find the optimal matching solution. The specific steps are as follows:

[0175] S3.31. Subtract the minimum value of its row from each element, and then subtract the minimum value of its column from each element, so that a new equivalent cost matrix can be generated;

[0176] S3.32. Try to cover all zeros with as few straight lines (rows or columns) as possible;

[0177] S3.33. If the number of lines is equal to the matrix order n, an optimal solution is found, otherwise, proceed to the next step;

[0178] S3.34. Find the minimum value of all elements not covered by the line, subtract the uncovered elements, and add to the elements at the two intersections;

[0179] Repeat the above steps S3.32 to S3.34 until the number of straight lines is equal to the order n of the matrix;

[0180] The Hungarian algorithm can help us achieve more accurate matching in multi-target tracking. By matching the observed data with known targets, we can ensure that each observation is correctly assigned to the corresponding target, thereby improving the tracking accuracy. This is very important to avoid false matches and missed matches, especially when there are similarities or interference between targets.

[0181] In step S3.3, the final cost matrix C′ ij The specific expression is:

[0182] C′ ij =C ij +a·p ij ;

[0183] in:

[0184] C ij = distance(z i , x i );

[0185]

[0186] Where a represents the prediction deviation p ij The weight factor of ij represents the basic cost; C′ ij represents the final cost matrix; p ij Represents the prediction deviation; distance represents the Euclidean distance, which uses the Euclidean distance metric to calculate the cost between the target and the observation; represents the predicted state of target j at the current time t; z i represents the observation value of target i, and represents the observation vector Z at time t t The i-th element of i represents the state vector of target i predicted by the model, and represents the state vector X at time t t The ith element of .

[0187] By introducing the deviation between the target's motion model prediction and the actual observation, and the similarity between the target and the observation to optimize the cost matrix, the target and the observation value can be matched more accurately, thereby improving the tracking accuracy, so that the system can not only handle target tracking in a static environment, but also cope with a dynamically changing environment, thereby enhancing the adaptability of the system;

[0188] Among them, the deviation between the target's motion model prediction and the actual observation is obtained through the Kalman filter;

[0189] In the application scenario of smart water cannon, the cost matrix C′ ij , which can ensure that the water cannon can quickly and accurately aim at the target, reduce unnecessary spraying times and waste of water resources, and thus achieve optimal allocation of resources;

[0190] Optimizing matching through the Hungarian algorithm can help the system allocate computing resources more efficiently. In multi-target tracking, multiple targets need to be processed simultaneously. The Hungarian algorithm can effectively manage these targets and ensure that each target is properly processed. Through reasonable matching, confusion between targets can be avoided, so that the trajectory of each target can be correctly tracked and recorded. In the application of smart water cannons, by accurately matching the targets, it can be ensured that the water cannons can quickly and accurately aim at the targets, reducing unnecessary spraying times and waste of water resources.

[0191] S3.4, use the state transfer equation to predict the target state at the next moment;

[0192] X t|t―1 =AX t―1 +Bu t―1 +W t―1 ;

[0193] Where, X t|t―1 represents the predicted value of the state vector at time t given time t-1; X t―1 represents the system state vector at time t-1; A represents the state transfer matrix, which is used to describe the change of system state over time; B represents the control input matrix, which is used to describe the influence of external control input on the state; u t―1 represents the control vector at time t-1; W t―1 represents process noise;

[0194] P t|t―1 =AP t―1 A T +Q;

[0195] Where P t|t―1 represents the predicted state covariance matrix, that is, the state covariance prediction matrix when the information at time t-1 is given at time t; Q represents the process noise covariance matrix; A Trepresents the transposed matrix of the state transfer matrix A; P t―1 represents the state covariance matrix at the previous time t-1;

[0196] S3.5, according to the weighted feature vector F of the current frame t and the weighted feature vector F of the previous frame t―1 The similarity calculation results between them are used to correct the predicted results for each target using the observation equation to obtain a new state estimate, which is used to update the target state information in real time.

[0197] Among them, the specific expression of the observation equation is:

[0198] Z t =HX t +V t ;

[0199] In the formula, Z t represents the observation vector at time t, which contains the data measured by the sensor; H represents the observation matrix, which maps the system state vector to the observation space; X t represents the system state vector at time t; V t Represents observation noise, which is used to reflect the uncertainty of sensor measurement;

[0200] Then calculate the Kalman gain K t , and update the state estimate:

[0201] K t =P t|t―1 H T (HP t|t―1 H T +R) ―1 ;

[0202] X′ t =X t|t―1 +K t (Z t ―HX t|t―1 );

[0203] Finally, update the state covariance matrix:

[0204] P′ t =(I-K t H)P t|t―1 ;

[0205] In the formula, H T represents the transpose of the observation matrix H; X′ t represents the updated state estimation vector; R represents the observation noise covariance matrix; P′ t represents the updated state covariance matrix; I represents the identity matrix, which is a matrix with all elements on the diagonal being 1 and the other elements being 0.

