Video linkage camera intelligent scheduling control method and system

Abstract

The invention relates to the technical field of intelligent video monitoring and multi-sensor cooperative control, in particular to a video linkage camera intelligent scheduling control method and system, and the method comprises the steps: presetting a semantic entropy trigger threshold value, a basic category weight and a behavior weight coefficient; maintaining the current monitoring state; triggering an emergency game scheduling process, and obtaining basic features of a moving target in a view field based on a panoramic video stream collected by a wide-angle monitoring array; combining the basic features with a preset knowledge base, and resolving to obtain target information value potential energy; according to the current attitude of the dome camera, the target position mapping angle and the edge calculation node load data, determining the mechanical-computing power combined cost; generating an aging risk ratio based on the target information value potential energy, the mechanical-computing power combined cost and the target escape time; outputting a physical holder control signal; outputting a virtual viewpoint synthesis instruction; according to the method, the problem that the system is slow to perceive nonlinear group emergencies is solved, and on-demand allocation and efficient utilization of resources are realized.

Description

Technical Field

[0001] This invention relates to the field of intelligent video surveillance and multi-sensor collaborative control technology, specifically to a method and system for intelligent scheduling and control of video-linked cameras. Background Technology

[0002] In the current wide-area security monitoring environment, multi-camera linkage systems face complex and ever-changing scene challenges, especially when there are non-linear group emergencies or high-density crowd gatherings, the targets in the field of view exhibit characteristics of high concurrency, strong dynamism and explosive growth of information entropy.

[0003] To acquire detailed features of moving targets in the field of view, existing camera scheduling solutions generally employ preset cruise paths or single rule-triggered mechanisms. This involves driving the PTZ camera to perform physical rotation, focusing, and zoom operations according to a fixed scanning logic or a simple first-in-first-out queue. While these solutions possess some monitoring capabilities in low-entropy steady-state scenarios, the inherent physical limits of the mechanical gimbal's rotation speed and zoom range, coupled with the lack of quantitative assessment of target value potential energy and escape risk in traditional scheduling logic, lead to a severe lag in mechanical response compared to target motion changes when the information burst rate exceeds conventional processing capabilities. This contradiction between the limited physical resources and the dynamic nature of targets results in untimely capture of high-value targets, a high rate of loss of critical evidence, and scheduling failures when faced with high-throughput information surges. Therefore, establishing a system capable of sensitively sensing the rate of information temperature rise in the scene and balancing physical and mechanical costs with computational synthesis costs through an adaptive game theory mechanism to maximize target acquisition efficiency with limited resources has become a pressing technical problem. Summary of the Invention

[0004] The purpose of this invention is to provide a method and system for intelligent scheduling and control of video-linked cameras, solving the problem of failure of traditional scheduling mechanisms in high-concurrency nonlinear group emergencies, and effectively resolving the contradiction between the limited physical resources and the dynamic nature of the target. Specifically, the technical solution of this invention is as follows:

[0005] A method for intelligent scheduling and control of video-linked cameras includes:

[0006] The system presets semantic entropy trigger thresholds, basic category weights, and behavior weight coefficients; initializes and generates a global field-of-view coordinate system based on the wide-angle monitoring coverage area; calculates scene semantic entropy flux in real time; maintains the current monitoring state when the scene semantic entropy flux is less than or equal to the semantic entropy trigger threshold; and triggers an emergency game-theoretic scheduling process when the scene semantic entropy flux is greater than the semantic entropy trigger threshold, including:

[0007] Based on the panoramic video stream acquired by the wide-angle monitoring array, the basic features of moving targets in the field of view are obtained through lightweight target detection algorithm. Combining the basic features with the preset knowledge base, the target information value potential energy is calculated through physical gravitational potential energy model analysis. The current attitude of the sphere, the target position mapping angle and the load data of the edge computing node are processed through the mechanical-computing power joint cost function to determine the mechanical-computing power joint cost.

[0008] Based on the target information value potential energy, the combined cost of mechanical and computing power, and the target escape time, a timeliness risk ratio is generated;

[0009] When the time-sensitive risk ratio is less than the preset safety ratio, a physical PTZ control signal is output.

[0010] When the timeliness risk ratio is greater than or equal to the preset safety ratio, a virtual viewpoint synthesis command is output.

[0011] Preferably, the real-time calculation of scene semantic entropy flux includes: calculating the displacement of adjacent frames using optical flow based on the panoramic video stream to obtain the target velocity vector;

[0012] Based on the panoramic video stream, a normalized target spatial density function is generated using the intersection-union heatmap of the target detection boxes.

[0013] Based on the normalized target space density function and the target velocity vector, the global kinetic energy integral term characterizing the energy of group motion is calculated;

[0014] Based on the target velocity vector, calculate the global divergence integral term that characterizes the discrete deformation rate of the population;

[0015] The global kinetic energy integral term and the global divergence integral term are multiplied by preset kinetic energy conversion coefficients and structural entropy conversion coefficients, respectively, and then summed to generate the scene semantic entropy flux.

[0016] Preferably, the target information value potential energy is calculated by: obtaining the basic category weights of the target from a preset knowledge base; generating a behavioral semantic score of the target using a behavioral analysis algorithm; and calculating the target's dwell time in the field of view.

