A combined active and passive flexible bridge anti-collision system

Abstract

The present invention discloses a combined active and passive flexible bridge anti-collision system, which includes a ship trajectory prediction subsystem for collecting navigation image data of target ships at sea through monitoring anti-collision platforms and realizing the detection and tracking of navigation target ships based on a constructed cascaded neural network; an active warning subsystem for confirming the predicted channel number of the current ship according to the estimated ship motion state output by the target tracking network, obtaining the first collision probability between the target ship and the bridge pier corresponding to the predicted channel number; and determining the position number of the bridge pier with a collision risk based on the first collision probability and the collision probability threshold; and sending a warning signal containing the position number to the target ship through the monitoring anti-collision platform; an orbital movement positioning assistance subsystem for driving the monitoring anti-collision platform to a warning positioning position according to the position number, and obtaining the second collision probability between the target ship and the corresponding collision pier when the monitoring anti-collision platform is located at the warning positioning position through the ship trajectory prediction subsystem and the active warning subsystem, and sending a start anti-collision signal to the passive flexible anti-collision subsystem when the second collision probability exceeds the collision probability threshold. It solves the problems that the existing single anti-collision method is prone to errors leading to dangerous accidents and cannot reflect the generality of early warning.

Description

Technical Field

[0001] The present invention relates to the technical field of flexible bridge anti-collision based on artificial intelligence, and particularly relates to an active and passive combined flexible bridge anti-collision system. Background Art

[0002] Bridges play an important role in land transportation. The increasing number of annual navigations of passing ships increases the probability of ship-bridge collisions, seriously affecting the safety of bridges. Existing bridge-ship anti-collision devices are mainly divided into passive anti-collision devices and active anti-collision devices. Passive anti-collision devices use various flexible energy dissipation devices, which are installed on the outside of the bridge. When a collision occurs, the impact force generated is reduced to lower the loss. Active anti-collision devices install various sensors or cameras and other electronic devices around the bridge area to establish an anti-collision warning system. By using artificial intelligence technology to monitor passing ships, the navigation route is predicted in advance to guide the navigation of ships and reduce or avoid the occurrence of collision accidents. These solutions mainly use sensors to monitor the movement trajectory of ships and issue warnings when dangerous ships are found.

[0003] However, due to the existing bridge-ship anti-collision devices being greatly affected by the environment and unable to distinguish the type and size of ships, it leads to corresponding insurmountable difficulties in aspects such as the protection level and cost of ships and bridges in different situations. A single anti-collision method is also prone to errors resulting in dangerous accidents and cannot reflect the generality of early warning. Therefore, how to use the same set of anti-collision devices to effectively complete the anti-collision function for different bridge types and different situations is a current difficulty. Summary of the Invention

[0004] The present invention provides an active and passive combined flexible bridge anti-collision system to overcome the above technical problems.

[0005] To achieve the above object, the technical solution of the present invention is:

[0006] An active and passive combined flexible bridge anti-collision system, comprising

[0007] a ship trajectory prediction subsystem, an active warning subsystem, an orbital movement positioning assistance subsystem, and a passive flexible anti-collision subsystem;

[0008] The ship trajectory prediction subsystem is used to continuously collect navigation image data of target ships at sea through a monitoring anti-collision platform equipped with a CCD monitoring camera, and realize the detection and tracking of navigation target ships based on a constructed cascaded neural network;

[0009] And the cascaded neural network includes a pre-trained target detection and recognition network for obtaining ship categories and confidence levels and a target tracking network for estimating the motion state of ships according to the output of the target detection and recognition network;

[0010] The active warning subsystem is used to confirm the predicted waterway number of the current ship according to the estimated ship motion state output by the target tracking network, and obtain the first collision probability between the target ship and the bridge pier corresponding to the predicted waterway number;

[0011] And based on the first collision probability and the collision probability threshold, determine the position number of the bridge pier with a collision risk; and send a warning signal containing the position number to the target ship through the monitoring anti-collision platform;

[0012] The rail moving and positioning assistance subsystem is used to drive the monitoring anti-collision platform to the warning positioning position according to the position number, and obtain the second collision probability between the target ship and the corresponding collision pier when the monitoring anti-collision platform is at the warning positioning position through the ship trajectory prediction subsystem and the active warning subsystem, and send a start anti-collision signal to the passive flexible anti-collision subsystem when the second collision probability exceeds the collision probability threshold.

[0013] Furthermore, the rail moving and positioning assistance subsystem is communicatively connected to the active warning subsystem and the passive flexible anti-collision subsystem respectively;

[0014] It includes a laser positioning system, a control module, a linear rail structure, and a servo motor for driving the monitoring anti-collision platform to reciprocate along the linear rail structure;

[0015] And the linear rail structure is fixedly installed at the bottom end of the bridge flange;

[0016] The laser positioning system includes a laser emitter installed on the monitoring anti-collision platform and a laser receiver installed at the bottom end of the bridge flange and parallel to the central axis of the corresponding collision pier;

[0017] The control module is used to confirm the location of the collision pier corresponding to the number when the first collision probability exceeds the collision probability threshold according to the warning signal sent by the active warning subsystem, and generate a positioning control instruction for driving the servo motor to move;

[0018] The servo motor drives the monitoring anti-collision platform to move along the linear rail structure through the received positioning control instruction, and performs position feedback through the laser positioning system to move the monitoring anti-collision platform to the warning positioning position.

[0019] Furthermore, the passive flexible anti-collision subsystem includes an anti-collision mechanism device and an anti-collision system encapsulated inside the monitoring anti-collision platform;

[0020] And the anti-collision system includes a power supply module for providing electric energy and a flexible airbag control module;

[0021] The anti-collision mechanism device includes a gas generator connected by a gas pipe and a flexible compression airbag with a one-way inflation valve. The flexible compression airbag can fall onto the water surface by its own weight after inflation to achieve anti-collision protection for the bridge pier;

[0022] The flexible airbag control module is used to receive the start anti-collision signal sent by the active early warning subsystem to control the gas generator to inflate the flexible compression airbag;

[0023] One side of the monitoring anti-collision platform is provided with a frangible plate structure, which is arranged opposite to the flexible compression airbag and is used to release the flexible compression airbag when it is crushed by the flexible compression airbag.

[0024] Further, the implementation method for detecting and tracking the navigation target ship based on the constructed cascaded neural network specifically includes the following steps:

[0025] S01: Based on the pre-trained object detection and recognition network, according to the pre-processed navigation image data, obtain the ship category and confidence; and the object detection and recognition network is the feature extraction neural network Darknet and the object recognition module;

[0026] Pre-processing is to perform feature annotation of the bounding box and ship category on the ship detection target in the navigation image data, and divide the annotated navigation image data into S*S grid images to obtain an optimized navigation map;

[0027] The feature extraction neural network Darknet is used to extract the annotated features in the optimized navigation map to obtain the target feature map; the object recognition module is used to obtain the object recognition feature map according to the YOLO5 algorithm based on the target feature map, and the object recognition feature map includes the coordinates of the bounding box, the ship category, and the confidence;

[0028] S02: Estimate the ship motion state through the object tracking network according to the output of the object detection and recognition network; and the object tracking network includes an optical flow algorithm module and a Kalman filter module;

[0029] Based on the optical flow algorithm module, obtain the motion optical flow vector of the key pixel points in the two-dimensional pixel feature map according to the optical flow constraint equation, and obtain the heading speed and navigation direction of the current position of the target ship according to the motion optical flow vector of the key pixel points;

[0030] Wherein the two-dimensional pixel feature map is a feature map including water flow texture obtained after graying and denoising the object recognition feature map;

[0031] Through the Kalman filter module, based on the course speed and the sailing direction, the course trajectory of the target ship is predicted according to the observation data of the current position of the target ship, so as to estimate the ship motion state and further realize the detection and tracking of the sailing target ship.