[0206] The whole process is a closed-loop system in which prediction and updating are continuously performed alternately, thus achieving continuous tracking of the target.

[0207] In this embodiment, a multi-target tracking control strategy is introduced to sort the tracking priorities of multiple target points. The specific steps involved are:

[0208] S3.6. Obtaining the initial thermal radiation intensity T of the target point through multi-source image data i (0), target point area A i (0), target point heat source combustion efficiency E i (0) and the temperature distribution D of the target point heat source i (0), and calculate the priority score according to the quantitative scoring algorithm;

[0209] In this embodiment, there are n target points. At the initial moment, each target point has a thermal radiation intensity T i (0), target point area A i (0), target point heat source combustion efficiency E i (0) and the temperature distribution D of the target point heat source i (0), since the state of the target point changes with time, the time variable t is introduced into the quantitative scoring algorithm, and the priority score is calculated. The priority score is used to determine the threat level of the target. The specific expression involved is:

[0210] P(T i (t),A i (t),E i (t),D i (t),t)=w T (t)·T i (t)+WA(t)·A i (t)+w E (t)·E i (t)+w D (t)·D i (t);

[0211] Where, T i (t) represents the thermal radiation intensity at time t; A i (t) represents the target point area at time t; E i (t) represents the combustion efficiency at time t; D i (t) represents the temperature distribution at time t; w T (t) represents the weight parameter of thermal radiation intensity at time t; w A (t) represents the weight parameter of the target point area at time t; w E(t) represents the weight parameter of combustion efficiency at time t; w D (t) represents the weight parameter of temperature distribution at time t; t represents time;

[0212] Temperature distribution of target point heat source D i (t) Through the analysis of flame images, it is found that in infrared images, the temperature distribution of different parts of the heat source can be extracted and a comprehensive temperature distribution score can be given accordingly;

[0213] in:

[0214] T i (t) = T i (0)+ΔT i t;

[0215] A i (t) = A i (0)+ΔA i t;

[0216] E i (t) = E i (0)+ΔE i t;

[0217] D i (t) = D i (0)+ΔD i t;

[0218] In the formula, ΔT i Indicates the change in thermal radiation intensity over time t; ΔA i Indicates the change in the target point area over time t; ΔE i Indicates the change in combustion efficiency over time t; ΔD i Represents the change of temperature distribution over time t;

[0219] And the weight parameters also change over time, as follows:

[0220] w T (t) = γ + δ·t;

[0221] w A (t) = ∈ + μ·t;

[0222] w E (t) = α + β·t;

[0223] w D (t) = η + θ·t;

[0224] Among them, γ, δ,∈, μ, α, β, η, and θ are constants with initial values ​​set according to expert opinions, which are used to describe how different factors change over time and the importance of these factors for priority assessment.

[0225] S3.7, use the quick sort algorithm to sort all target points from high to low priority scores, and sort multiple target points in descending order;

[0226] S3.8, select the target with the highest priority from the sorted list for tracking;

[0227] S3.9. When new temperature data arrives, repeat steps S3.6 to S3.8 to keep the priority sorting up to date.

[0228] Based on the above multi-target tracking control strategy, the tracking priority of multiple target points is sorted to improve the efficiency, accuracy and emergency response capability of the intelligent water cannon system. Through priority sorting, the system can give priority to the targets with the greatest threat, thereby more efficiently utilizing limited resources to deal with multiple targets.

[0229] In a multi-target environment, the status of targets will change over time. By dynamically adjusting priorities, the system can adapt to the changing environment more flexibly and ensure that the most urgent tasks are always focused on. In emergency situations such as fires, prioritizing the targets with the greatest threat can minimize losses and protect the safety of people and property.

[0230] By comprehensively considering factors such as thermal radiation intensity, target area, combustion efficiency and temperature distribution, the system can more comprehensively assess the threat level of the target and make decisions accordingly. When targeting non-fire source targets, there is no need to consider the impact of combustion efficiency.

[0231] And by using a quick sorting algorithm to sort all target points from high to low according to priority scores, and selecting the highest priority target for tracking, the system can respond quickly to new changes and improve its ability to deal with emergencies.

