Intelligent security system for industrial park

Abstract

An intelligent security and protection system for an industrial park comprises a video monitoring system for monitoring the internal environment of the industrial park in real time; the cargo transportation and prevention and control management system monitors the operation parameters of the trucks in the park and the parameters of the cargos on the trucks in real time, performs prevention and control management when the operation parameters of the trucks or the parameters of the cargos on the trucks are abnormal, and performs scheduling management on the trucks; the personnel behavior intelligent analysis system is used for performing behavior analysis on personnel in the park; the alarm system is used for giving an alarm when the personnel behavior intelligent analysis system analyzes that the personnel in the park have abnormal conditions; the intelligent patrol system patrols the park regularly according to a specified route; the access control system controls access control of the entrance and exit of the park; and the intelligent security and protection system is used for performing security and protection alarm and emergency command in case of fire emergency. The security and protection system integrates freight monitoring, personnel monitoring and environment monitoring, and the security and protection system of the industrial park can be intelligently enhanced.

Description

Technical Field

[0001] The present invention relates to the field of security protection, and in particular to an intelligent security protection system for an industrial park. Background Art

[0002] With the rapid development of the global economy and the increasing demand for resources, the construction of smart parks has become an important development goal of many countries and regions. It is to connect various devices and systems with each other through the application of information technology and Internet of Things technology, and organically combine industrial agglomeration development with different spaces for urban life and residence, to form a functional complex urban space area with community value association, circle resource sharing, and full-time land utilization. The operation of logistics within the park is a key link in the operation of the park, which is of great significance for improving transportation efficiency, reducing costs, and optimizing resource allocation.

[0003] However, the complexity and large scale of logistics business today make traditional monitoring and management methods unable to meet the needs, which is specifically reflected in the following aspects: First, in terms of logistics transportation route recommendation, traditional route recommendation is usually based on experience and rules, while logistics activities within the park are often subject to various restrictions and special requirements. Traditional route recommendation is difficult to take these factors into account and may not provide the best route selection. Second, when transporting goods within the park, the goods are prone to cross-border overturning and causing the goods to fall, which can easily cause cargo loss and casualties. Therefore, for trucks, how to achieve real-time perception of the environment during operation, accurately detect the crossing of goods, avoid the risk of cargo overturning, and ensure operational efficiency and safety is extremely important. Existing cargo overturning detection methods and devices generally have problems such as complex structure, high cost and unsatisfactory detection effect. Third, most existing industrial parks use security personnel to patrol to prevent illegal intrusion, fire hazards and other problems. However, ground prevention and control through manual patrols is not only inefficient, but also consumes a lot of manpower and material resources. Summary of the invention

[0004] In order to solve the above problems existing in the prior art, the present invention proposes the following technical solutions:

[0005] The present invention provides an intelligent security system for an industrial park, comprising: a video monitoring system for real-time monitoring of the internal environment of the industrial park; a cargo transportation and prevention and control management system for real-time monitoring of the operating parameters of trucks in the park and the parameters of the cargo on the trucks, and for performing prevention and control management when the operating parameters of the trucks or the parameters of the cargo on the trucks are abnormal, as well as for dispatching management of the trucks; a personnel behavior intelligent analysis system for performing behavior analysis on personnel in the park; an alarm system for performing an alarm when the personnel behavior intelligent analysis system analyzes that an abnormal situation occurs among personnel in the park; an intelligent patrol system for regularly patrolling the park along a designated route; an access control system for controlling the access control at the entrances and exits of the park; and an intelligent security system for performing security alarms and emergency commands in the event of a fire emergency.

[0006] Preferably, the video surveillance system includes a first camera, a display and a memory; a plurality of the first cameras are arranged in an industrial park to perform all-round monitoring of the park, and the display displays the monitoring screen of the first camera in real time; and the memory stores the monitoring screen.

[0007] Preferably, the cargo transportation and prevention and control management system includes: a cargo monitoring module, which uses on-board sensors to obtain the driving data of the truck, uses GPS to obtain the location of the truck and tracks the driving trajectory of the truck in real time, and uses an on-board camera installed on the truck to obtain the status information of the cargo on the truck; an abnormality detection module, which processes and analyzes the driving data of the truck obtained by the on-board sensors and the images obtained by the on-board camera, determines whether the driving data and the status information of the cargo on the truck are abnormal, and issues an alarm to the driver when an abnormality occurs; a route planning and scheduling module, which plans the optimal transportation route for the truck in the park according to the transportation demand of the cargo, road condition information, and the busyness of the loading and unloading locations.