[0032] Further, in S02, through the Kalman filter module, based on the course speed and the sailing direction, the course trajectory of the target ship is predicted according to the observation data of the current position of the target ship, so as to estimate the ship motion state, which specifically includes the following steps:

[0033] S021: According to the predicted output of the optical flow algorithm module, the state vector of the target ship at the current moment is expressed as X = [x, y, vx, vy]; where x represents the abscissa of the position of the target ship at the current moment; y represents the ordinate of the position of the target ship at the current moment; vx represents the lateral speed component of the sailing of the position of the target ship; vy represents the longitudinal speed component of the sailing of the position of the target ship.

[0034] S022: Define the state prediction equation and the position observation equation according to the state vector as

[0035] Xk+1|k = AXk|k + Bkuk

[0036] Zk = HXk + vk, Zk = [zx, zy]

[0037] In the formula: Xk+1|k represents the predicted state vector at the k+1 moment; Xk|k represents the estimated state vector at the k moment; A represents the state transition matrix determined by the motion optical flow vector in the optical flow algorithm module; Bk represents the control input matrix; uk represents the control input vector; Zk represents the position observation vector; H represents the observation matrix; vk represents the observation noise vector; zx represents the abscissa of the position observation; zy represents the ordinate of the position observation.

[0038] S023: Based on the state prediction equation and the position observation equation, obtain the target state estimation equation and the initial covariance matrix, and their expressions are

[0039] Xk+1|k+1 = Xk+1|k + Kk+1(Zk - HXk+1|k)

[0040] Pk+1|k = APk|kRk + Qk

[0041] In the formula: Kk+1 represents the Kalman gain used to determine the weights of the observation data and the predicted state in updating the state estimation; Pk|k represents the estimated covariance matrix at the k moment; Pk+1|k represents the predicted covariance matrix at the k+1 moment; Qk represents the process noise covariance matrix; Rk represents the observation noise covariance matrix.

[0042] And use the output of the target state estimation equation as the motion state estimation of the target ship under the current iteration;

[0043] S024: Obtain the updated covariance matrix according to the initial covariance matrix;

[0044] Pk+1|k+1 = (1 - Kk+1H)Pk+1|k

[0045] In the formula: Pk+1|k+1 represents the updated covariance matrix at time k + 1;

[0046] S025: Obtain the target state estimation at the next moment according to the covariance error matrix and the target state estimation equation;

[0047] S026: Use the target state estimation at the next moment and the updated covariance matrix as the input of the Kalman filter module again, and repeat steps S023 to S025.

[0048] Furthermore, in S02, the motion optical flow vector of the key pixel points in the two-dimensional pixel feature map is obtained based on the optical flow constraint equation, so as to obtain the heading speed and sailing direction of the current position of the target ship according to the motion optical flow vector of the key pixel points. The specific steps are as follows:

[0049] S001: Obtain the change relationship between the motion optical flow vector of the key pixel points in the two-dimensional pixel feature map and the image grayscale through the optical flow constraint equation, and the acquisition method of the optical flow constraint equation is:

[0050] Obtain the grayscale image expression of the two-dimensional pixel feature map as

[0051] I(x, y, t) = I(x + uΔt, y + vΔt, t + Δt)

[0052] In the formula: I(x, y, t) represents the image grayscale value at time t and position (x, y); u and v respectively represent the components of the motion optical flow vector in the x and y directions; Δt represents the time interval between two frames of images;

[0053] Perform Taylor expansion on the grayscale image expression and ignore the high-order terms to obtain the Taylor expansion formula as

[0054]

[0055] In the formula: represents the partial derivative of the grayscale image in the x direction; represents the partial derivative of the grayscale image in the y direction; represents the partial derivative of the grayscale image in time t;

[0056] According to the grayscale image expression and the Taylor expansion formula, obtain the optical flow constraint equation as

[0057]

[0058] S002: Set a sliding window to estimate the motion optical flow vector by minimizing the sum of the squares of the pixel intensity errors within the window, and the error function of the pixel intensity error is

[0059] E = Σ[(I(x + uΔt, y + vΔt, t + Δt) - I(x, y, t))^2]

[0060] where: Σ represents the summation over all pixel points within the sliding window; E represents the pixel intensity error;

[0061] Perform a Taylor expansion on the error function and set its first derivative to zero to obtain a system of linear equations for solving the motion optical flow vector, and its expression is

[0062] A * [u; v] = -b

[0063] where: A represents a 2n×2 matrix, and n is the number of pixel points within the window; b represents a 2n×1 vector; [u; v] represents the motion optical flow vector to be solved, and then confirm the heading speed and navigation direction of the current position of the target ship according to the solved motion optical flow vector.

[0064] Furthermore, the method for selecting key pixel points in the two-dimensional pixel feature map described in S001 is

[0065] S0010: Calculate the autocorrelation matrix of each pixel point in the two-dimensional pixel feature map, and its expression is

[0066]

[0067] where: C(x, y) represents the autocorrelation matrix of the pixel point (x, y); Ix and Iy respectively represent the partial derivatives of the image gray value in the x direction and y direction obtained through the Sobel operator or edge detection operator;

[0068] S0011: Construct a pixel point feature equation based on the autocorrelation matrix as

[0069] |C(x, y) - λI| = 0

[0070] where: I represents the identity matrix; λ represents the eigenvalue of the pixel point;

[0071] And based on the corner response function, obtain the corner response value according to the pixel point eigenvalue, and use the pixel point corresponding to the pixel point eigenvalue whose corner response value exceeds the preset response threshold as the key pixel point;

[0072] And the corner response function is R = min(λ1, λ2).