[0232] S4, sending the position information of the tracked target point to the water cannon actuator, which controls the steering and water jetting of the water cannon based on the PID control algorithm;

[0233] Among them, the influence of water jet velocity on the steering accuracy of the water cannon is introduced into the PID control algorithm to achieve the optimal control of the steering and water jet of the water cannon;

[0234] The water velocity will affect the steering accuracy of the water cannon, because the water flow will exert force on the mechanical structure of the water cannon, which may cause the water cannon to deviate from the target. Therefore, the influence coefficient of the water velocity k is v The influence coefficient k of the priority score on the water jet velocityp As a compensation item, it is used to improve control accuracy;

[0235] By adjusting the water flow rate according to the threat level of the target, water resources can be used more effectively and unnecessary injection can be reduced, thus saving resources and improving firefighting efficiency;

[0236] In this embodiment, the steering and water jet control of the water cannon is realized by a PID control algorithm, and the specific steps involved are:

[0237] S4.1. Initialize the parameter K of the PID control algorithm p , K i , K d , and the influence coefficient k of water jet velocity v The influence coefficient k of the priority score on the water jet velocity p As a compensation item;

[0238] S4.2. Based on the feature subset M of the target point, obtain the target position r i (t) and the current position of the water monitor y i (t), and calculate the error e i (t);

[0239] S4.3, using PID control algorithm to calculate PID control output signal;

[0240]

[0241] In the formula, u i (t) represents the PID control output signal of the i-th target at time t; K p K represents the proportional gain, which is used to determine the direct impact of the error e(t) on the control signal u(t); i Represents the integral gain, which is used to eliminate steady-state errors and accumulate past errors; K d represents the differential gain, which is used to reduce overshoot and improve the response speed of the system; e i (t) represents the error signal of the i-th target at time t; e i (τ) represents the error signal of the i-th target at a time point τ in the past, reflecting the difference between the actual position and the expected position of the target; represents the error e from time 0 to time t i The integral of (τ); Indicates the error e i (t) rate of change at time t; P i Indicates the priority score of the target;

[0242] Among them, e i (t) = r i (t)―yi (t); k v k P P i Indicates the water jet velocity v i Compensation item for the influence on the steering accuracy of water cannon;

[0243] S4.4, the calculated control signal u i (t) Applied to the water cannon actuator (servo motor) to adjust the angle of the water cannon;

[0244] S4.5. Score P according to the priority of the target i , adjust the water jet velocity v i ;

[0245] S4.6. Repeat steps S4.2 to S4.5 in the next sampling cycle.

[0246] Specifically, first calculate the target priority score P according to the quantitative scoring algorithm i , target priority score P i Reflects the threat level of target i, the target priority score P i The higher the value, the greater the threat level. According to the threat level, adjust the water jet speed v i ,The higher the threat level, the greater the water jet speed;

[0247] And the water jet velocity v i With priority score P i is proportional to, specifically expressed as:

[0248] v i =k P P i ;

[0249] In the formula, k P represents the influence coefficient of priority score on water jet velocity, and k P is a constant, and its specific value is set according to expert opinion;

[0250] Fully consider the influence of water flow velocity on the steering accuracy of water cannon and adjust the water flow velocity according to the priority score to determine the target threat level. Dynamically adjust the water flow velocity according to the threat level of the target, so as to more effectively extinguish the heat source. Introducing the influence of water flow velocity on the steering accuracy of water cannon into the PID control algorithm can make the intelligent water cannon system have higher accuracy, stronger adaptability and better resource utilization efficiency when facing complex environments.

[0251] Embodiment 2:

[0252] This embodiment provides a target tracking system applied to an intelligent water cannon, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement any of the steps of the target tracking method applied to an intelligent water cannon described above.

[0253] The above shows and describes the basic principles, main features and advantages of the present invention. Those skilled in the art should understand that the present invention is not limited by the above embodiments. The above embodiments and descriptions are only preferred examples of the present invention and are not intended to limit the present invention. Without departing from the spirit and scope of the present invention, the present invention may have various changes and improvements, which fall within the scope of the present invention. The scope of protection of the present invention is defined by the attached claims and their equivalents.

Claims

1. A target tracking method applied to an intelligent water cannon, characterized in that: The following steps are involved: S1. Use sensors to collect multi-source data of the target and target environment, pre-process the collected multi-source data, and use data fusion technology to fuse the multi-source image data; S2, using target extraction algorithm to extract target features from multi-source image data; S3. Establish a multi-target tracking model to track the target features extracted from multi-source image data in real time, and introduce a multi-target tracking control strategy to sort the tracking priorities for multiple target points; S4, sending the position information of the tracked target point to the water cannon actuator, which controls the steering and water jetting of the water cannon based on the PID control algorithm; Among them, the influence of water flow velocity on the steering accuracy of the water cannon is introduced into the PID control algorithm to achieve the optimal control of the steering and water flow injection of the water cannon.