[0017] The completeness of acquired information is obtained based on the quality assessment of historical frame images; a potential energy field equation is constructed with the basic category weight and behavioral semantic score as gain terms, and the target dwell time and the completeness of acquired information as attenuation terms, and the potential energy of target information value is calculated by solving the equation.

[0018] Preferably, the determination of the combined mechanical and computing costs includes: calculating the angle difference between the target mapping angle and the current attitude of the PTZ camera, and determining the mechanical time delay by combining the preset maximum angular velocity of the gimbal; and obtaining the zoom motor travel time as the zoom time delay.

[0019] Monitor the real-time load of edge computing nodes to obtain the average inference latency as the computing power lag; superimpose the mechanical lag, zoom lag and computing power lag to determine the joint mechanical-computing power cost.

[0020] Preferably, the generation of the time-risk ratio includes: determining the effective line-of-sight distance based on the straight-line distance from the target's current position along the target velocity vector direction to the boundary of the effective coverage area; and calculating the target's effective escape time based on the ratio of the effective line-of-sight distance to the magnitude of the target velocity vector.

[0021] Divide the combined cost of machine and computing power by the target's effective escape time to generate the timeliness risk ratio.

[0022] Preferably, the virtual viewpoint synthesis instruction is output, including: determining the set of neighboring cameras that overlap with the current main PTZ camera's field of view; calculating the cross-union ratio (CUP) between the neighboring camera's field of view and the target's predicted position; obtaining the effective resolution of the neighboring cameras at the target's predicted position; and calculating the virtual synthesis feasibility score by combining the CUP, effective resolution, and preset projection transformation distortion weights.

[0023] In response to a virtual synthesis feasibility score greater than a preset synthesis threshold, a virtual close-up image of the target is synthesized through homography matrix transformation;

[0024] If the virtual synthesis feasibility score is less than or equal to the preset synthesis threshold, the synthesis is abandoned and the target is marked as invisible.

[0025] Preferably, the system counts the number of targets lost due to untimely scheduling and the total number of high-value targets within the current time period; calculates the actual loss rate and compares the difference between the actual loss rate and the maximum loss rate allowed by the system.

[0026] In response to the actual loss rate being greater than the maximum loss rate allowed by the system, the semantic entropy trigger threshold is reduced based on the product of the difference and the preset learning rate.

[0027] In response to an actual loss rate less than or equal to the system's maximum allowable loss rate, the semantic entropy trigger threshold is increased or maintained.

[0028] Preferably, the parameter initialization module is used to preset the semantic entropy trigger threshold, basic category weights and behavior weight coefficients, and establish a global field of view coordinate system; the entropy perception trigger module is used to calculate the scene semantic entropy flux in real time, and perform conditional judgment based on the scene semantic entropy flux to determine whether to trigger the emergency game scheduling process.

[0029] The emergency game scheduling module includes: a feature extraction unit, which is used to process the panoramic video stream acquired by the wide-angle monitoring array through a lightweight target detection algorithm to obtain the basic features of moving targets in the field of view;

[0030] The value assessment unit is used to combine basic features with a pre-set knowledge base and analyze the target information value potential energy through a physical gravitational potential energy model.

[0031] The cost calculation unit is used to process the current attitude of the sphere, the target position mapping angle, and the load data of the edge computing nodes through the mechanical-computing power joint cost function to determine the mechanical-computing power joint cost.

[0032] The decision generation unit is used to generate a timeliness risk ratio based on the target information value potential energy, the combined cost of mechanical and computing power, and the target escape time.

[0033] The execution feedback unit is used to select either a physical PTZ control signal or a virtual viewpoint synthesis command based on the comparison result between the timeliness risk ratio and the preset safety ratio.

[0034] Compared with the prior art, the present invention has the following beneficial effects:

[0035] 1. This invention introduces the scene semantic entropy flux index, which integrates fluid mechanics and information entropy theory to quantify the information generation rate and disorder level of the monitoring area in real time. Unlike traditional fixed patrol, this method can keenly perceive the information burst trend in high-concurrency scenarios and flexibly switch between low-entropy steady state and high-entropy emergency state. It solves the problem of the system's slow perception of nonlinear group sudden events and realizes the on-demand allocation and efficient utilization of resources.

[0036] 2. This invention constructs a target value assessment system based on a physical gravity model and a joint cost function of mechanical computing power to generate a time-risk ratio to guide game decision-making. This mechanism breaks the rigid logic of traditional first-in-first-out and transforms target category, behavior and dwell time into scheduling priority, ensuring that high-risk and high-value targets are locked first when physical resources are limited, effectively solving the technical problem of key evidence being lost due to scheduling delay.