[0073] Further, the method for obtaining the first collision probability or the second collision probability is as follows

[0074] S100: Determine the predicted trajectory of the target ship based on the estimated ship motion state obtained by the ship trajectory prediction subsystem;

[0075] S200: And confirm whether a first collision prediction is satisfied according to the trajectory point coordinates in the predicted trajectory of the ship within the first set target area and the ship width of the predicted ship;

[0076] Where the rule for confirming whether the first collision prediction is satisfied is

[0077] Judge whether the current predicted collision distance of the ship meets the preset distance threshold;

[0078] If the current predicted collision distance of the ship meets the preset distance threshold, confirm the position number of the corresponding bridge pier where the ship may collide with the predicted waterway number, and calculate and obtain the first correct rate of the predicted ship trajectory;

[0079] The calculation formula for the first correct rate of the predicted ship trajectory is

[0080] P1 = 1 - ζ1

[0081] In the formula: P1 represents the first correct rate of the predicted ship trajectory, that is, the first collision probability; ζ1 represents the corresponding element in the error covariance matrix of the Kalman filter, that is, the error of the current predicted ship track;

[0082] Otherwise, re-detect other target ships sailing at sea through the ship trajectory prediction subsystem, and repeat S100;

[0083] S300: Judge whether the first correct rate of the predicted ship trajectory meets the preset correct rate threshold, that is, the collision probability threshold. If it meets, send a warning signal containing the position number to the target ship through the monitoring anti-collision platform; otherwise, repeat S100;

[0084] S400: Through the track movement positioning assistance subsystem, drive the monitoring anti-collision platform to the warning positioning position according to the position number, and obtain the current predicted trajectory of the current target ship confirmed according to the current estimated ship motion state through the ship trajectory prediction subsystem;

[0085] S500: And confirm whether a second collision prediction is satisfied according to the trajectory point coordinates in the predicted trajectory of the ship within the second set target area and the ship width of the predicted ship;

[0086] Where the rule for confirming whether the second collision prediction is satisfied is

[0087] Determine whether the predicted collision distance of the current ship meets the preset distance threshold;

[0088] If the predicted collision distance of the current ship meets the preset distance threshold, confirm the position number of the corresponding bridge pier where the ship may collide with the predicted waterway number, and calculate and obtain the second correct rate of the predicted ship trajectory;

[0089] The calculation formula for the second correct rate of the predicted ship trajectory is

[0090] P2 = 1 - ζ2

[0091] In the formula: P2 represents the second correct rate of the predicted ship trajectory, that is, the second collision probability; ζ2 represents the corresponding element in the error covariance matrix of the Kalman filter, that is, the error of the predicted ship trajectory of the current ship;

[0092] Otherwise, re-detect other target ships sailing at sea through the ship trajectory prediction subsystem, and repeat S100;

[0093] S600: Determine whether the second correct rate of the predicted ship trajectory meets the preset correct rate threshold, that is, the collision probability threshold. If it meets, send a start anti-collision signal to the passive flexible anti-collision subsystem through the monitoring anti-collision platform; otherwise, repeat S100.

[0094] Beneficial effects: The present invention provides a combined active and passive flexible bridge anti-collision system. By integrating an active warning subsystem, an orbital movement positioning assistance subsystem, and a passive flexible anti-collision subsystem, it can issue a warning prompt before a collision occurs, reducing the risk of collision. That is, through the active warning subsystem, based on the ship lane number, the predicted lane number of the current ship is confirmed according to the ship's movement state, and the collision piers corresponding to the numbers with the first collision probability exceeding the collision probability threshold are obtained. Then, a warning signal is sent to the target ship through the monitoring anti-collision platform, realizing active warning of the ship to facilitate the target ship to respond in a timely manner, so that it can be in a warning state when the anti-collision program is not triggered, greatly improving the navigability of the bridge. Through the orbital movement positioning assistance subsystem, the monitoring anti-collision platform is moved to the warning positioning position at the bottom of the bridge flange; and the ship trajectory prediction subsystem is called to obtain the navigation image data of the target ship again. When the second collision probability between the target ship and the corresponding collision pier exceeds the collision probability threshold, a start anti-collision signal is sent to the passive flexible anti-collision subsystem, and the damage of the target ship to the bridge pier in an inevitable situation is realized through the anti-collision mechanism device carried by the monitoring anti-collision platform. That is, in the case of an inevitable collision, the flexible anti-collision device is deployed, which can not only effectively protect the bridge structure but also maximize the protection of ship safety. Moreover, the combined active and passive method solves the problem that a single anti-collision method is prone to errors leading to dangerous accidents and cannot reflect the universality of warning. In addition, through the anti-collision mechanism device carried by the monitoring anti-collision platform, the disadvantages of high cost, affecting navigation, and insufficient protection of ships of traditional anti-collision devices are avoided, greatly improving the accuracy of collision avoidance prediction and warning between different ships and different bridge types. Brief Description of the Drawings

[0095] In order to more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the following will briefly introduce the drawings required for use in the description of the embodiments or the prior art. Obviously, the drawings in the following description are some embodiments of the present invention. For those of ordinary skill in the art, without creative efforts, other drawings can be obtained based on these drawings.

[0096] Figure 1 Schematic diagram of the combined active and passive flexible bridge anti-collision system of the present invention;

[0097] Figure 2 Flow schematic diagram of the combined active and passive flexible bridge anti-collision system in this embodiment;

[0098] Figure 3 Schematic diagram of the principle of the trajectory prediction and collision system in this embodiment;

[0099] Figure 4 Schematic diagram of the principle of the laser positioning system in this embodiment;

[0100] Figure 5 It is a schematic diagram of the principle of the track movement positioning assistance subsystem in this embodiment;

[0101] Figure 6 It is a schematic diagram of the principle of the flexible compression airbag excitation control in this embodiment;

[0102] Figure 7 It is a schematic diagram of the internal structure of the flexible anti-collision device in this embodiment;

[0103] Figure 8 It is a schematic diagram of the expansion of the compression airbag in this embodiment;

[0104] Figure 9 It is a simplified internal diagram of the flexible compression airbag in this embodiment. Specific implementation manners

[0105] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present invention. Apparently, the described embodiments are some, but not all, of the embodiments of the present invention. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts shall fall within the protection scope of the present invention.

[0106] This embodiment provides a main-passive combined flexible bridge anti-collision system, as Figure 1 shown, including a ship trajectory prediction subsystem, an active warning subsystem, a track movement positioning assistance subsystem, and a passive flexible anti-collision subsystem;

[0107] The ship trajectory prediction subsystem is used to continuously collect the navigation image data of the target ship at sea through a monitoring anti-collision platform equipped with a CCD monitoring camera, and realize the detection and tracking of the navigation target ship based on the constructed cascaded neural network; wherein the initial position of the monitoring anti-collision platform is set at the center position of the bottom end of the bridge flange where it is located, and image data within a set sea area is continuously collected and obtained at the initial position;

[0108] And the cascaded neural network includes a pre-trained target detection and recognition network for obtaining the ship category and confidence level and a target tracking network for estimating the ship motion state according to the output of the target detection and recognition network;

[0109] Specifically, as Figure 2 shown, the implementation method for realizing the detection and tracking of the navigation target ship based on the constructed cascaded neural network specifically includes the following steps:

[0110] S01: Based on the pre-trained object detection and recognition network, obtain the ship category and confidence according to the pre-processed navigation image data; and the object detection and recognition network is the feature extraction neural network Darknet and the object recognition module;

[0111] The pre-processing is to perform feature annotation of the bounding box and ship category on the ship detection target in the navigation image data, and segment the annotated navigation image data into S*S grid images to obtain an optimized navigation map;

