Intelligent multi-target extraction method and system for intelligent security monitoring video images

Abstract

The invention discloses an intelligent multi-target extraction method and system for intelligent security monitoring video images, and relates to the field of video monitoring technology. The system comprises a data acquisition module, a target detection and tracking module, a behavior analysis module and an evaluation and decision-making module. The technical key points are as follows: by giving priority to pedestrians and closing gates at road intersections, the safety of pedestrian passage is improved, the opening sequence of gates is dynamically controlled, vehicles at various intersections are ensured to pass in an orderly manner, traffic chaos is avoided, and then the congestion situation at road intersections is calculated and monitored in real time, and management is carried out by dispatching personnel, so that traffic pressure is promptly and effectively relieved, and smooth roads are ensured. According to the congestion situation and the specific situation of the intersection, a recommended value of the number of dispatched people is generated, personnel are reasonably arranged for traffic management, and resource allocation is optimized.

Description

Technical Field

[0001] The present invention relates to the technical field of video monitoring, and in particular to an intelligent multi-target extraction method and system for intelligent security monitoring video images. Background Art

[0002] Video surveillance is a system that uses cameras, networks, storage devices, and management software to monitor, record, and playback specific areas in real time. Its main purpose is to ensure safety, prevent crime, manage operations, and provide evidence when necessary. Specifically, a video surveillance system includes the following key parts:

[0003] Camera equipment: including various types of cameras, such as fixed cameras, PTZ cameras, infrared cameras, etc., used to capture real-time video images in the monitoring area; Transmission equipment: Video signals can be transmitted to the monitoring center via wired (such as network cables, optical fibers) or wireless (such as Wi-Fi, 4G / 5G networks); Storage devices: used to store video data, common ones include DVR (digital video recorder), NVR (network video recorder), NAS (network attached storage) and other devices; Display devices: devices used to display video images in real time, such as monitors, TV walls, etc.; Management software: Video management system (VMS) software is used to manage, control and schedule various parts of the entire monitoring system, including video recording, playback, intelligent analysis and other functions; Power supply: Power supply to ensure the normal operation of all equipment, including power adapters, POE (Power over Ethernet) devices, etc.

[0004] The existing application publication number is CN112966164A, and the name is a fast positioning method and device for extracting targets from massive remote sensing images. The technical solution provided in the method and device includes: obtaining an image to be processed containing a target area and a target extraction range surface and a multi-target area seed surface of a target to be monitored; using preset features and based on the target extraction range surface to filter the image to be processed to obtain a filtered image; using the multi-target area seed surface to construct a target area spatial index for the filtered image to obtain a search image; cropping the search image to extract targets from the cropped image, and quickly matching and cropping the image to be processed of the target area using corresponding seed surface primitives to obtain corresponding slice files, so as to solve the problem of low retrieval efficiency in the process of extracting targets from massive remote sensing images, but it cannot simultaneously track or extract multiple features for monitoring video images.

[0005] In combination with the above documents and existing technologies, security cameras are usually installed at the intersections in underground garages to achieve the purpose of security monitoring. At the same time, relevant video image recognition technology is also used to conduct multi-target tracking of the monitored features, i.e. pedestrians or vehicles. However, when the underground garage encounters holidays or opening hours, there are usually multiple streams of traffic at the road intersection, and even pedestrians. If not controlled, it will cause traffic jams or even traffic accidents. If only one manager is sent there, it cannot meet the actual needs of different road intersections. Summary of the invention

[0006] 1. Technical issues to be solved

[0007] In view of the deficiencies in the prior art, the present invention provides an intelligent multi-target extraction method and system for intelligent security monitoring video images. The system can effectively manage complex intersections in the direction of underground garage exits, give priority to pedestrian safety, monitor and respond to traffic congestion in real time, optimize traffic flow, and improve the efficiency and effectiveness of overall traffic management, thereby solving the problems raised in the background technology.

[0008] (II) Technical solution

[0009] To achieve the above objectives, the present invention is implemented through the following technical solutions:

[0010] Intelligent multi-target extraction system for intelligent security monitoring video images, including:

[0011] The data acquisition module obtains the original data stream from the security camera and pre-processes the original data stream;

[0012] The target detection and tracking module uses a target detection algorithm to identify the location and category of the target in the original data stream, and uses a target tracking algorithm to continuously locate and track the target;

[0013] The behavior analysis module uses motion analysis technology to analyze the motion trajectory of each target in each raw data stream at the road intersection. It uses the pre-built rule engine to execute the priority prompt strategy when pedestrians are found among the targets in the corresponding raw data stream. If pedestrians are not found among the targets in the corresponding raw data stream, an early warning signal is triggered.