2. The target tracking method applied to an intelligent water cannon according to claim 1, characterized in that: The sensor includes at least a visual sensor and an infrared sensor; The visual sensor is used to collect video streams containing visual features of the target and the target environment. The video streams are used to provide information about the contour of the surrounding environment, the contour of the target, the movement trajectory and speed of the target, the surface texture features of the target, and the distance between the target and the sensor. Infrared sensors are used to detect the infrared characteristics of the target environment, which specifically include heat source location information and thermal radiation intensity.

3. The target tracking method applied to an intelligent water cannon according to claim 2, characterized in that: In step S1, multi-source image data are fused using data fusion technology. The specific steps involved in data fusion are: S1.1, use ICP algorithm to align the preprocessed visual features and infrared features in the same spatial frame; S1.2, merging the visual features and infrared features aligned in the same spatial frame to construct a complete feature vector F; S1.3, using a recursive feature elimination algorithm to filter out a feature subset M in the feature vector F; S1.

4. Different weights are assigned to each feature according to its importance in the feature subset M, and a weighted feature vector F is obtained that combines visual features and infrared features. w ; S1.

5. Use the Dempster-Shafer evidence theory algorithm to integrate the information of visual and infrared sensors and calculate the values ​​of the trust function Bel and the likelihood function Pl.

4. The target tracking method applied to an intelligent water cannon according to claim 1, characterized in that: In step S2, the target extraction algorithm is used to extract target features from multi-source image data, and the specific steps involved are: Prepare a training data set, including target samples and non-target samples. The target samples contain the feature vector of the heat source, and the non-target samples do not contain the feature vector of the heat source. For each target sample, there is a positive class label y i = +1, each non-target sample has a negative class label y i = -1; Select the kernel function K and regularization parameter C; Known weighted eigenvector F w , and each feature vector has a corresponding label y, using the weighted feature vector F w And the corresponding target label y trains the SVM model, where the target label y represents the category to which the feature vector belongs; Use the trained SVM model to classify multi-source image data to detect and separate heat sources: In the formula, y(F w ) represents the predicted weighted feature vector F including the data points w The output label of ; sign(.) represents the sign function; N represents the total number of support vectors; a i represents the Lagrange multiplier; y i represents the label of the i-th support vector in the training data set; K(F w，i , F w ) represents the kernel function; i represents the index variable; F w，i represents the weighted feature vector of the i-th support vector in the training data set; F w represents the weighted feature vector of the new input data point, that is, the data point that needs to be classified; b represents the bias term.

5. The target tracking method applied to an intelligent water cannon according to claim 1, characterized in that: The multi-target tracking model is based on a Kalman filter and is used to track multiple target features extracted from multi-source image data in real time. The specific steps involved are: S3.

1. For each target detected for the first time, use the SVM classifier to confirm the target and initialize the basic information of the target's position, size, and shape characteristics; S3.

2. In subsequent frames, extract the corresponding weighted feature vector F for each target t , calculate the weighted feature vector F of the current frame t and the weighted feature vector F of the previous frame t―1 The similarity between two feature vectors is quantified by Euclidean distance. S3.3, construct a cost matrix, and include the deviation between the target's motion model prediction and the actual observation, and the similarity between the target and the observation into the cost matrix, and obtain the final cost matrix C′ ij , and apply the Hungarian algorithm to find matching solutions for multiple target features; S3.4, use the state transfer equation to predict the target state at the next moment; X t|t―1 =AX t―1 +Bu t―1 +W t―1 ; Where, X t|t―1 represents the predicted value of the state vector at time t given time t-1; X t―1 represents the system state vector at time t-1; A represents the state transfer matrix; B represents the control input matrix; u t―1 represents the control vector at time t-1; W t―1 represents process noise; P t|t―1 =AP t―1 From T +Q; Where P t|t―1 represents the predicted state covariance matrix; Q represents the process noise covariance matrix; A T represents the transposed matrix of the state transfer matrix A; P t―1 represents the state covariance matrix at the previous time t-1; S3.5, according to the weighted feature vector F of the current frame t and the weighted feature vector F of the previous frame t―1 The similarity calculation results between them are used to correct the predicted results for each target using the observation equation to obtain a new state estimate, which is used to update the target state information in real time. Among them, the specific expression of the observation equation is: Z t =HX t +V t ; In the formula, Z t represents the observation vector at time t; H represents the observation matrix; X t represents the system state vector at time t; V t represents the observation noise; Then calculate the Kalman gain K t , and update the state estimate: K t =P t|t―1 H T (HP t|t―1 H T +R) ―1 ; X′ t =X t|t―1 +K t (Z t ―HX t|t―1 ); Finally, update the state covariance matrix: P′ t =(I―K t H)P t|t―1 ; In the formula, H T represents the transpose of the observation matrix H; X′ t represents the updated state estimation vector; R represents the observation noise covariance matrix; P′ t represents the updated state covariance matrix; I represents the identity matrix.