[0112] The feature extraction neural network Darknet is used to extract the annotated features in the optimized navigation map to obtain the target feature map; the object recognition module is used to obtain the object recognition feature map according to the YOLO5 algorithm based on the target feature map, and the object recognition feature map includes the coordinates of the bounding box, the ship category, and the confidence; among them, calculating the confidence of each bounding box represents the probability that the bounding box contains an object and the probability of containing an object; the definition of confidence is Pr(Object)IoUpredtruth, where Pr(Object) represents the probability that the bounding box contains an object, and IoUpredtruth represents the intersection over union (Intersection Over Union) between the predicted bounding box and the true bounding box. The confidence calculation formula is: confidence = Pr(Object)×IoUpredtruth; if the true position of the object falls within a certain grid, then Pr(Object) takes the value of 1, otherwise 0; IoUpredtruth is the intersection over union between the predicted bounding box and the true bounding box. The calculation formula for the probability that the bounding box contains an object is:

[0113] Pr(Classi|Object)×Pr(Object)×IoUpredtruth = Pr(Classi)×IoUpredtruth; where Pr(Classi|Object) represents the probability that the grid cell belongs to a certain category under the condition of containing an object; Pr(Object) represents the probability that the grid cell contains an object;

[0114] In this embodiment, the pre-trained object detection and recognition network classifies ships using the neural network in the YOLO5 algorithm. The network is trained using the reverse ship training function, and the input vector and the corresponding target vector are used to train the neural network until the input vector can be classified in a defined appropriate manner. Through the pre-trained object detection and recognition network, ships can be classified into three categories: large, medium, and small, corresponding to less than 5,000 tons, 5,000 - 10,000 tons, and more than 10,000 tons respectively. In addition, the tonnage level (DWT) of the ship can also be predicted by inputting the ship size into the network; the recognized ship information is compared with the ship types in the database using the training method of the volov5 algorithm to obtain the ship type and size information; then, based on the target tracking network, the optical flow algorithm is used to calculate the actual position, actual navigation speed, and direction of the ship from the obtained ship information and pixel positions. The dataset of the previous voyages of various ships, including movement trajectories, speeds, etc., is saved for subsequent purposes of predicting ship trajectories, predicting channel numbers, and collision probabilities;

[0115] S02: According to the output of the object detection and recognition network through the target tracking network, estimate the ship motion state; and the target tracking network includes an optical flow algorithm module and a Kalman filter module;

[0116] Based on the optical flow algorithm module, obtain the motion optical flow vector of the key pixel points in the two-dimensional pixel feature map according to the optical flow constraint equation, so as to obtain the course speed and navigation direction of the target ship's current position according to the motion optical flow vector of the key pixel points;

[0117] Wherein the two-dimensional pixel feature map is a feature map including water flow texture obtained after grayscale and denoising processing of the target recognition feature map;

[0118] In a specific embodiment, in S02, obtaining the motion optical flow vector of the key pixel points in the two-dimensional pixel feature map according to the optical flow constraint equation, and obtaining the course speed and navigation direction of the target ship's current position according to the motion optical flow vector of the key pixel points specifically includes the following steps:

[0119] S001: Obtain the change relationship between the motion optical flow vector of the key pixel points in the two-dimensional pixel feature map and the image grayscale according to the optical flow constraint equation;

[0120] Specifically, the method for selecting the key pixel points in the two-dimensional pixel feature map in S001 is

[0121] S0010: Calculate the autocorrelation matrix of each pixel point in the two-dimensional pixel feature map, and its expression is

[0122]

[0123] Where: C(x, y) represents the autocorrelation matrix of the pixel point (x, y); Ix and Iy respectively represent the partial derivatives of the image gray value in the x-direction and y-direction obtained by the Sobel operator or edge detection operator;

[0124] S0011: Construct the pixel point feature equation according to the autocorrelation matrix as

[0125] |C(x, y) - λI| = 0

[0126] Where: I represents the identity matrix; λ represents the eigenvalue of the pixel point;

[0127] And based on the corner response function, obtain the corner response value according to the pixel point eigenvalue, and use the pixel point corresponding to the pixel point eigenvalue whose corner response value exceeds the preset response threshold as the key pixel point;

[0128] And the corner response function is R = min(λ1, λ2);

[0129] The acquisition method of the optical flow constraint equation is:

[0130] The obtained gray image expression of the two-dimensional pixel feature map is

[0131] I(x, y, t) = I(x + uΔt, y + vΔt, t + Δt)

[0132] Where: I(x, y, t) represents the image gray value at time t and position (x, y); u and v respectively represent the components of the motion optical flow vector in the x-direction and y-direction; Δt represents the time interval between two frames of images;

[0133] Perform Taylor expansion on the gray image expression and ignore the high-order terms to obtain the Taylor expansion formula as

[0134]

[0135] Where: Represents the partial derivative of the gray image in the x-direction; Represents the partial derivative of the gray image in the y-direction; Represents the partial derivative of the gray image in time t;

[0136] According to the gray image expression and the Taylor expansion formula, obtain the optical flow constraint equation as

[0137]

[0138] S002: Set a sliding window to estimate the motion optical flow vector by minimizing the sum of squares of the pixel intensity errors within the window, and the error function of the pixel intensity error is

[0139] E = Σ[(I(x + uΔt, y + vΔt, t + Δt) - I(x, y, t))^2]

[0140] Where: Σ represents the summation over all pixel points within the sliding window; E represents the pixel intensity error;

[0141] Perform a Taylor expansion on the error function and set its first derivative to zero to obtain a system of linear equations for solving the motion optical flow vector, and its expression is

[0142] A * [u; v] = -b

[0143] Where: A represents a 2n×2 matrix, and n is the number of pixel points within the window; b represents a 2n×1 vector; [u; v] represents the motion optical flow vector to be solved, and then confirm the course speed and sailing direction of the target ship's current position based on the solved motion optical flow vector;

[0144] Through the Kalman filter module, based on the course speed and sailing direction, predict the course trajectory of the target ship according to the observed data of the target ship's current position, so as to estimate the ship's motion state and then realize the detection and tracking of the sailing target ship.

[0145] In a specific embodiment, in S02, through the Kalman filter module, based on the course speed and sailing direction, predict the course trajectory of the target ship according to the observed data of the target ship's current position to estimate the ship's motion state, which specifically includes the following steps:

[0146] S021: According to the predicted output of the optical flow algorithm module, represent the state vector of the target ship at the current moment as X = [x, y, vx, vy]; where x represents the abscissa of the target ship's position at the current moment; y represents the ordinate of the target ship's position at the current moment; vx represents the transverse speed component of the target ship's position during sailing; vy represents the longitudinal speed component of the target ship's position during sailing;

[0147] S022: Define the state prediction equation and the position observation equation according to the state vector as

[0148] Xk+1|k = AXk|k + Bkuk

[0149] Zk = HXk + vk, Zk = [zx, zy]

[0150] Where: \(X_{k + 1|k}\) represents the predicted state vector at time \(k + 1\); \(X_{k|k}\) represents the estimated state vector at time \(k\); \(A\) represents the state transition matrix determined by the motion optical flow vector in the optical flow algorithm module; \(B_k\) represents the control input matrix; \(u_k\) represents the control input vector; \(Z_k\) represents the position observation vector; \(H\) represents the observation matrix; \(v_k\) represents the observation noise vector; \(z_x\) represents the abscissa of the position observation; \(z_y\) represents the ordinate of the position observation.