[0014] The evaluation and decision module, under the condition of receiving the warning signal, marks the center of the road intersection as the target point. When a single target at any intersection is detected passing the target point, the gate at the next intersection is opened. When a single target at the next intersection passes the target point, the gate at another intersection is opened, and the cycle repeats.

[0015] When any gate is open, the other gates are closed;

[0016] The traffic volume passing through the target point and the number of vehicles in the road intersection are counted within a unit time, the statistical data are normalized, the congestion coefficient Ir of the current road intersection is calculated, and the congestion coefficient Ir is compared with the preset standard threshold Bz. When the congestion coefficient Ir exceeds the standard threshold Bz, the management dispatch strategy is executed.

[0017] Furthermore, the monitoring image of a single security camera covers the corresponding road intersection in the underground parking lot, and the original data stream includes the real-time monitoring image data of the corresponding security camera.

[0018] Furthermore, the process of preprocessing the original data stream is as follows:

[0019] Decoding: Decoding the received video stream data into a processable image sequence;

[0020] Image quality enhancement: enhance the image sequence and remove the noise;

[0021] Time synchronization processing: Use timestamps to synchronize video streams obtained by different security cameras.

[0022] Furthermore, the targets include vehicles and pedestrians, the target detection algorithm includes any one of YOLO and Faster R-CNN, and the target tracking algorithm includes any one of a Kalman filter and a multi-target tracking algorithm.

[0023] Furthermore, the road intersection represents the crossroads in the exit direction of the underground garage. If the road intersection is three intersections, the number of gates configured is 2; if the road intersection is four intersections, the number of gates configured is 3, and so on. The positions of the gates are all at the intersections of the corresponding roads and the exits, and the number of gates configured in the corresponding road intersections is always 1 less than the number of intersections.

[0024] Furthermore, the process of the priority prompt strategy is: a voice prompt is given at the road intersection to give pedestrians priority, and the vehicle gate at the intersection is closed simultaneously, and the content of the voice prompt is: "Pedestrians have priority".

[0025] Furthermore, the process of detecting whether a single target at any intersection passes through the target point is as follows: use the target detection and tracking module to track the motion trajectory of each target, mark the center of the road intersection as the target point, detect the current position of the target in each video frame in the corresponding original data stream, and determine whether the motion trajectory of the target crosses the set target point area through geometric relationships. In this way, it can be detected whether a single target at any intersection passes through the target point.

[0026] Furthermore, the congestion coefficient Ir of the current road intersection is calculated based on the following formula:

[0027]

[0028] Where N c represents the current number of vehicles in the road intersection, C represents the capacity of the road intersection, that is, the maximum number of vehicles that the road intersection can accommodate without congestion, and F t Represents the traffic volume passing through the target point within unit time t.

[0029] Furthermore, the management dispatch strategy process executed is:

[0030] According to the current congestion coefficient and the number of intersections in the current road intersection, a basic calculation model is built to generate a recommended value for the number of people to be dispatched, and a corresponding number of people are arranged to go to the current road intersection according to the recommended value for the number of people to be dispatched. After the preset time T, if the congestion coefficient at this time is on a downward trend compared to the congestion coefficient before the preset time T, no response action is taken. If the congestion coefficient at this time is on an upward trend compared to the congestion coefficient before the preset time T, a secondary dispatch strategy is executed, and the number of people to be dispatched is increased by 1 each time, until the current congestion coefficient is on a downward trend compared to the previous congestion coefficient;

[0031] The process of generating the recommended value of the number of dispatched personnel after building the basic calculation model is as follows:

[0032] Preset variables and parameters: the number of intersections in the road intersection is recorded as N, the basic dispatched number is recorded as D0, that is, a basic dispatched number, and the error correction factor is recorded as K;

[0033] Formula establishment: Based on the normalized variables and parameters, the recommended value of the number of dispatched personnel is obtained, and the formula is as follows:

[0034] D = D0 + int(K*(Ir-Bz)*N);

[0035] In the formula, D represents the recommended value of the number of dispatched personnel.