[0008] Preferably, a visual algorithm is used to analyze the status information of the cargo on the truck, wherein the visual algorithm includes data acquisition and preprocessing, target semantic segmentation, cross-border detection, overturning detection and alarm and recording; the data acquisition and preprocessing, through the RGBD cameras installed on the four positions of the truck, collects the color and depth information of the cargo boxes to be loaded and in the target area, generates an RGB image and a depth map, and converts the depth map into a point cloud according to the calibrated internal and external parameters of the camera, and performs data enhancement on the RGB image and the point cloud; the target semantic segmentation, inputs the enhanced RGB image and point cloud data into an improved version of the 3D target semantic segmentation network, obtains the point cloud bounding box of the cargo box, and uses the point cloud denoising algorithm to segment the cargo point cloud from the environment; the cross-border detection, detects whether the cargo exceeds the boundary of the truck; the overturning detection, detects whether the cargo on the truck is tilted or overturned; the alarm and recording, if it is judged that the cargo is tilted or overturned, the system sends an alarm to the main control through the CAN communication module, and records relevant data for further analysis and processing.

[0009] Preferably, the planning of the optimal transportation route for trucks in the industrial park based on the transportation demand of the goods, road condition information, and the busyness of the loading and unloading locations includes: obtaining industrial park map data; establishing a topological road network based on the industrial park map data to simulate the map environment, and setting the starting and ending points and loading and unloading points of the industrial park; establishing an industrial park truck path planning model based on the topological road network, constructing an improved ant colony algorithm based on the industrial park truck path planning model, and completing truck path planning in the topological road network based on the improved ant colony algorithm.

[0010] Preferably, the intelligent personnel behavior analysis system includes: a face recognition module, which performs face recognition on personnel in the park and determines whether the person is an intruder; a behavior monitoring module, which monitors the behavior of personnel in the park and determines whether the person has committed any illegal behavior; and a group behavior detection module, which monitors the density and flow speed of personnel and determines whether abnormal gathering occurs.

[0011] Preferably, the facial recognition of persons in the park and judging whether the persons are intruders include: loading a local face library and feature vectors; normalizing the face image to an angle that conforms to the feature extraction of the face image; marking the feature parts of the face used for facial recognition, accurately calibrating the position and size of the face in the image, and screening key features, and using the features as a benchmark to extract and preprocess the features in a targeted manner; using geometric correction, median filtering, sharpening, and standardized face images with the same grayscale value range; calculating the Euclidean distance between the key features and the feature vectors in the local face library, and performing a threshold judgment; when all Euclidean distances are greater than the threshold, the human body is judged to be an illegal intruder.

[0012] Preferably, the monitoring of the behavior of the personnel in the park and determining whether the personnel have violated the rules includes: converting the video stream acquired by the first camera into a video frame sequence, analyzing and detecting the video frame sequence images according to the target detection analysis training model, analyzing and detecting the coordinate lists of three target categories: human figure, safety helmet, and reflective vest, and the coordinate lists are recorded in the following format;

[0013] {(x1,y1,w1,h1),(x2,y2,w2,h2),......,(xn,yn,wn,hn)}, where x and y are the center coordinates of the detection target, and w and h are the width and height of the detection target; the target detection analysis training model detects whether the operator wears the safety helmet and reflective vest correctly based on the coordinate list of three target categories: human figure, safety helmet and reflective vest; the steps of detecting whether the operator wears the safety helmet correctly are as follows:

[0014] A1), according to the target detection alarm module, analyze the video to obtain the human figure coordinate list I human figure and the safety helmet coordinate list I safety helmet; A2), take a human figure coordinate (xj, yj, wj, hj), traverse the safety helmet coordinate list (xi, yi, wi, hi), and judge whether the worker corresponding to the human figure coordinate is wearing the safety helmet correctly. The judgment standard is:

[0015] If the above conditions are met at the same time, it is determined that the worker is wearing the safety helmet correctly, otherwise, it is considered that the worker is not wearing the safety helmet correctly;

[0016] The steps to detect whether the operator is wearing a reflective vest correctly are:

[0017] B1), analyzing the video according to the target detection alarm module to obtain a human coordinate list I human and a reflective vest coordinate list I reflective vest;

[0018] B2) Take a human figure coordinate (xj, yj, wj, hj), traverse the safety helmet coordinate list (xi, yi, wi, hi), and determine whether the worker corresponding to the human figure coordinate is wearing a reflective vest correctly. The judgment standard is:

[0019]

[0020] If the above conditions are met at the same time, it is determined that the worker is wearing the reflective vest correctly. Otherwise, it is considered that the worker is not wearing the reflective vest correctly.