[0037] 3. This invention establishes a dual-modal execution mechanism that complements physical PTZ control and virtual viewpoint synthesis; for targets with extremely high escape risk and untimely mechanical response, the system automatically switches to computational photography mode and uses data from nearby cameras to synthesize virtual close-ups; this design breaks through the physical limits of the rotation speed and zoom travel of a single PTZ camera, and fills the physical blind spot by trading computing power for time, significantly improving the system's ability to capture highly dynamic targets;

[0038] 4. This invention designs an entropy threshold adaptive drift mechanism based on the actual loss rate; the system can dynamically correct the semantic entropy trigger threshold according to the target loss caused by scheduling lag; when the loss rate exceeds the standard, the system sensitivity is automatically increased, and otherwise redundant computing power is released; this closed-loop self-evolution capability enables the system to adapt to monitoring environments of different time periods and densities, ensuring long-term operational stability and robustness in complex and ever-changing scenarios. Attached Figure Description

[0039] The present invention will be further explained below with reference to the accompanying drawings and embodiments:

[0040] Figure 1 This is a flowchart of the method of the present invention;

[0041] Figure 2 This is a structural diagram of the system of the present invention. Detailed Implementation

[0042] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to specific embodiments.

[0043] Example 1:

[0044] Please see Figure 1 A video-linked intelligent scheduling and control method for cameras includes: pre-setting a semantic entropy trigger threshold, basic category weights, and behavior weight coefficients; initializing and generating a global field-of-view coordinate system based on the wide-angle monitoring coverage area; calculating the scene semantic entropy flux in real time; maintaining the current monitoring state in response to the scene semantic entropy flux being less than or equal to the semantic entropy trigger threshold; and triggering an emergency game-theoretic scheduling process in response to the scene semantic entropy flux being greater than the semantic entropy trigger threshold, including:

[0045] Based on the panoramic video stream acquired by the wide-angle monitoring array, the basic features of moving targets in the field of view are obtained through lightweight target detection algorithm.

[0046] Combining basic features and a pre-set knowledge base, the target information value potential energy is calculated through analysis using a physical gravitational potential energy model; the current attitude of the sphere, the target position mapping angle, and the load data of the edge computing nodes are processed through a mechanical-computing power joint cost function to determine the mechanical-computing power joint cost.

[0047] Based on the target information value potential energy, the combined cost of mechanical and computing power, and the target escape time, a timeliness risk ratio is generated; in response to the timeliness risk ratio being less than the preset safety ratio, a physical gimbal control signal is output; in response to the timeliness risk ratio being greater than or equal to the preset safety ratio, a virtual viewpoint synthesis command is output.

[0048] This embodiment provides a video-linked camera intelligent scheduling and control method, which aims to solve the problem of traditional scheduling mechanisms failing in high-concurrency nonlinear group emergencies;

[0049] The system performs initialization and parameter calibration steps; in this step, the semantic entropy trigger threshold is preset. Basic category weights and behavioral weighting coefficient To ensure that the semantic entropy trigger threshold is met. The value of conforms to the physical laws of the actual scenario. This embodiment adopts a statistical calibration method based on historical data. Specifically, a sample video set containing historical emergencies is selected, and the sample semantic entropy flux of each frame in the sample set is calculated. and will The preset quantile value of the probability density function is used as the semantic entropy trigger threshold. For example, the 95th percentile value is taken to ensure that only significantly abnormal entropy increase events will be triggered; at the same time, the global field of view coordinate system is initialized and generated based on the wide-angle monitoring coverage area. This step constructs the mapping relationship from the pixel coordinate system to the world coordinate system by reading the intrinsic and extrinsic parameter matrices of the wide-angle camera, thereby ensuring that the position information of all targets is unified under the same physical space reference.

[0050] The system enters the real-time monitoring phase, calculating the scene semantic entropy flux in real time. This indicator is a physical quantity that characterizes the rate of information generation and the degree of chaos within the monitored area; the system will calculate the scene semantic entropy flux in real time. With the preset semantic entropy trigger threshold Compare; if scene semantic entropy flux Less than or equal to the semantic entropy trigger threshold This indicates that the current scene is in a low-entropy steady state, and the system maintains its current monitoring state, i.e., performs a routine scan according to the preset cruise path; if the scene semantic entropy throughput... Greater than the semantic entropy trigger threshold This indicates that the information burst rate in the scene exceeds the normal processing capacity, and the system immediately triggers the emergency game scheduling process.

[0051] The emergency game scheduling process includes the following steps: Based on the panoramic video stream acquired by the wide-angle monitoring array, a lightweight target detection algorithm is used to process the data and obtain the basic features of moving targets in the field of view; the lightweight target detection algorithm preferably adopts the YOLOv8-Nano network model, and its output basic features include the first... One goal is pixel coordinates at time and Velocity vector and target category confidence Combining basic features and a pre-set knowledge base, the potential energy of the target information value is calculated through analysis using a physical gravitational potential energy model. This potential energy value is used to quantify the target's scheduling priority for the camera; the current attitude of the PTZ camera, the target position mapping angle, and the edge computing node load data are processed through a joint mechanical and computing cost function to determine the joint mechanical and computing cost. This cost characterizes the first Camera number 1 switched to number 2 The total time cost required to achieve the target;

[0052] Based on the value potential of target information The combined cost of machinery and computing power and target escape time Generate a timeliness risk ratio; this timeliness risk ratio is the combined cost of machinery and computing power. escape time of the target The system calculates the quotient value of the time-risk ratio and performs flow control based on the time-risk ratio: if the time-risk ratio is less than the preset safety ratio, it indicates that physical capture is feasible, and the system outputs a physical PTZ control signal to drive the PTZ camera to rotate and focus to lock onto the target; if the time-risk ratio is greater than or equal to the preset safety ratio, it indicates that the risk of physical capture failure is extremely high, and the system outputs a virtual viewpoint synthesis instruction to call the computational photography module to generate a virtual image of the target; the above preset safety ratio can be set according to Shannon's sampling theorem and system response delay, and is usually between 0.8 and 1.0.