[0151] S023: Based on the state prediction equation and the position observation equation, obtain the target state estimation equation and the initial covariance matrix, and their expressions are

[0152] \(X_{k + 1|k+1}=X_{k + 1|k}+K_{k + 1}(Z_k - HX_{k + 1|k})\)

[0153] \(P_{k + 1|k}=AP_{k|k}R_k + Q_k\)

[0154] Where: \(K_{k + 1}\) represents the Kalman gain used to determine the weights of the observed data and the predicted state in updating the state estimation; \(P_{k|k}\) represents the estimated covariance matrix at time \(k\); \(P_{k + 1|k}\) represents the predicted covariance matrix at time \(k + 1\); \(Q_k\) represents the process noise covariance matrix; \(R_k\) represents the observation noise covariance matrix.

[0155] And take the output of the target state estimation equation as the motion state estimation of the target ship in the current iteration.

[0156] S024: Obtain the updated covariance matrix according to the initial covariance matrix;

[0157] \(P_{k + 1|k+1}=(1 - K_{k + 1}H)P_{k + 1|k}\)

[0158] Where: \(P_{k + 1|k+1}\) represents the updated covariance matrix at time \(k + 1\);

[0159] S025: According to the covariance error matrix and the target state estimation equation, obtain the target state estimation at the next moment;

[0160] S026: Take the target state estimation at the next moment and the updated covariance matrix as the input of the Kalman filter module again, and repeat steps S023 to S025.

[0161] The active warning subsystem is used to confirm the predicted waterway number of the current ship according to the estimated ship motion state output by the target tracking network, and obtain the first collision probability between the target ship and the bridge pier corresponding to the predicted waterway number; if the predicted trajectory coincides with the pier position, the following operation procedure is executed, that is, obtain the first collision probability between the ship and the corresponding bridge pier, otherwise return to the initial procedure, that is, return to the ship trajectory prediction subsystem to detect and track other ships sailing at sea.

[0162] Based on the first collision probability and the collision probability threshold, the position numbers of the bridge piers with collision risks are confirmed, that is, the position numbers of the collision piers corresponding to the first collision probability exceeding the collision probability threshold are obtained; and a warning signal is sent to the target ship through the monitoring anti-collision platform.

[0163] In this embodiment, the active anti-collision device uses real-time camera monitoring as the main data source for judging ship navigation information and early warning, and uses water flow information as an auxiliary means to achieve the ability of accurate prediction and early warning. In this embodiment, a binocular (multi-) camera is used as the image acquisition device to record the videos of the navigation of each passing ship, and the videos are input into the processing center in the system based on the ship trajectory prediction subsystem, that is, analyzed by means of artificial intelligence to obtain the predicted ship trajectory, and then the track movement positioning auxiliary subsystem is used to accurately move the monitoring anti-collision platform on the track. The principle of the trajectory prediction and collision system is as Figure 3 shown.

[0164] The track movement positioning auxiliary subsystem is used to drive the monitoring anti-collision platform to the warning positioning position according to the position number, that is, the track movement positioning auxiliary subsystem is used to move the monitoring anti-collision platform to the bottom end of the bridge flange and the warning positioning position parallel to the central axis of the corresponding collision pier according to the corresponding number of the collision pier;

[0165] Specifically, the track movement positioning auxiliary subsystem is respectively communicatively connected with the active warning subsystem and the passive flexible anti-collision subsystem; it includes a laser positioning system, a control module, a linear track structure, and a servo motor for driving the monitoring anti-collision platform to reciprocate along the linear track structure; and the linear track structure is fixedly installed at the bottom end of the bridge flange;

[0166] The laser positioning system includes a laser transmitter installed on the monitoring anti-collision platform and a laser receiver installed at the bottom end of the bridge flange and parallel to the central axis of the corresponding collision pier;

[0167] The control module is used to confirm the location of the collision pier corresponding to the number with the first collision probability exceeding the collision probability threshold according to the warning signal sent by the active warning subsystem, and generate a positioning control command for driving the servo motor to move; the servo motor drives the monitoring anti-collision platform to move along the linear track structure by receiving the positioning control command, and performs position feedback through the laser positioning system to drive the monitoring anti-collision platform to the warning positioning position;

[0168] In this embodiment, the track movement positioning assistance subsystem receives the command of the ship trajectory prediction subsystem, uses the servo motor as the power source, and accurately and quickly moves the monitoring and anti-collision platform to the preset position. In this embodiment, in order to eliminate the influence of accidental factors, a fine-tuning program for the servo motor is added, and the precise positioning of the monitoring and anti-collision platform is achieved in combination with the laser positioning system. The principle of the fine-tuning program is as follows: Laser receivers are arranged at the positioning positions corresponding to each waterway. When the laser emitted by the laser generator of the monitoring and anti-collision platform is received by the laser receiver, the laser receiver will transmit a signal indicating that the predetermined position has been reached to the platform. At this time, the servo motor stops working, and the programming method of the fine-tuning program is a well-known prior art means and will not be elaborated here; The principle of the track movement positioning system is as Figures 4 to 5 shown. Its working process is as follows: After receiving the movement instruction of the active warning subsystem, the track movement positioning assistance subsystem starts to work. The track movement positioning assistance subsystem processes the received predicted waterway number into the monitored - anti-collision platform number to be mobilized (each waterway signal corresponds to a corresponding platform number, and the signal conversion is achieved through the corresponding relationship) and the movement distance through the microprocessor, and transmits the instruction to the servo motor controller. Then, the servo motor operates according to the instruction, and at the same time, the laser positioning system starts to work. When the predetermined position is reached, the laser positioning system is used for inspection. If the laser receiver receives the laser signal, it means that the position of the monitoring and anti-collision platform is correct; If the laser receiver does not receive the signal, it means that there is a deviation in the position. At this time, the fine-tuning program is used to drive the servo motor for fine-tuning to reach the precise position;

[0169] And when the ship trajectory prediction subsystem and the active warning subsystem are used to obtain the second collision probability between the target ship and the corresponding collision pier when the monitoring and anti-collision platform is located at the warning positioning position, and when the second collision probability exceeds the collision probability threshold, a start anti-collision signal is sent to the passive flexible anti-collision subsystem, that is, the ship trajectory prediction subsystem is called to obtain the navigation image data of the target ship again, and the navigation target ship is detected and tracked again based on the cascaded neural network,

[0170] When the active warning subsystem is called to obtain the second collision probability between the target ship and the corresponding collision pier when the monitoring and anti-collision platform is located at the warning positioning position, and when the second collision probability exceeds the collision probability threshold, a start anti-collision signal is sent to the passive flexible anti-collision subsystem, and the damage of the target ship to the bridge pier in an inevitable situation is realized through the anti-collision mechanism device carried inside the monitoring and anti-collision platform;

[0171] Among them, if it is confirmed that the second collision probability does not exceed the collision probability threshold, a reset signal of the monitoring and anti-collision platform is sent to the track movement positioning assistance subsystem through the active warning subsystem. After resetting the command through the reset signal, the program controls the motor to move the monitoring and anti-collision platform to the initial position and wait for the next instruction; The principle of the track movement positioning assistance subsystem is asFigure 5 as shown