[0036] The intelligent multi-target extraction method of intelligent security monitoring video images includes the following steps:

[0037] S1, obtain the original data stream from the security camera and pre-process the original data stream;

[0038] S2, using the target detection algorithm to identify the location and category of the target in the original data stream, and using the target tracking algorithm to continuously locate and track the target;

[0039] S3. At the road intersection, motion analysis technology is used to analyze the motion trajectory of each target in each raw data stream. When pedestrians are found among the targets in the corresponding raw data stream, a priority prompt strategy is executed using a pre-built rule engine. When pedestrians are not found among the targets in the corresponding raw data stream, an early warning signal is triggered.

[0040] S4. When receiving the warning signal, the center of the road intersection is marked as the target point. When a single target at any intersection is detected to pass the target point, the gate at the next intersection is opened. When the single target at the next intersection passes the target point, the gate at another intersection is opened, and the cycle repeats.

[0041] When any gate is open, the other gates are closed;

[0042] The traffic volume passing through the target point and the number of vehicles in the road intersection are counted within a unit time, the statistical data are normalized, the congestion coefficient Ir of the current road intersection is calculated, and the congestion coefficient Ir is compared with the preset standard threshold Bz. When the congestion coefficient Ir exceeds the standard threshold Bz, the management dispatch strategy is executed.

[0043] (III) Beneficial effects

[0044] The present invention provides an intelligent multi-target extraction method and system for intelligent security monitoring video images, which has the following beneficial effects:

[0045] The present invention improves the image quality by preprocessing the original video stream, making the subsequent target detection and tracking more accurate and efficient. By combining the use of target detection algorithm and target tracking algorithm, the intersection in the underground parking lot can be effectively monitored, and multiple targets, namely vehicles and pedestrians, can be detected and tracked in real time, thereby improving the overall security performance and accuracy of the monitoring system.

[0046] The present invention improves the safety of pedestrian traffic by giving priority to pedestrians and closing gates at road intersections, dynamically controls the opening order of gates, ensures orderly traffic of vehicles at each intersection, avoids traffic chaos, and then calculates and monitors the congestion situation at road intersections in real time, and manages it by dispatching personnel, thus promptly and effectively alleviating traffic pressure and ensuring smooth roads. According to the congestion situation and the specific conditions of the intersection, a recommended value for the number of dispatched people is generated, personnel are reasonably arranged for traffic management, and resource allocation is optimized. BRIEF DESCRIPTION OF THE DRAWINGS

[0047] Figure 1 It is a schematic diagram of the intelligent modular structure of the intelligent security monitoring video image of the present invention;

[0048] Figure 2A real-life picture of the shooting range of the security camera in the present invention;

[0049] Figure 3 The present invention is a flowchart of the overall steps of the intelligent multi-target extraction method of intelligent security monitoring video images. DETAILED DESCRIPTION

[0050] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0051] Example 1: Please refer to Figure 1 to Figure 2 This embodiment provides an intelligent multi-target extraction system for intelligent security monitoring video images; the design concept of the system is as follows:

[0052] The system includes a data acquisition module, a target detection and tracking module, a behavior analysis module, and an evaluation and decision-making module. The application scenario of the entire system is directed to underground parking lots or some other scenarios. Security monitoring is performed on these scenarios. By collecting video image data, detecting and tracking selected targets based on the image data, multi-target extraction of vehicles or pedestrians is performed, and the movement direction and trajectory of each target are analyzed. Subsequent evaluation and decision-making operations are performed based on the analysis results, playing a role in intelligent security and alert.

[0053] The data acquisition module obtains the original data stream from the security camera and pre-processes the original data stream;

[0054] Among them, security cameras are pre-installed in designated locations;

[0055] For underground parking lots, the surveillance images of security cameras need to cover all intersections in the underground parking lot. The center of the intersection is the center of the fork road, and this position is also the designated position.