[0053] Through the above method, the present invention can achieve keen perception and intelligent scheduling of high-entropy scenarios, effectively solving the contradiction between the limited physical resources and the dynamic nature of the target.

[0054] Example 2:

[0055] Real-time calculation of scene semantic entropy flux: Based on panoramic video stream, optical flow method is used to calculate the displacement of adjacent frames and obtain the target velocity vector; Based on panoramic video stream, target detection box intersection-union heatmap is used to generate normalized target spatial density function; Based on normalized target spatial density function and target velocity vector, global kinetic energy integral term representing group motion energy is calculated;

[0056] Based on the target velocity vector, calculate the global divergence integral term that characterizes the discrete deformation rate of the population;

[0057] The global kinetic energy integral term and the global divergence integral term are multiplied by preset kinetic energy conversion coefficients and structural entropy conversion coefficients, respectively, and then summed to generate the scene semantic entropy flux.

[0058] This embodiment is a further specification of the step of real-time calculation of scene semantic entropy flux in Embodiment 1; real-time calculation of scene semantic entropy flux. The process follows the logic of integrating the fluid mechanics flux conservation law and information entropy theory; the specific steps are as follows:

[0059] Based on panoramic video streams, optical flow is used to calculate the displacement between adjacent frames, thereby obtaining the target velocity vector. To ensure dimensional uniformity, the system introduces a pre-calibrated homography matrix. The optical flow vector of the image plane Projected onto the physical ground plane, the physical velocity vector is calculated. Unit: m / s, used as the reference for subsequent calculations. The velocity vector Characterizes spatial location The target's direction and speed of motion at the location;

[0060] Based on panoramic video streams, and utilizing the distribution of target detection box center points, a Gaussian kernel density estimation algorithm is employed to generate a continuous normalized target spatial density function. The specific formula is as follows ,in, This represents the total number of moving targets detected within the field of view. Let be the coordinate vector of any spatial position within the field of view. For the first The coordinates of the grounding point of each target. For bandwidth The Gaussian kernel function; this function The unit is Its value is directly proportional to the degree of crowding of targets within the area;

[0061] Based on the normalized objective space density function With the target velocity vector Calculate the global kinetic energy integral term representing the energy of group motion; specifically, this global kinetic energy integral term is... ,in For wide-angle surveillance coverage of the two-dimensional field of view, This is the velocity square field; its physical significance lies in quantifying the total kinetic energy of all targets in the scene.

[0062] Based on the target velocity vector Calculate the global divergence integral term characterizing the discrete deformation rate of the population; specifically, this global divergence integral term is... ,in The velocity field divergence characterizes the discrete or convergent rate of a group of targets.

[0063] Multiply the global kinetic energy integral term and the global divergence integral term by a preset kinetic energy conversion coefficient. and structural entropy conversion coefficient Then, summation is performed to generate the scene semantic entropy flux. The calculation formula is: ;in, This indicates the two-dimensional field of view covered by wide-angle surveillance. Surface integral, kinetic energy conversion coefficient The unit is structural entropy conversion coefficient The unit is also The introduction of both ensures a unified mapping from physical dimensions to information processing load dimensions.

[0064] To determine the specific values ​​of the aforementioned coefficients, this embodiment employs a dual-scenario normalization calibration method: a standard test video set is pre-set, comprising a sparse, fast-moving population dominated by kinetic energy and a high-density, low-speed crawling population dominated by structural entropy; adjustments are made... and The ratio ensures that the peak semantic entropy flux of the system remains on the same order of magnitude in both scenarios; as a preferred empirical value, it is set to... for , The range of values ​​is to This allows for a moderate increase in the system's sensitivity to crowding deformation; through this computational model, the system can accurately perceive the rate of information temperature rise in the scene.

[0065] Example 3:

[0066] The target information value potential energy is calculated, including: obtaining the basic category weights of the target from a pre-set knowledge base; generating the target's behavioral semantic score using a behavior analysis algorithm; and calculating the target's dwell time in the field of view.

[0067] The completeness of acquired information is obtained based on the quality assessment of historical frame images; a potential energy field equation is constructed with the basic category weight and behavioral semantic score as gain terms, and the target dwell time and the completeness of acquired information as attenuation terms, and the potential energy of target information value is calculated by solving the equation.

[0068] This embodiment is a further specification of the steps in Embodiment 1 for calculating the target information value potential energy; calculating the target information value potential energy The process specifically includes the following steps: obtaining the target's basic category weights from a pre-defined knowledge base. The knowledge base stores preset priorities for different categories of targets, for example, setting the weight of pedestrians to 1, vehicles to 0.8, and fire sources to 10.

[0069] Generate behavioral semantic scores for targets using behavioral analysis algorithms This rating It is derived from real-time analysis of the target's movement trajectory and human posture. For example, when running behavior is detected, the score is recorded as 2, and when walking normally, the score is recorded as 0.