[0172] In a specific embodiment, the method for obtaining the first collision probability or the second collision probability is

[0173] S100: Determine the predicted trajectory of the target ship based on the estimated ship motion state obtained by the ship trajectory prediction subsystem;

[0174] S200: And confirm whether a first collision prediction is satisfied according to the trajectory point coordinates in the predicted ship trajectories within the first set target area and the ship width of the predicted ship;

[0175] Where the rule for confirming whether the first collision prediction is satisfied is

[0176] Judge whether the current predicted ship collision distance satisfies a preset distance threshold;

[0177] Wherein, the preset distance threshold is determined according to the ship width and the abscissa of the position of the bridge pier corresponding to the ship channel axis; the expression of the current predicted ship collision distance A is A = B + C, where B represents the ship width, that is, the maximum distance from the initial position point of the ship center to the ship edge; C represents the distance between the ship edge with the maximum distance from the ship center and the axis of the bridge pier;

[0178] If the current predicted ship collision distance satisfies the preset distance threshold, confirm the position number of the corresponding bridge pier where the ship may collide with the predicted channel number, and calculate and obtain the first correct rate of the ship predicted trajectory;

[0179] The calculation formula for the first correct rate of the ship predicted trajectory is

[0180] P1 = 1 - ζ1

[0181] In the formula: P1 represents the first correct rate of the ship predicted trajectory, that is, the first collision probability; ζ1 represents the corresponding element in the error covariance matrix of the Kalman filter, that is, the error of the current predicted ship trajectory;

[0182] Otherwise, re-detect other target ships sailing at sea through the ship trajectory prediction subsystem, and repeat S100;

[0183] S300: Judge whether the first correct rate of the ship predicted trajectory satisfies a preset correct rate threshold, that is, a collision probability threshold. If it is satisfied, send a warning signal containing the position number to the target ship through the monitoring and anti-collision platform; otherwise, repeat S100;

[0184] S400: Drive the monitoring and anti-collision platform to the warning and positioning position according to the position number through the track movement and positioning assistance subsystem, and obtain the predicted trajectory of the current target ship confirmed according to the current estimated ship motion state through the ship trajectory prediction subsystem;

[0185] S500: And confirm whether it meets the secondary collision prediction according to the trajectory point coordinates in the predicted trajectory of the ship in the second set target area and the ship width of the predicted ship;

[0186] Where the rule for confirming whether it meets the secondary collision prediction is

[0187] Judge whether the current predicted ship collision distance meets the preset distance threshold;

[0188] If the current predicted ship collision distance meets the preset distance threshold, confirm the position number of the corresponding bridge pier where the ship may collide with the predicted waterway number, and calculate and obtain the second accuracy rate of the predicted ship trajectory;

[0189] The calculation formula for the second accuracy rate of the predicted ship trajectory is

[0190] P2 = 1 - ζ2

[0191] In the formula: P2 represents the second accuracy rate of the predicted ship trajectory, that is, the second collision probability; ζ2 represents the corresponding element in the error covariance matrix of the Kalman filter, that is, the error of the current predicted ship track;

[0192] Otherwise, re-detect other target ships sailing at sea through the ship trajectory prediction subsystem, and repeat S100;

[0193] S600: Judge whether the second accuracy rate of the predicted ship trajectory meets the preset accuracy rate threshold, that is, the collision probability threshold. If it meets, send a start anti-collision signal to the passive flexible anti-collision subsystem through the monitoring and anti-collision platform; otherwise, repeat S100.

[0194] Specifically, the passive flexible anti-collision subsystem includes an anti-collision mechanism device and an anti-collision system encapsulated inside the monitoring and anti-collision platform;

[0195] And the anti-collision system includes a power supply module for providing electrical energy and a flexible airbag control module;

[0196] The anti-collision mechanism device includes a gas generator connected by a gas pipe and a flexible compression airbag provided with a one-way inflation valve;

[0197] The flexible airbag control module is used to receive the start anti-collision signal sent by the active early warning subsystem to control the gas generator to inflate the flexible compression airbag;

[0198] One side of the monitoring and anti-collision platform is provided with a frangible plate structure, which is arranged opposite to the flexible compression airbag. When the volume of the flexible compression airbag increases during the inflation process, it squeezes and destroys the frangible plate structure to unfold and surround the corresponding collision pier. After the flexible compression airbag is inflated, it falls onto the water surface under its own weight to achieve anti-collision protection for the pier.

[0199] In this embodiment, the passive flexible anti-collision subsystem uses a flexible airbag control module (i.e., the reference automotive airbag control system ACU) as the control center, communicates with the active warning subsystem, and exchanges information through connection. When the anti-collision program is started, the ignition device and sodium azide inflator are used to control the activation of the compression airbag through the gas generator. The flexible compression airbag and the ignition device are located at the front end of the flexible anti-collision device, as Figure 6 shown. The activation process is as follows: the control system powers on the power supply, the ignition coil is powered on to generate high temperature, causing the primer to catch fire. Immediately afterwards, the inflator burns and generates nitrogen gas, which is sent into the flexible compression airbag to make the flexible compression airbag unfold. The control principle of the flexible airbag activation is as Figure 3 shown. Among them, the overall structure of the passive flexible anti-collision subsystem is rigidly connected in parallel with the ship trajectory prediction subsystem and the active warning subsystem to ensure that the displacement modes of the two subsystems are consistent. The internal structure of the passive flexible anti-collision subsystem mainly consists of a flexible compression airbag, a control circuit, an igniter, a power supply system, a backup power supply, and a load block, etc. The flexible airbag control module includes a control circuit, a wireless signal transmission module, and a microprocessor, which can receive instructions and start the gas generator. A large amount of nitrogen gas generated by the gas generator in a short time is input into the compression airbag through the air duct to achieve rapid inflation. One side of the monitoring and anti-collision platform is provided with a frangible plate structure, which is arranged opposite to the flexible compression airbag. When it is crushed by the flexible compression airbag, it releases the flexible compression airbag, quickly unfolds and wraps the pier structure, and the flexible compression airbag can fall onto the water surface under its own weight after inflation to achieve anti-collision protection for the pier. The schematic diagram of the flexible airbag unfolding is as Figures 8 to 9 shown; the internal structure of the flexible anti-collision system is as Figure 7 shown.