[0056] The original data stream includes the real-time monitoring screen data of the corresponding security camera;

[0057] The process of preprocessing the original data stream is:

[0058] Decoding: Decoding received video stream data into a processable image sequence; Example: Using a video codec (such as H.264, H.265) to decode the video stream from a compressed format into raw image data;

[0059] Image quality enhancement: Enhance the image to improve the accuracy of subsequent analysis and processing; examples: remove noise from video, adjust contrast and brightness, or perform image stabilization to reduce jitter;

[0060] Video format conversion: converting a video stream into a format suitable for subsequent processing; example: converting a raw video stream into a specific image format (such as JPEG, PNG) or video format (such as AVI) so that subsequent algorithms can process it effectively;

[0061] Time synchronization processing: ensuring that video streams from different sources are synchronized in time for subsequent cross-camera analysis; Example: using timestamps to synchronize video streams from different cameras so that they can be merged or cross-analyzed;

[0062] In addition, it can also include frame rate adjustment and resolution adjustment;

[0063] Adjust the video frame rate according to system requirements to balance the real-time performance and processing efficiency of the video stream; Example: Reduce the high frame rate video stream to a frame rate suitable for analysis to save processing resources; Adjust the video resolution according to monitoring requirements and processing capabilities; Example: Reduce high-resolution video to a reasonable resolution to increase processing speed and save storage space;

[0064] Through these preprocessing steps, the data acquisition module can optimize the raw video stream into a format and quality suitable for subsequent target detection, tracking, and behavior analysis, thereby improving the overall performance and accuracy of the system.

[0065] The target detection and tracking module uses a target detection algorithm to identify the location and category of the target in the original data stream, and uses a target tracking algorithm to continuously locate and track the target;

[0066] Among them, targets include vehicles and pedestrians;

[0067] Object detection algorithms include deep learning-based object detectors, such as YOLO and Faster R-CNN;

[0068] YOLO:

[0069] Principle: The entire raw data stream, i.e., the image, is fed into the neural network at once to directly predict the bounding box and category probability; YOLO transforms the detection problem into a regression problem and completes the detection in one forward pass;

[0070] Advantages: fast, suitable for real-time applications;

[0071] Example: Detect pedestrians, vehicles and other objects in surveillance videos and annotate their bounding boxes and category labels;

[0072] Faster R-CNN:

[0073] Principle: Generate candidate regions (Region Proposals) based on convolutional neural networks (CNN), and then further perform classification and regression on these candidate regions to accurately locate the target;

[0074] Advantages: High detection accuracy, suitable for applications requiring high accuracy;

[0075] Example: Accurately detect and classify different types of objects, such as pedestrians and vehicles, in high-resolution surveillance videos;

[0076] Target tracking algorithms include Kalman filter and multi-target tracking algorithm;

[0077] Kalman Filter:

[0078] Principle: Based on the target's motion model and measurement data, the target's position and velocity are predicted and updated;

[0079] Advantages: high computational efficiency, suitable for linear motion target tracking;

[0080] Example: Tracking a continuously moving vehicle in a surveillance video, predicting its next location, and maintaining continuous monitoring;

[0081] Multi-target tracking algorithm:

[0082] Principle: Combined with data association methods, such as the Hungarian algorithm and JPDA (Joint Probabilistic Data Association), the detected targets are associated with the existing tracks and the independent tracking of each target is maintained;

[0083] Advantages: Ability to track multiple targets simultaneously and handle occlusion and interaction between targets;

[0084] Example: Tracking each individual in a crowded crowd and analyzing crowd dynamics.

[0085] The specific steps of using target detection algorithm and target tracking algorithm are as follows:

[0086] S101, object detection: extract images from the video frames of the original data stream, use the object detection algorithm to identify the location and category of the object, and generate a detection result including the object bounding box and category label;

[0087] S102, target initialization: Initialize the detected targets, assign unique tracking IDs, and obtain the initial position, speed, and tracking ID of each target;

[0088] S103, target prediction: predict the position of the target using a target tracking algorithm, consider the motion model of the target, and generate a predicted target position;

[0089] S104, data association: match the new detection result with the predicted position and update the tracking trajectory; if no suitable target is matched, start a new tracking instance to obtain the updated target position, speed and tracking ID;

[0090] S105, trajectory update: update the tracking trajectory of each target, process the target's entry, exit and occlusion, and obtain a continuous target trajectory and real-time position;

[0091] Through the above steps, the target detection and tracking module can effectively identify and track multiple targets in the video stream, and can maintain tracking even when the target is blocked or there is visual discontinuity, ensuring that the system can monitor and respond to various dynamic changes in real time.

[0092] By adopting the above technical solution, the original video stream is preprocessed to improve the image quality, making subsequent target detection and tracking more accurate and efficient. Combined with the use of target detection algorithm and target tracking algorithm, it is possible to effectively monitor intersections in underground parking lots, detect and track multiple targets, namely vehicles and pedestrians, in real time, thereby improving the overall security performance and accuracy of the monitoring system.