[0070] Calculate the target's dwell time in the field of view. that time The duration of existence of the unique identifier ID generated by the tracking algorithm is used to reflect the scheduling logic that prioritizes newly emerging targets;

[0071] The completeness of acquired information is obtained based on the quality assessment of historical frame images. The completeness The value ranges from 0 to 1, and is derived from a weighted evaluation of the sharpness and resolution of historical captured images of the target.

[0072] Based on basic category weights and behavioral semantic scoring For the gain term, with the target residence time and the completeness of the information obtained As the attenuation term, a potential energy field equation is constructed, and the potential energy of the target information value is calculated. The specific potential energy field equation is as follows:

[0073] in, The sensitivity index for missing information is recommended to range from 1.5 to 3.0; it applies only if the completeness of the obtained information is... Lower When the value is large, the target dwell time Only then will the gain effect on potential energy be significantly amplified; if If the value approaches 1, the target has been clearly captured, the dwell time term is suppressed, and the system no longer pays too much attention to the target, thus conforming to the technical logic of prioritizing the scheduling of high-risk targets that have not been clearly captured.

[0074] in, These are the behavioral weighting coefficients; This is the time-sensitivity decay factor, and its physical unit is set to Hertz. or the reciprocal of seconds Used to offset target dwell time The model uses time dimensions to ensure consistency in the physical meaning of the denominator; through dimensional normalization, it ensures that high-risk targets that are not clearly captured have a reasonable maximum potential energy value after prolonged stay.

[0075] Example 4:

[0076] Determine the joint mechanical-computing cost, including: calculating the angle difference between the target mapping angle and the current attitude of the PTZ camera, and combining it with the preset maximum angular velocity of the gimbal to determine the mechanical time delay; obtaining the zoom motor travel time as the zoom time delay; monitoring the real-time load of the edge computing node to obtain the average inference latency as the computing power time delay; and superimposing the mechanical time delay, zoom time delay and computing power time delay to determine the joint mechanical-computing cost.

[0077] This embodiment is a further specification of the step in Example 1 for determining the combined cost of machinery and computing power; determining the combined cost of machinery and computing power The process aims to normalize and superimpose the physical motion delay and the computational processing delay. The specific steps are as follows:

[0078] Before calculating the angle difference, the system performs a mapping step from pixel coordinates to physical angles; based on the intrinsic parameter matrix of the wide-angle surveillance camera. Including focal length and principal point coordinates Using the inverse transformation formula of the pinhole imaging model, the pixel coordinates of the target in the panoramic image are determined. Converted to target mapping angle relative to the optical center The specific calculation formula is as follows:

[0079]

[0080] in, and The system maps the absolute azimuth and pitch angles of the wide-angle camera itself. After mapping, it calculates the angle difference between the target mapping angle and the current attitude of the PTZ camera. Combined with the preset maximum angular velocity of the gimbal, it determines the mechanical time delay. In this step, to ensure consistency of physical dimensions, the system converts all angle values ​​to radians. Specifically, the mechanical time delay consists of horizontal rotation time delay and vertical rotation time delay, calculated using the following formula: ,in and These are the target's horizontal azimuth and elevation angles, respectively, in Rad. and These are the current horizontal azimuth and elevation angles of the PTZ camera, in Rad. and These are the maximum angular velocities of the gimbal in the horizontal and vertical directions, respectively, in Rad / s. These values ​​are derived from the hardware specifications.

[0081] The mechanical time delay, zoom time delay, and computing power time delay are superimposed to determine the combined cost of mechanical and computing power. The calculation formula is: This cost function achieves a unified measure of multimodal delay, with its final physical unit being seconds (s).

[0082] Example 5:

[0083] The generation of timeliness risk ratio includes: determining the effective line of sight distance based on the straight-line distance from the target's current position along the target velocity vector direction to the boundary of the effective coverage area;

[0084] The effective escape time of the target is calculated based on the ratio of the effective line-of-sight distance to the magnitude of the target velocity vector; the time-risk ratio is generated by dividing the combined mechanical and computational cost by the effective escape time of the target.

[0085] This embodiment is a further specification of the step of generating the timeliness risk ratio in Embodiment 1;

[0086] The process of generating the timeliness-risk ratio is mainly used to assess the feasibility of physical capture, and the specific steps are as follows:

[0087] Based on the target's current location Along the target velocity vector The effective line-of-sight distance from the direction to the boundary of the effective coverage area is used to determine the effective line-of-sight distance. The effective coverage area boundary is defined as the camera's field of view boundary or the minimum recognizable pixel density boundary.

[0088] Based on effective field of vision modulus of the target velocity vector The ratio of the two values ​​is used to calculate the target's effective escape time. To more accurately reflect the radial escape velocity, the angle between the velocity vector and the boundary normal is introduced in the calculation. The correction, among which The angle between the target velocity vector and the boundary normal pointing out of the region is given by the following formula: In particular, when At this time, the target moves inward into the area, moves parallel to the boundary, or remains stationary. The system will The system's maximum time constant is set to a preset value, such as 9999 seconds, thereby reducing the urgency of physical capture of the target in subsequent risk ratio calculations; this time represents the remaining time window before the target moves out of the monitoring range or becomes unrecognizable.