[0200] In this embodiment, by integrating an active warning subsystem, an orbit movement positioning assistance subsystem, and a passive flexible anti-collision subsystem, it is possible to issue a warning prompt before a collision occurs, reducing the risk of collision. In the case where a collision is unavoidable, a flexible anti-collision device is deployed, which can not only effectively protect the bridge structure but also maximize the protection of ship safety. Its flexible anti-collision airbag has the characteristics of good energy absorption effect, long buffering time, small peak impact force, reusable, and low cost compared with traditional rigid anti-collision devices. The addition of an active warning system enables it to be in a folded state when the anti-collision program is not triggered, greatly improving the navigability of the bridge and making it quite feasible. Its subsystem, the ship trajectory prediction and collision warning system, has information processing capabilities and can independently process and make decisions with the support of artificial intelligence. At the same time, it has a communication module inside, which can be upgraded and controlled at any time, increasing the controllability and transparency of the system. The equipment used in this system is easy to obtain, making large-scale promotion possible. According to simple statistics, the cost of a single-bridge collision warning system is about 700,000 yuan, while the cost of a rigid anti-collision device using composite materials as the energy-absorbing material reaches more than one million yuan. This system uses high-speed cameras and flexible anti-collision devices, greatly reducing the cost. The cost of a single bridge is approximately 130,000 yuan for (2500 + 1500 + 3500 + 5000 + 800) * 10, significantly increasing the possibility of wide installation. This embodiment integrates multiple emerging technologies such as active warning, mobile platforms, and artificial intelligence, avoiding the disadvantages of traditional anti-collision devices such as high price, affecting navigation, and insufficient protection for ships, promoting the wide use of bridge anti-collision devices, and providing a new research direction for the future research of bridge anti-collision devices.

[0201] Finally, it should be noted that: the above embodiments are only used to illustrate the technical solutions of the present invention, and are not intended to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those of ordinary skill in the art should understand that they can still modify the technical solutions described in the foregoing embodiments, or perform equivalent replacements for some or all of the technical features; and these modifications or replacements do not make the essence of the corresponding technical solutions deviate from the scope of the technical solutions of the embodiments of the present invention.

Claims

1. An active and passive combined flexible bridge anti-collision system, characterized in that: include Ship trajectory prediction subsystem, active warning subsystem, orbital mobile positioning auxiliary subsystem and passive flexible collision avoidance subsystem; The ship trajectory prediction subsystem is used to continuously collect navigation image data of target ships at sea by monitoring the anti-collision platform, and realize detection and tracking of navigation target ships based on the constructed cascade neural network; The cascade neural network includes a pre-trained target detection and recognition network for obtaining the ship category and confidence and a target tracking network for estimating the ship motion state according to the output of the target detection and recognition network; The active warning subsystem is used to confirm the predicted channel number of the current ship according to the estimated ship motion state output by the target tracking network, and obtain the first collision probability between the target ship and the bridge pier corresponding to the predicted channel number; Based on the first collision probability and the collision probability threshold, the position number of the bridge pier with collision risk is determined; and a warning signal containing the position number is sent to the target ship through the monitoring anti-collision platform; The track movement positioning auxiliary subsystem is used to drive the monitoring and anti-collision platform to the warning positioning position according to the position number, and obtain the second collision probability between the target ship and the corresponding collision pier when the monitoring and anti-collision platform is located at the warning positioning position through the ship trajectory prediction subsystem and the active early warning subsystem, and send a start-up anti-collision signal to the passive flexible anti-collision subsystem when the second collision probability exceeds the collision probability threshold; The passive flexible anti-collision subsystem includes an anti-collision mechanism device and an anti-collision system encapsulated inside the monitoring anti-collision platform; And the anti-collision system includes a power module for providing electrical energy and a flexible airbag control module; The anti-collision mechanism device includes a gas generator connected by an air guide tube and a flexible compressed air bag provided with a one-way inflation valve. After inflation, the flexible compressed air bag can fall to the water surface under its own weight to achieve anti-collision protection for the bridge pier. The flexible airbag control module is used to receive the start-up anti-collision signal sent by the active warning subsystem to control the gas generator to inflate the flexible compression airbag; A fragile plate structure is provided on one side of the monitoring and anti-collision platform and is arranged opposite to the flexible compressed airbag, so as to release the flexible compressed airbag when the flexible compressed airbag is squeezed and broken.

2. The active-passive combined flexible bridge anti-collision system according to claim 1 is characterized in that: The track movement positioning auxiliary subsystem is respectively connected to the active early warning subsystem and the passive flexible collision avoidance subsystem in communication; It includes a laser positioning system, a control module, a linear track structure, and a servo motor for driving the monitoring and anti-collision platform to reciprocate along the linear track structure; And the linear track structure is fixedly installed at the bottom end of the bridge flange; The laser positioning system includes a laser transmitter installed on the monitoring and anti-collision platform, and a laser receiver installed at the bottom end of the bridge flange and parallel to the central axis of the corresponding collision pier; The control module is used to confirm the location of the collision pier with a number corresponding to the collision probability threshold value according to the warning signal sent by the active warning subsystem, and generate a positioning control instruction for driving the servo motor to move; The servo motor drives the monitoring and anti-collision platform to move along the linear track structure through the received positioning control instruction, and performs position feedback through the laser positioning system to move the driven monitoring and anti-collision platform to the warning positioning position.

3. The active-passive combined flexible bridge anti-collision system according to claim 2 is characterized in that: The method for realizing detection and tracking of a navigation target ship based on the constructed cascade neural network specifically comprises the following steps: S01: Based on the pre-trained target detection and recognition network, the ship category and confidence are obtained according to the pre-processed navigation image data; and the target detection and recognition network is the feature extraction neural network Darknet and the target recognition module; The preprocessing is to annotate the ship detection targets in the navigation image data with bounding boxes and ship category features, and segment the annotated navigation image data into S*S grid images to obtain an optimized navigation map; The feature extraction neural network Darknet is used to extract the annotation features in the optimized navigation map to obtain the target feature map; the target recognition module is used to obtain the target recognition feature map based on the target feature map through the YOLO5 algorithm, and the target recognition feature map includes the coordinates of the bounding box, the ship category and the confidence level; S02: Estimate the motion state of the ship according to the output of the target detection and recognition network through the target tracking network; and the target tracking network includes an optical flow algorithm module and a Kalman filter module; Based on the optical flow algorithm module, the motion optical flow vector of the key pixel point in the two-dimensional pixel feature map is obtained according to the optical flow constraint equation, so as to obtain the heading speed and navigation direction of the current position of the target ship according to the motion optical flow vector of the key pixel point; The two-dimensional pixel feature map is a feature map including water flow texture obtained by graying and denoising the target recognition feature map; Through the Kalman filter module, based on the heading speed and sailing direction, the heading trajectory of the target ship is predicted according to the observation data of the current position of the target ship, so as to estimate the ship's motion state and thus realize the detection and tracking of the sailing target ship.