[0093] The behavior analysis module uses motion analysis technology to analyze the motion trajectory of each target in each raw data stream at the road intersection. It uses the pre-built rule engine to execute the priority prompt strategy when pedestrians are found among the targets in the corresponding raw data stream. If pedestrians are not found among the targets in the corresponding raw data stream, an early warning signal is triggered.

[0094] The road intersection refers to the crossroads in the exit direction of the underground garage. If the road intersection is a three-road intersection, the number of gates is 2; if the road intersection is a four-road intersection, the number of gates is 3, and so on. The positions of the gates are all at the intersections of the corresponding roads and exits. Figure 2 It can be seen that;

[0095] Motion analysis technology: Analyze the target's motion trajectory and speed through the motion vectors between video frames in the original data stream;

[0096] The process of the priority prompt strategy is:

[0097] At road intersections, a voice prompt is given to pedestrians to have priority, and the vehicle gates at the intersection are closed simultaneously. The content of the voice prompt is: "Pedestrians have priority."

[0098] The evaluation and decision module, under the condition of receiving the warning signal, marks the center of the road intersection as the target point. When a single target at any intersection is detected passing the target point, the gate at the next intersection is opened. When a single target at the next intersection passes the target point, the gate at another intersection is opened, and the cycle repeats.

[0099] When any gate is open, the other gates are closed;

[0100] Count the traffic volume passing through the target point and the number of vehicles in the road intersection within a unit time, normalize the statistical data, calculate the congestion coefficient Ir of the current road intersection, and compare the congestion coefficient Ir with the preset standard threshold Bz. When the congestion coefficient Ir exceeds the standard threshold Bz, it means that the current road intersection is more congested, and the management dispatch strategy is executed. When the congestion coefficient Ir does not exceed the standard threshold Bz, no response action is taken;

[0101] The process of detecting whether a single target at any intersection passes through the target point is as follows:

[0102] Use the target detection and tracking module to track the movement trajectory of each target, ensure that each target has a unique identifier (ID), and mark it as a target point at the center of the road intersection. In each video frame, detect the current position of the target and determine whether the target's movement trajectory crosses the set target point area through geometric relationships. This can detect whether a single target at any intersection passes through the target point.

[0103] Specific description: Figure 2 As shown in the figure, under the condition of receiving the warning signal, the center of the road intersection is marked as the target point. When a single target at the left intersection passes the target point, the gate at the lower intersection is opened. When a single target at the lower intersection passes the target point, the gate at the left intersection is reopened. Figure 2 For a four-fork road intersection, where the fourth fork is on the right, the gate at the right intersection will continue to be opened. When a single target at the right intersection passes the target point, the gate at the left intersection will be reopened, and the cycle will continue in a predetermined counterclockwise order.

[0104] When counting the traffic volume passing through the target point and the number of vehicles in the road intersection within a unit time, a traffic volume sensor can be installed at the target point. For example, if the traffic volume is 4 vehicles / minute, the number of vehicles in the road intersection is counted at the end of the unit time, that is, Figure 2 The number of vehicles within the range of the security cameras is 42;

[0105] The congestion coefficient Ir of the current road intersection is calculated based on the following formula:

[0106]

[0107] Where N c represents the current number of vehicles in the intersection, C represents the capacity of the intersection, that is, the maximum number of vehicles that the intersection can accommodate without congestion, and F t It represents the traffic volume passing the target point within the unit time t, that is, the number of vehicles, N c / C represents the ratio of the number of preceding vehicles to the intersection capacity. If the ratio is close to 1 or greater than 1, it means that the intersection is close to saturation or overload. c / F t It represents the ratio of the current number of vehicles to the traffic flow per unit time. This ratio reflects the digestibility of the traffic flow to the current number of vehicles. c / C multiplied by 1+N c / F t In order to increase the complexity, when the number of cars in the intersection increases, the congestion coefficient will increase nonlinearly, reflecting the increasing degree of congestion;

[0108] For example:

[0109] Assume that the capacity of a road intersection is 100 vehicles, the traffic volume passing the target point within a unit time (such as 1 minute) is 50 vehicles, and the number of vehicles in the current intersection is 60 vehicles. Substitute these values ​​into the formula and calculate the final congestion coefficient = 1.32; Result analysis: Congestion coefficient = 1.32 indicates that the current road intersection is in a certain degree of congestion, because the congestion coefficient exceeds the standard threshold, that is, 1.32>1. When the congestion coefficient is less than 1, that is, the standard threshold, it means that the current road intersection is relatively unobstructed. When the congestion coefficient exceeds 1, it means that the intersection is severely congested and traffic diversion measures need to be taken. Through the above formula and examples, the congestion situation of road intersections can be calculated and monitored in real time, providing important data support for traffic management and optimization.