6. The target tracking method applied to an intelligent water cannon according to claim 5, characterized in that: In S3.3, the final cost matrix C′ ij The specific expression is: C′ ij =C ij +a·p ij ; in: C ij =distance(z i ,x i ); Where a represents the prediction deviation p ij The weight factor of ij represents the basic cost; C′ ij represents the final cost matrix; p ij represents the prediction deviation; distance represents the Euclidean distance; represents the predicted state of target j at the current time t; z i represents the observed value of target i; x i represents the state vector of target i predicted by the model.

7. The target tracking method applied to an intelligent water cannon according to claim 3, characterized in that: In S3, by introducing a multi-target tracking control strategy, multiple target points are prioritized for tracking. The specific steps involved are: S3.

6. Obtaining the initial thermal radiation intensity T of the target point through multi-source image data i (0), target point area A i (0), target point heat source combustion efficiency E i (0) and the temperature distribution D of the target point heat source i (0), and calculate the priority score according to the quantitative scoring algorithm; S3.7, use the quick sort algorithm to sort all target points from high to low priority scores, and sort multiple target points in descending order; S3.8, select the target with the highest priority from the sorted list for tracking; S3.

9. When new temperature data arrives, repeat steps S3.6 to S3.8 to keep the priority sorting up to date.

8. The target tracking method applied to an intelligent water cannon according to claim 7, characterized in that: In S3.6, there are n target points. At the initial moment, each target point has a thermal radiation intensity T i (0), target point area A i (0), target point heat source combustion efficiency E i (0) and the temperature distribution D of the target point heat source i (0), since the state of the target point changes with time, the time variable t is introduced into the quantitative scoring algorithm, and the priority score is calculated. The specific expression involved is: P(T i (t),A i (t),E i (t),D i (t),t)=w T (t)·T i (t)+w A (t)·A i (t)+w E (t)·E i (t)+w D (t)·D i (t); Where, T i (t) represents the thermal radiation intensity at time t; A i (t) represents the target point area at time t; E i (t) represents the combustion efficiency at time t; D i (t) represents the temperature distribution at time t; w T (t) represents the weight parameter of thermal radiation intensity at time t; w A (t) represents the weight parameter of the target point area at time t; w E (t) represents the weight parameter of combustion efficiency at time t; w D (t) represents the weight parameter of temperature distribution at time t; t represents time.

9. The target tracking method applied to an intelligent water cannon according to claim 7, characterized in that: In S4, the steering and water jet control of the water cannon is realized through the PID control algorithm, and the specific steps involved are: S4.

1. Initialize the parameter K of the PID control algorithm p , K i , K d , and the influence coefficient k of water jet velocity v The influence coefficient k of the priority score on the water jet velocity p As a compensation item; S4.

2. Based on the feature subset M of the target point, obtain the target position r i (t) and the current position of the water monitor y i (t), and calculate the error e i (t); S4.3, using PID control algorithm to calculate PID control output signal; In the formula, u i (t) represents the PID control output signal of the i-th target at time t; K p Represents proportional gain; K i Indicates the integral gain; K d represents the differential gain; e i (t) represents the error signal of the i-th target at time t; e i (τ) represents the error signal of the i-th target at a time point τ in the past; represents the error e from time 0 to time t i The integral of (τ); Indicates the error e i (t) rate of change at time t; P i Indicates the priority score of the target; S4.4, the calculated control signal u i (t) Applied to water monitor actuators to adjust the angle of water monitors; S4.

5. Score P according to the priority of the target i , adjust the water jet velocity v i ; S4.

6. Repeat steps S4.2 to S4.5 in the next sampling cycle.

10. A target tracking system for an intelligent water cannon, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that: The processor executes a computer program to implement the steps of a target tracking method applied to an intelligent water cannon as described in any one of claims 1 to 9.

Images