[0021] Preferably, it is characterized in that the intelligent patrol system includes: a drone, which patrols the park regularly along a designated route, and when the personnel behavior intelligent analysis system detects abnormal personnel behavior, the drone warns and expels the personnel; a storage module, which is used to store initial images of fire protection, water pipes, and infrastructure in the park. During the patrol process, the drone obtains image information of fire protection, water pipes, and infrastructure in the park and compares it with the initial image to check for fire hazards in the park.

[0022] Preferably, the alarm system includes a broadcast alarm and a signal transmission module. When abnormal behavior of personnel or fire protection occurs in the park, a broadcast alarm is issued through the broadcast alarm and security personnel are notified through the signal transmission module.

[0023] Description of the drawings

[0024] Figure 1 is a schematic diagram of an industrial park intelligent security system provided by an embodiment of the present invention;

[0025] Figure 2 It is a flowchart of a visual algorithm analysis method provided by an embodiment of the present invention;

[0026] Figure 3 It is a schematic diagram of the face recognition process provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0027] See also Figure 1 The present invention provides an industrial park intelligent security system 100, including a video monitoring system 101, which monitors the internal environment of the industrial park in real time; the video monitoring system 101 includes a first camera, a display and a memory; multiple first cameras are arranged in the industrial park to monitor the park in all directions without blind spots, and the display is set in the security room, which is used to display the monitoring screen of the first camera in real time so that the security personnel can observe the situation in the park in real time; the memory is set in the computer room of the security room, which is used to store the monitoring screen so as to retrieve the corresponding monitoring screen when an emergency occurs.

[0028] The intelligent security system 100 of the industrial park also includes a cargo transportation and prevention and control management system 102, which monitors the operating parameters of trucks in the park and the parameters of the cargo on the trucks in real time, and performs prevention and control management when the operating parameters of the trucks or the parameters of the cargo on the trucks are abnormal, and performs dispatch management on the trucks. The cargo transportation and prevention and control management system 102 includes: a cargo monitoring module, which uses on-board sensors to obtain the driving data of the trucks, uses GPS to obtain the location of the trucks and tracks the driving trajectory of the trucks in real time, and uses on-board cameras set on the trucks to obtain the status information of the cargo on the trucks; an abnormality detection module, which processes and analyzes the driving data of the trucks obtained by the on-board sensors and the images obtained by the on-board cameras, determines whether the driving data and the status information of the cargo on the trucks are abnormal, and issues an alarm to the driver when an abnormality occurs; a route planning and scheduling module, which plans the optimal transportation route for the trucks in the park according to the transportation demand of the cargo, road condition information, and the busyness of the loading and unloading locations.

[0029] like Figure 2 The cargo transportation and prevention and control management system 102 uses a visual algorithm to analyze the status information of the cargo on the truck, wherein the visual algorithm includes five steps: data acquisition and preprocessing, target semantic segmentation, cross-boundary detection, overturning detection, and alarm and recording.

[0030] S1: Data acquisition and preprocessing. The RGBD cameras installed on the four sides of the truck collect the color and depth information of the cargo boxes to be loaded and in the target area, generate RGB images and depth maps, and convert the depth maps into point clouds according to the calibrated camera internal and external parameters. The RGBD cameras can simultaneously obtain color and depth information, providing a more comprehensive object perception capability. By collecting color information, the appearance features of the object, such as color and texture, can be obtained; by collecting depth information, the spatial structure and size information of the object can be obtained. Compared with traditional RGB cameras, RGBD cameras can To provide three-dimensional information of objects, the distance and depth distribution of objects can be obtained through depth images, thereby realizing three-dimensional perspective observation and analysis of objects, which helps to understand the shape and position of objects more accurately. According to the calibrated internal and external parameters of the camera, the collected depth images can be accurately calibrated, which can eliminate distortion and improve image quality and accuracy. At the same time, the depth image is converted into point cloud data, and the two-dimensional image information is converted into three-dimensional point cloud space, which is helpful for subsequent point cloud processing and analysis. Data enhancement of the collected RGB images and point clouds can increase the diversity and richness of data through image processing and point cloud processing technologies.