[0110] The management dispatch policy process performed is:

[0111] According to the current congestion coefficient and the number of intersections in the current road intersection, a basic calculation model is built to generate a recommended value for the number of people to be dispatched, and a corresponding number of people are arranged to go to the current road intersection according to the recommended value for the number of people to be dispatched. After the preset time T, if the congestion coefficient at this time is on a downward trend compared to the congestion coefficient before the preset time T, no response action is taken. If the congestion coefficient at this time is on an upward trend compared to the congestion coefficient before the preset time T, a secondary dispatch strategy is executed, and the number of people to be dispatched is increased by 1 each time, until the current congestion coefficient is on a downward trend compared to the previous congestion coefficient;

[0112] The process of generating the recommended value of the number of dispatched personnel after building the basic calculation model is as follows:

[0113] Preset variables and parameters: The number of crossings in the road intersection is recorded as N, the basic dispatch number is recorded as D0, that is, a basic dispatch number, assuming that the minimum number of dispatches required under the most ideal situation (no congestion and one crossing), which is set according to the situation, and the error correction factor is recorded as K, which is used to adjust the parameters of the dispatch number;

[0114] Formula establishment: Based on the normalized variables and parameters, the recommended value of the number of dispatched personnel is obtained, and the formula is as follows:

[0115] D = D0 + int(K*(Ir-Bz)*N);

[0116] In the formula, D represents the recommended value of the number of dispatched personnel;

[0117] Example: Assume the following data:

[0118] <![CDATA[Basic dispatched number D0]]> 1 Error correction factor K 2 Current congestion factor 1.32 The number of intersections in the road intersection N 4

[0119] Substituting the above data into the formula, we can calculate the recommended number of people to be dispatched D = 1 + 2 = 3;

[0120] Therefore, according to the current congestion coefficient and the number of intersections, the recommended number of dispatched personnel is 3, and the dispatched personnel are used to direct the traffic at the current road intersection to avoid road congestion and speed up road clearing.

[0121] By adopting the above technical solution, pedestrians are given priority to pass and the gates are closed at the road intersections, which improves the safety of pedestrian traffic, dynamically controls the opening order of the gates, ensures the orderly passage of vehicles at each intersection, and avoids traffic chaos. Then, the congestion situation at the road intersection is calculated and monitored in real time, and management is carried out by dispatching personnel, which effectively alleviates the traffic pressure and ensures smooth roads. According to the congestion situation and the specific conditions of the intersection, the recommended value of the number of dispatched people is generated, and personnel are reasonably arranged for traffic management to optimize resource allocation.

[0122] In summary, the system can effectively manage complex intersections in the direction of the underground garage exit, give priority to the safety of pedestrians, monitor and respond to traffic congestion in real time, optimize traffic flow, and improve the efficiency and effectiveness of overall traffic management.

[0123] Example 2: Please refer to Figure 3 Based on Example 1, this embodiment also provides an intelligent multi-target extraction method for intelligent security monitoring video images, including the following specific steps:

[0124] S1, obtain the original data stream from the security camera and pre-process the original data stream;

[0125] S2, using the target detection algorithm to identify the location and category of the target in the original data stream, and using the target tracking algorithm to continuously locate and track the target;

[0126] S3. At the road intersection, motion analysis technology is used to analyze the motion trajectory of each target in each raw data stream. When pedestrians are found among the targets in the corresponding raw data stream, a priority prompt strategy is executed using a pre-built rule engine. When pedestrians are not found among the targets in the corresponding raw data stream, an early warning signal is triggered.

[0127] S4. When receiving the warning signal, the center of the road intersection is marked as the target point. When a single target at any intersection is detected to pass the target point, the gate at the next intersection is opened. When the single target at the next intersection passes the target point, the gate at another intersection is opened, and the cycle repeats.