[0089] Based on the combined cost of mechanical and computing power Target's effective escape time and the target information value potential energy calculated above. Generate a weighted timeliness risk ratio This step aims to establish a ternary game model of value, cost, and risk. Considering that high-value objectives should have higher scheduling tolerance, this embodiment constructs the following computational model:

[0090] in, The base of the natural logarithm is used; the logarithmic function is introduced to smooth the impact of extremely high-value targets on the sharp drop in the risk ratio, preventing systemic overreaction. This ratio... This intuitively reflects the game relationship between the system response speed and the target escape velocity under target value weighting: target value The higher the calculated risk ratio, the better. The lower the value, the easier it is to meet the condition of being less than the preset safety ratio, even when the escape time is short, prompting the system to prioritize physical capture of high-value targets.

[0091] Example 6:

[0092] Output virtual viewpoint synthesis instructions, including: determining the set of neighboring cameras that overlap with the current main PTZ camera's field of view; calculating the intersection-union ratio (IUU) between the neighboring camera's field of view and the target's predicted position; and obtaining the effective resolution of the neighboring cameras at the target's predicted position.

[0093] The virtual synthesis feasibility score is calculated by combining the intersection-union ratio, effective resolution, and preset projection transformation distortion weights.

[0094] In response to a virtual synthesis feasibility score greater than a preset synthesis threshold, a virtual close-up image of the target is synthesized through homography matrix transformation;

[0095] If the virtual synthesis feasibility score is less than or equal to the preset synthesis threshold, the synthesis is abandoned and the target is marked as invisible.

[0096] This embodiment is a further specification of the step of outputting virtual viewpoint synthesis instructions in Embodiment 1;

[0097] Outputting virtual viewpoint synthesis instructions is a remedial measure when physical scheduling cannot meet timeliness requirements. The specific steps are as follows:

[0098] Determine the set of neighboring cameras that overlap with the current field of view of the main PTZ camera. This set is derived from the system's pre-set camera topology matrix; calculate the field of view of neighboring cameras. Predicted location of the target intersection ratio ; Obtain the location of nearby cameras at the predicted target position Effective resolution at The resolution is measured in pixels per meter; combined with the intersection-union ratio (IU). Effective resolution and preset projection transformation distortion weights and introduce system reference resolution. Normalization is performed to calculate the virtual synthesis feasibility score. The calculation formula is:

[0099] Among them, the projection transformation distortion weight The weighting is inversely proportional to the angle between the principal optical axis of the adjacent camera and the target; the specific weighting calculation function is set as follows: ,in For the first The angle between the principal optical axis vector of the camera and the vector pointing from the optical center of the camera to the target position; the principal optical axis vector mentioned here specifically refers to the instantaneous physical direction of the neighboring camera before it rotates, not the theoretical direction after rotation; this is because the virtual viewpoint synthesis command is a millisecond-level remedial measure when physical scheduling cannot respond in time, and must be sampled in real time based on the current actual posture of the neighboring camera; when the angle is 0 degrees, it is a direct view, and the weights are... The distortion is minimal when the angle increases; as the angle increases, the weight decreases according to the cosine law, thereby quantifying the negative impact of perspective distortion on the quality of image synthesis. The more positive the angle, the greater the weight. The preset baseline pixel density that meets the minimum requirements of the synthesis algorithm, for example This normalization term eliminates the influence of hardware specification differences on the score dimensions, making It becomes a dimensionless evaluation indicator;

[0100] Perform the synthesis decision: if the virtual synthesis feasibility score is... If the value exceeds a preset synthesis threshold, the system synthesizes a virtual close-up image of the target using homography matrix transformation to fill in the physical blind spot; if the virtual synthesis feasibility score is... If the value is less than or equal to the preset synthesis threshold, it indicates that effective visual information cannot be obtained, and the system abandons the synthesis and marks the target as invisible.

[0101] Example 7:

[0102] Calculate the number of targets lost due to untimely scheduling and the total number of high-value targets within the current time period; calculate the actual loss rate and compare the difference between the actual loss rate and the maximum loss rate allowed by the system;

[0103] In response to the actual loss rate being greater than the maximum loss rate allowed by the system, the semantic entropy trigger threshold is reduced based on the product of the difference and the preset learning rate.

[0104] In response to an actual loss rate less than or equal to the system's maximum allowable loss rate, the semantic entropy trigger threshold is increased or maintained.

[0105] This embodiment adds an adaptive adjustment mechanism, namely an entropy threshold adaptive drift step, based on embodiment 1.

[0106] This step specifically includes: calculating the number of targets lost due to untimely scheduling within the current time period. and the total number of high-value targets Among them, the number of targets lost. The interruption count originates from the tracking algorithm;

[0107] Calculate the actual loss rate and compare it with the system's maximum allowable loss rate. Perform a difference comparison; the actual loss rate is and The ratio;

[0108] Adjust the semantic entropy trigger threshold based on the comparison results. If the actual loss rate is greater than the system's maximum allowable loss rate This indicates that the system's current sensitivity is insufficient, and the threshold needs to be lowered to capture more details; if the actual loss rate is less than or equal to the system's maximum allowable loss rate... This indicates that the system performance has a margin, and the threshold can be appropriately increased to save computing power; in order to comply with the principle of dimensional conservation, this embodiment adopts a proportional adaptive feedback correction formula:

[0109] in, To be a function that maximizes the value, The minimum positive entropy threshold preset for the system, for example This lower bound constraint prevents system crashes caused by the threshold flipping to a negative value due to a negative calculation result within the parentheses under extremely high loss rates. The preset learning rate is dimensionless, for example, between 0.1 and 0.5; this formula ensures the corrected learning rate through multiplicative gain. Still maintain with Consistent physical units This ensures the mathematical rigor of the physical model while achieving dynamic optimization.