4. The active-passive combined flexible bridge anti-collision system according to claim 3 is characterized in that: In S02, the Kalman filter module is used to predict the heading trajectory of the target ship based on the heading speed and the sailing direction according to the observation data of the current position of the target ship to estimate the motion state of the ship, which specifically includes the following steps: S021: According to the prediction output of the optical flow algorithm module, the state vector of the target ship at the current moment is represented as X=[x, y, vx, vy]; wherein x represents the abscissa of the target ship position at the current moment; y represents the ordinate of the target ship position at the current moment; vx represents the lateral velocity component of the target ship position; vy represents the longitudinal velocity component of the target ship position; S022: According to the state vector, the state prediction equation and the position observation equation are defined as follows: Xk+1|k=AXk|k+Bkuk Zk=HXk+vk,Zk=[zx,zy] Where: Xk+1|k represents the predicted state vector at time k+1; Xk|k represents the estimated state vector at time k; A represents the state transfer matrix determined by the motion optical flow vector in the optical flow algorithm module; Bk represents the control input matrix; uk represents the control input vector; Zk represents the position observation vector; H represents the observation matrix; vk represents the observation noise vector; zx represents the horizontal coordinate of the position observation; zy represents the vertical coordinate of the position observation; S023: Based on the state prediction equation and the position observation equation, the target state estimation equation and the initial covariance matrix are obtained, and the expression is: Xk+1|k+1=Xk+1|k+Kk+1(Zk-HXk+1|k) Pk+1|k=APk|kRk+Qk Where: Kk+1 represents the Kalman gain used to determine the weights of the observed data and the predicted state in the updated state estimation; Pk|k represents the estimated covariance matrix at time k; Pk+1|k represents the predicted covariance matrix at time k+1; Qk represents the process noise covariance matrix; Rk represents the observation noise covariance matrix; The output of the target state estimation equation is used as the motion state estimation of the target ship in the current iteration; S024: Obtain an updated covariance matrix according to the initial covariance matrix; Pk+1|k+1=(1-Kk+1H)Pk+1|k Where: Pk+1|k+1 represents the updated covariance matrix at time k+1; S025: Obtain the target state estimation at the next moment according to the covariance error matrix and the target state estimation equation; S026: Use the target state estimation and updated covariance matrix at the next moment as input to the Kalman filter module again, and repeat steps S023 to S025.

5. The active-passive combined flexible bridge anti-collision system according to claim 3 is characterized in that: In S02, the motion optical flow vector of the key pixel point in the two-dimensional pixel feature map is obtained based on the optical flow constraint equation, so as to obtain the heading speed and navigation direction of the current position of the target ship according to the motion optical flow vector of the key pixel point, which specifically includes the following steps: S001: Obtain the change relationship between the motion optical flow vector of the key pixel point in the two-dimensional pixel feature map and the image grayscale through the optical flow constraint equation, and the method for obtaining the optical flow constraint equation is: The grayscale image expression for obtaining the two-dimensional pixel feature map is: I(x,y,t)=I(x+uΔt,y+vΔt,t+Δt) Where: I(x,y,t) represents the grayscale value of the image at time t and position (x,y); u and v represent the components of the motion optical flow vector in the x and y directions respectively; Δt represents the time interval between two frames of images; Perform Taylor expansion on the grayscale image expression and ignore the higher-order terms to obtain the Taylor expansion formula: Where: Represents the partial derivative of the grayscale image in the x direction; Represents the partial derivative of the grayscale image in the y direction; Represents the partial derivative of the grayscale image at time t; According to the grayscale image expression and Taylor expansion formula, the optical flow constraint equation is obtained as follows: S002: Set a sliding window to estimate the motion optical flow vector by minimizing the sum of squares of pixel intensity errors within the window, and the error function of the pixel intensity error is E=Σ[(I(x+uΔt,y+vΔt,t+Δt)-I(x,y,t))^2] Where: Σ represents the sum of all pixels in the sliding window; E represents the pixel intensity error; Taylor expansion is performed on the error function and its first-order derivative is set to zero to obtain a linear equation system for solving the motion optical flow vector, which is expressed as follows: A*[u;v]=-b Wherein: A represents the state transfer matrix determined by the motion optical flow vector in the optical flow algorithm module; b represents a 2n×1 vector; [u; v] represents the motion optical flow vector to be solved, and then the heading speed and navigation direction of the current position of the target ship are confirmed according to the solved motion optical flow vector.

6. The active-passive combined flexible bridge anti-collision system according to claim 5 is characterized in that: The method for selecting key pixels in the two-dimensional pixel feature map in S001 is: S0010: Calculate the autocorrelation matrix of each pixel in the two-dimensional pixel feature map, and its expression is Where: C(x,y) represents the autocorrelation matrix of the pixel point (x,y); Ix and Iy represent the partial derivatives of the image grayscale value in the x direction and y direction obtained by the Sobel operator or edge detection operator respectively; S0011: Construct pixel feature equation based on autocorrelation matrix as |C(x,y)-λI|=0 Where: I represents the unit matrix; λ represents the eigenvalue of the pixel; Based on the corner point response function, the corner point response value is obtained according to the pixel point feature value, and the pixel point corresponding to the pixel point feature value whose corner point response value exceeds the preset response threshold is taken as the key pixel point; And the corner point response function is R=min(λ1,λ2).

7. The active-passive combined flexible bridge anti-collision system according to claim 1 is characterized in that: The method for obtaining the first collision probability or the second collision probability is as follows: S100: determining a predicted trajectory of a target ship through an estimated ship motion state obtained by a ship trajectory prediction subsystem; S200: confirming whether a collision prediction is satisfied based on the coordinates of the track points in the predicted track of the ship in the first set target area and the predicted ship width; The rule for confirming whether a collision prediction is satisfied is: Determine whether the current predicted collision distance of the ship meets the preset distance threshold; If the current predicted collision distance of the ship meets the preset distance threshold, the position number of the corresponding bridge pier where the ship may collide with the predicted channel number is confirmed, and the first accuracy of obtaining the predicted trajectory of the ship is calculated; The calculation formula for the first accuracy of the ship's predicted trajectory is: P1=1-ζ1 Where: P1 represents the first accuracy of the ship's predicted trajectory, that is, the first collision probability; ζ1 represents the corresponding element in the error covariance matrix of the Kalman filter, that is, the error of the current ship's predicted track; Otherwise, the ship trajectory prediction subsystem re-detects other target ships sailing at sea and repeats S100; S300: determining whether the first accuracy of the predicted trajectory of the ship meets a preset accuracy threshold, i.e., a collision probability threshold; if so, sending a warning signal containing the position number to the target ship through the monitoring anti-collision platform; Otherwise, repeat S100; S400: driving the monitoring and anti-collision platform to the warning positioning position according to the position number through the track movement positioning auxiliary subsystem, and obtaining the current target ship prediction trajectory confirmed according to the current estimated ship motion state through the ship trajectory prediction subsystem; S500: confirming whether the secondary collision prediction is satisfied according to the coordinates of the track points in the predicted track of the ship in the second set target area and the predicted ship width; The rule for confirming whether the secondary collision prediction is satisfied is: Determine whether the current predicted collision distance of the ship meets the preset distance threshold; If the current predicted collision distance of the ship meets the preset distance threshold, the position number of the corresponding bridge pier where the ship may collide with the predicted channel number is confirmed, and the second accuracy of obtaining the predicted trajectory of the ship is calculated; The calculation formula for the second accuracy of the ship's predicted trajectory is: P2=1-ζ2 Where: P2 represents the second accuracy of the ship's predicted trajectory, that is, the second collision probability; ζ2 represents the corresponding element in the error covariance matrix of the Kalman filter, that is, the error of the current ship's predicted track; Otherwise, the ship trajectory prediction subsystem re-detects other target ships sailing at sea and repeats S100; S600: Determine whether the second accuracy of the ship's predicted trajectory meets a preset accuracy threshold, i.e., a collision probability threshold. If so, send a start-up collision avoidance signal to the passive flexible collision avoidance subsystem through the collision avoidance monitoring platform; otherwise, repeat S100.

Images