[0128] When any gate is open, the other gates are closed;

[0129] The traffic volume passing through the target point and the number of vehicles in the road intersection are counted within a unit time, the statistical data are normalized, the congestion coefficient of the current road intersection is calculated, and the congestion coefficient is compared with the preset standard threshold. When the congestion coefficient exceeds the standard threshold, the management dispatch strategy is executed.

[0130] In the application, the several formulas involved are all calculated by removing the dimensions and taking their numerical values. The formula is a formula obtained by collecting a large amount of data and performing software simulation to obtain the most recent real situation. The formula is set by technical personnel in this field according to actual conditions.

[0131] The above embodiments may be implemented in whole or in part by software, hardware, firmware or any other combination thereof. When implemented using software, the above embodiments may be implemented in whole or in part in the form of a computer program product. A person of ordinary skill in the art may appreciate that the units and algorithm steps of each example described in conjunction with the embodiments disclosed herein may be implemented in electronic hardware, or in a combination of computer software and electronic hardware. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution.

[0132] The units described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, and may be located in one place or distributed on multiple network units. Some or all of the units may be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0133] The above description is only a specific implementation manner of the present application, but the protection scope of the present application is not limited thereto. Any technician familiar with the technical field can easily think of changes or substitutions within the technical scope disclosed in the present application, which should be included in the protection scope of the present application.

Claims

1. Intelligent multi-target extraction system for intelligent security monitoring video images, characterized in that: The system includes: The data acquisition module obtains the original data stream from the security camera and pre-processes the original data stream; The target detection and tracking module uses a target detection algorithm to identify the location and category of the target in the original data stream, and uses a target tracking algorithm to continuously locate and track the target; The behavior analysis module uses motion analysis technology to analyze the motion trajectory of each target in each raw data stream at the road intersection. It uses the pre-built rule engine to execute the priority prompt strategy when pedestrians are found among the targets in the corresponding raw data stream. If pedestrians are not found among the targets in the corresponding raw data stream, an early warning signal is triggered. The evaluation and decision module, under the condition of receiving the warning signal, marks the center of the road intersection as the target point. When a single target at any intersection is detected passing the target point, the gate at the next intersection is opened. When a single target at the next intersection passes the target point, the gate at another intersection is opened, and the cycle repeats. When any gate is open, the other gates are closed; Count the traffic volume passing through the target point and the number of vehicles in the road intersection within a unit time, normalize the statistical data, calculate the congestion coefficient Ir of the current road intersection, and compare the congestion coefficient Ir with the preset standard threshold Bz. When the congestion coefficient Ir exceeds the standard threshold Bz, execute the management dispatch strategy; The congestion coefficient Ir of the current road intersection is calculated based on the following formula: ; In the formula, Indicates the current number of cars in the road intersection. It represents the capacity of a road intersection, that is, the maximum number of vehicles that the road intersection can accommodate without congestion. It represents the traffic volume passing the target point in unit time t; The management dispatch policy process performed is: According to the current congestion coefficient and the number of intersections in the current road intersection, a basic calculation model is built to generate a recommended value for the number of people to be dispatched, and a corresponding number of people are arranged to go to the current road intersection according to the recommended value for the number of people to be dispatched. After the preset time T, if the congestion coefficient at this time is on a downward trend compared to the congestion coefficient before the preset time T, no response action is taken. If the congestion coefficient at this time is on an upward trend compared to the congestion coefficient before the preset time T, a secondary dispatch strategy is executed, and the number of people to be dispatched is increased by 1 each time, until the current congestion coefficient is on a downward trend compared to the previous congestion coefficient; The process of generating the recommended value of the number of dispatched personnel after building the basic calculation model is as follows: Preset variables and parameters: The number of intersections in the road intersection is recorded as N, the number of basic dispatched personnel is recorded as , which is a basic number of dispatches, and the error correction factor is denoted by K; Formula establishment: Based on the normalized variables and parameters, the recommended value of the number of dispatched personnel is obtained, and the formula is as follows: ; In the formula, D represents the recommended value of the number of dispatched personnel.

2. The intelligent multi-target extraction system for intelligent security monitoring video images according to claim 1 is characterized in that: The monitoring image of a single security camera covers the corresponding road intersection in the underground parking lot, and the original data stream includes the real-time monitoring image data of the corresponding security camera.

3. The intelligent multi-target extraction system for intelligent security monitoring video images according to claim 2 is characterized in that: The process of preprocessing the original data stream is: Decoding: Decoding the received video stream data into a processable image sequence; Image quality enhancement: enhance the image sequence and remove the noise; Time synchronization processing: Use timestamps to synchronize video streams obtained by different security cameras.