[0110] Example 8:

[0111] Please see Figure 2 The parameter initialization module is used to preset the semantic entropy trigger threshold, basic category weights and behavior weight coefficients, and establish a global field of view coordinate system; the entropy perception trigger module is used to calculate the scene semantic entropy flux in real time and perform conditional judgment based on the scene semantic entropy flux to determine whether to trigger the emergency game scheduling process.

[0112] The emergency game scheduling module includes:

[0113] The feature extraction unit is used to process the panoramic video stream acquired by the wide-angle monitoring array and obtain the basic features of moving targets in the field of view through a lightweight target detection algorithm.

[0114] The value assessment unit is used to combine basic features with a pre-set knowledge base and analyze the target information value potential energy through a physical gravitational potential energy model.

[0115] The cost calculation unit is used to process the current attitude of the sphere, the target position mapping angle, and the load data of the edge computing nodes through the mechanical-computing power joint cost function to determine the mechanical-computing power joint cost.

[0116] The decision generation unit is used to generate a timeliness risk ratio based on the target information value potential energy, the combined cost of mechanical and computing power, and the target escape time.

[0117] The execution feedback unit is used to select either a physical PTZ control signal or a virtual viewpoint synthesis command based on the comparison result between the timeliness risk ratio and the preset safety ratio.

[0118] This embodiment provides a video-linked camera intelligent scheduling and control system, which is applied to the method described in any one of embodiments 1 to 7 above;

[0119] The system includes a parameter initialization module, an entropy-aware triggering module, and an emergency game scheduling module; the parameter initialization module is used to preset the semantic entropy triggering threshold. Basic category weights and behavioral weighting coefficient And establish a global field-of-view coordinate system;

[0120] The entropy-aware triggering module is used to calculate the semantic entropy flux of the scene in real time. And perform scene-based semantic entropy flux. The conditions are used to determine whether to trigger the emergency game scheduling process;

[0121] The emergency game scheduling module further includes a feature extraction unit, a value assessment unit, a cost calculation unit, a decision generation unit, and an execution feedback unit. The feature extraction unit is used to process the panoramic video stream acquired by the wide-angle monitoring array using a lightweight target detection algorithm to obtain the basic features of moving targets in the field of view. The value assessment unit is used to combine the basic features with a preset knowledge base and calculate the target information value potential energy through analysis using a physical gravitational potential energy model. The cost calculation unit is used to process the current attitude of the PTZ camera, the target position mapping angle, and the load data of the edge computing nodes through a joint mechanical and computing cost function to determine the joint mechanical and computing cost. The decision generation unit is used to generate value potential based on target information. The combined cost of machinery and computing power and target escape time The system generates a timeliness risk ratio; the execution feedback unit is used to select one of the physical PTZ control signal or virtual viewpoint synthesis command based on the comparison result between the timeliness risk ratio and the preset safety ratio; the various modules of the system work together to realize intelligent perception and efficient scheduling for high-entropy scenarios.

[0122] Example 9:

[0123] This embodiment uses security monitoring for large-scale sporting events, such as marathons, as an example to illustrate the practical application effect of this method. In high-density crowd gathering areas such as the starting area, when abnormal collisions or nonlinear diffusion occur among the crowd, the global divergence integral term... It will spike rapidly, causing the scene semantic entropy flux to increase. Exceeding the threshold At this point, the system triggers emergency dispatch, and the value assessment unit marks the identified fall behavior or fire source as high-weight. And through timeliness risk ratio Judgment; if the crowd moves too fast, causing Extremely short and with physical rotation of the gimbal If there is no time to respond, the system will immediately synthesize a virtual close-up using a nearby camera array, thereby obtaining clear evidence images of high-value targets without losing physical tracking time, thus solving the scheduling failure problem under the impact of high-throughput information.

[0124] 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 it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention.

Claims

1. A method for intelligent scheduling and control of video-linked cameras, characterized in that, include: Preset semantic entropy trigger threshold, basic category weight, and behavior weight coefficient; A global field-of-view coordinate system is initialized based on the wide-angle surveillance coverage area. Real-time calculation of scene semantic entropy flux; maintain the current monitoring state in response to scene semantic entropy flux being less than or equal to the semantic entropy trigger threshold. In response to the scenario semantic entropy flux exceeding the semantic entropy trigger threshold, an emergency game scheduling process is triggered, including: Based on the panoramic video stream acquired by the wide-angle monitoring array, the basic features of moving targets in the field of view are obtained through lightweight target detection algorithm. Combining the basic features with the preset knowledge base, the target information value potential energy is calculated through physical gravitational potential energy model analysis. The current attitude of the sphere, the target position mapping angle and the load data of the edge computing node are processed through the mechanical-computing power joint cost function to determine the mechanical-computing power joint cost. Based on the target information value potential energy, the combined cost of mechanical and computing power, and the target escape time, a timeliness risk ratio is generated; When the time-sensitive risk ratio is less than the preset safety ratio, a physical PTZ control signal is output. When the timeliness risk ratio is greater than or equal to the preset safety ratio, a virtual viewpoint synthesis command is output.