4. The intelligent multi-target extraction system for intelligent security monitoring video images according to claim 3 is characterized in that: The targets include vehicles and pedestrians, the target detection algorithm includes any one of YOLO and Faster R-CNN, and the target tracking algorithm includes any one of Kalman filter and multi-target tracking algorithm.

5. The intelligent multi-target extraction system of intelligent security monitoring video images according to claim 4 is characterized in that: The road intersection refers to the crossroads in the exit direction of the underground garage. If the road intersection is three intersections, the number of gates configured is 2. If the road intersection is four intersections, the number of gates configured is 3. And so on. The positions of the gates are all at the junctions of the corresponding roads and the exits, and the number of gates configured in the corresponding road intersections is always 1 less than the number of intersections.

6. The intelligent multi-target extraction system for intelligent security monitoring video images according to claim 5 is characterized in that: The process of the priority prompt strategy is: voice prompts pedestrians to have priority at road intersections, and the vehicle gates at the intersections are closed simultaneously, and the content of the voice prompt is: "Pedestrians have priority." 7. The intelligent multi-target extraction system for intelligent security monitoring video images according to claim 6 is characterized in that: The process of detecting whether a single target at any intersection passes through the target point is as follows: Use the target detection and tracking module to track the motion trajectory of each target, mark the center of the road intersection as the target point, detect the current position of the target in each video frame in the corresponding original data stream, and determine whether the target's motion trajectory crosses the set target point area through geometric relationships. In this way, it can be detected whether a single target at any intersection passes through the target point.

8. An intelligent multi-target extraction method for intelligent security monitoring video images, using any system described in claims 1 to 7, characterized in that: The steps include: S1, obtain the original data stream from the security camera and pre-process the original data stream; S2, using the target detection algorithm to identify the location and category of the target in the original data stream, and using the target tracking algorithm to continuously locate and track the target; S3. At the road intersection, motion analysis technology is used to analyze the motion trajectory of each target in each raw data stream. When pedestrians are found among the targets in the corresponding raw data stream, a priority prompt strategy is executed using a pre-built rule engine. When pedestrians are not found among the targets in the corresponding raw data stream, an early warning signal is triggered. S4. When receiving the warning signal, the center of the road intersection is marked as the target point. When a single target at any intersection is detected to pass the target point, the gate at the next intersection is opened. When the single target at the next intersection passes the target point, the gate at another intersection is opened, and the cycle repeats. When any gate is open, the other gates are closed; Count the traffic volume passing through the target point and the number of vehicles in the road intersection within a unit time, normalize the statistical data, calculate the congestion coefficient Ir of the current road intersection, and compare the congestion coefficient Ir with the preset standard threshold Bz. When the congestion coefficient Ir exceeds the standard threshold Bz, execute the management dispatch strategy; The congestion coefficient Ir of the current road intersection is calculated based on the following formula: ； In the formula, Indicates the current number of cars in the road intersection. It represents the capacity of a road intersection, that is, the maximum number of vehicles that the road intersection can accommodate without congestion. It represents the traffic volume passing the target point in unit time t; The management dispatch policy process performed is: According to the current congestion coefficient and the number of intersections in the current road intersection, a basic calculation model is built to generate a recommended value for the number of people to be dispatched, and a corresponding number of people are arranged to go to the current road intersection according to the recommended value for the number of people to be dispatched. After the preset time T, if the congestion coefficient at this time is on a downward trend compared to the congestion coefficient before the preset time T, no response action is taken. If the congestion coefficient at this time is on an upward trend compared to the congestion coefficient before the preset time T, a secondary dispatch strategy is executed, and the number of people to be dispatched is increased by 1 each time, until the current congestion coefficient is on a downward trend compared to the previous congestion coefficient; The process of generating the recommended value of the number of dispatched personnel after building the basic calculation model is as follows: Preset variables and parameters: The number of intersections in the road intersection is recorded as N, the number of basic dispatched personnel is recorded as , which is a basic number of dispatches, and the error correction factor is denoted by K; Formula establishment: Based on the normalized variables and parameters, the recommended value of the number of dispatched personnel is obtained, and the formula is as follows: ； In the formula, D represents the recommended value of the number of dispatched personnel.

Images