2. The intelligent scheduling and control method for video-linked cameras according to claim 1, characterized in that, Real-time computation of scene semantic entropy flux, including: Based on the panoramic video stream, the displacement between adjacent frames is calculated using the optical flow method to obtain the target velocity vector; Based on the panoramic video stream, a normalized target spatial density function is generated using the intersection-union heatmap of the target detection boxes. Based on the normalized target space density function and the target velocity vector, the global kinetic energy integral term characterizing the energy of group motion is calculated; Based on the target velocity vector, calculate the global divergence integral term that characterizes the discrete deformation rate of the population; The global kinetic energy integral term and the global divergence integral term are multiplied by preset kinetic energy conversion coefficients and structural entropy conversion coefficients, respectively, and then summed to generate the scene semantic entropy flux.

3. The intelligent scheduling and control method for video-linked cameras according to claim 1, characterized in that, The target information value potential energy is calculated by: obtaining the basic category weights of the target from a preset knowledge base; generating the target's behavioral semantic score using a behavior analysis algorithm; calculating the target's dwell time in the field of view; obtaining the completeness of the acquired information based on the quality assessment of historical frame images; and constructing a potential energy field equation with the basic category weights and behavioral semantic score as gain terms and the target dwell time and the completeness of the acquired information as attenuation terms, and then solving the target information value potential energy.

4. The intelligent scheduling and control method for video-linked cameras according to claim 1, characterized in that, Determine the combined mechanical and computing costs, including: calculating the angle difference between the target mapping angle and the current attitude of the PTZ camera, and combining it with the preset maximum angular velocity of the gimbal to determine the mechanical time delay; obtaining the zoom motor travel time as the zoom time delay; and monitoring the real-time load of edge computing nodes to obtain the average inference latency as the computing time delay. The mechanical time delay, zoom time delay, and computing power time delay are superimposed to determine the combined cost of mechanical and computing power.

5. The intelligent scheduling and control method for video-linked cameras according to claim 1, characterized in that, The generation of timeliness risk ratio includes: determining the effective line of sight distance based on the straight-line distance from the target's current position along the target velocity vector direction to the boundary of the effective coverage area; The effective escape time of the target is calculated based on the ratio of the effective line-of-sight distance to the magnitude of the target velocity vector. Divide the combined cost of machine and computing power by the target's effective escape time to generate the timeliness risk ratio.

6. The intelligent scheduling and control method for video-linked cameras according to claim 1, characterized in that, Output virtual viewpoint synthesis instructions, including: determining the set of neighboring cameras that overlap with the current main PTZ camera's field of view; calculating the cross-union ratio (CUP) between the neighboring camera's field of view and the target's predicted position; obtaining the effective resolution of the neighboring cameras at the target's predicted position; and calculating the virtual synthesis feasibility score by combining the CUP, effective resolution, and preset projection transformation distortion weights. In response to a virtual synthesis feasibility score greater than a preset synthesis threshold, a virtual close-up image of the target is synthesized through homography matrix transformation; If the virtual synthesis feasibility score is less than or equal to the preset synthesis threshold, the synthesis is abandoned and the target is marked as invisible.

7. The intelligent scheduling and control method for video-linked cameras according to claim 1, characterized in that, The method further includes: counting the number of targets lost due to untimely scheduling and the total number of high-value targets within the current time period; calculating the actual loss rate and comparing the difference between the actual loss rate and the maximum loss rate allowed by the system; In response to the actual loss rate being greater than the maximum loss rate allowed by the system, the semantic entropy trigger threshold is reduced based on the product of the difference and the preset learning rate. In response to an actual loss rate less than or equal to the system's maximum allowable loss rate, the semantic entropy trigger threshold is increased or maintained.

8. A video-linked camera intelligent scheduling and control system, applied to the video-linked camera intelligent scheduling and control method according to any one of claims 1-7, characterized in that, include: The parameter initialization module is used to preset the semantic entropy trigger threshold, basic category weights and behavior weight coefficients, and to establish a global field of view coordinate system; The entropy-aware triggering module is used to calculate the scene semantic entropy flux in real time and perform conditional judgments based on the scene semantic entropy flux to determine whether to trigger the emergency game scheduling process. The emergency game scheduling module includes: The feature extraction unit is used to process the panoramic video stream acquired by the wide-angle monitoring array and obtain the basic features of moving targets in the field of view through a lightweight target detection algorithm. The value assessment unit is used to combine basic features with a pre-set knowledge base and analyze the target information value potential energy through a physical gravitational potential energy model. The cost calculation unit is used to process the current attitude of the sphere, the target position mapping angle, and the load data of the edge computing nodes through the mechanical-computing power joint cost function to determine the mechanical-computing power joint cost. The decision generation unit is used to generate a timeliness risk ratio based on the target information value potential energy, the combined cost of mechanical and computing power, and the target escape time. The execution feedback unit is used to select either a physical PTZ control signal or a virtual viewpoint synthesis command based on the comparison result between the timeliness risk ratio and the preset safety ratio.

Images