[0031] S2: Target semantic segmentation: input the enhanced RGB image and point cloud data into the improved 3D target semantic segmentation network to obtain the point cloud bounding box of the cargo box. Using the point cloud denoising algorithm, the cargo point cloud is segmented from the environment. Through the target semantic segmentation network, the RGB image and point cloud can be comprehensively analyzed and processed to achieve accurate identification and segmentation of the cargo box. This helps to distinguish the area between the cargo box and the surrounding environment, accurately extract the target object, and thus achieve effective detection and analysis of the cargo. Using point cloud data for target semantic segmentation can provide more accurate three-dimensional information. Point cloud data contains not only color information, but also information such as the spatial position and geometric shape of the object. Through the three-dimensional features of the point cloud, the cargo point cloud and the environmental point cloud can be better segmented, reducing misjudgment and interference, and improving the accuracy of the segmentation results.

[0032] S3: Cross-border detection, when the goods exceed the boundaries of the truck, there may be a risk of falling, collision or accidental injury. Cross-border detection can detect this situation in time and warn the operator or relevant personnel so that appropriate measures can be taken, such as rearranging the goods or adjusting the loading method to ensure the stability of the goods and the safety of the workplace. Through cross-border detection, goods that exceed the boundaries of the truck can be detected in time, so as to adjust the loading method, increase support or re-plan the storage layout, which helps to optimize the loading and storage methods of goods, maximize the use of space, and improve logistics efficiency. When the goods exceed the boundaries of the truck, they may be at risk of being squeezed, damaged or broken. Cross-border detection can help identify cross-border goods in time and notify the operator through the alarm system so that appropriate measures can be taken to protect the goods from further damage.

[0033] S4: Overturning detection. Overturning refers to the situation where goods tilt or fall during transportation or storage. Overturning may cause the goods to fall, be damaged, collide or cause accidental injury. Through overturning detection, the tilting of goods can be discovered in time, and measures can be taken in advance to prevent further overturning, thereby ensuring the safety of the workplace. The tilting or overturning of goods may cause damage or destruction of the goods. Overturning detection can identify tilted goods in time and notify the operator or relevant personnel through the alarm system to take measures, such as rearranging the goods or adjusting the support structure, to protect the goods from further damage.

[0034] S5: Alarm and record. If the cargo is judged to be tilted or overturned, the system will send an alarm to the main control through the CAN communication module and record the relevant data for further analysis and processing. The alarm system can promptly notify the operator or relevant personnel of the tilt or overturn of the cargo, and take appropriate measures immediately, such as stopping the operation, rearranging the cargo, or adjusting the loading method, to avoid further danger and loss. Immediate warning helps to ensure personnel safety and cargo integrity, and recording relevant data can provide valuable basis for further analysis and processing. The recorded data can include the time, location, tilt angle, cargo information, etc. of the tilt or overturning event. By analyzing these data, the frequency, cause and trend of the tilt or overturning event can be understood, so as to take targeted improvement measures to improve work efficiency and safety, and the recording of relevant data can also be used for troubleshooting and equipment maintenance. If the system frequently alarms or records abnormal data, the root cause of the problem can be traced and repairs or adjustments can be made in time. This helps to maintain the normal operation of the equipment and reduce downtime and maintenance costs.

[0035] Further, data acquisition and preprocessing, this module performs Gaussian smoothing, bilateral filtering, image cropping and scaling on the image, which greatly improves the processing speed of subsequent steps. Through techniques such as Gaussian smoothing and bilateral filtering, the noise and details in the image can be reduced, making the image smoother. This helps to reduce interference and misjudgment in subsequent processing steps and improve the robustness and accuracy of the algorithm. Image cropping and scaling can be based on actual needs. Cropping and scaling the image can remove unnecessary background information and focus on the area of ​​interest, which can reduce the amount of processing and time, and improve the processing speed and efficiency of subsequent steps. Data standardization uses image processing technology to crop and scale the image, so that different images have consistent sizes and proportions. This helps to unify the representation of data and facilitate the training and comparison of subsequent algorithms. At the same time, standardized data is also conducive to the generalization ability and portability of the algorithm. Accelerating the processing speed Preprocessing the image can optimize and prepare the image in advance, making the subsequent processing steps more efficient. Reducing noise, smoothing the image, and cropping and scaling operations can reduce the amount of processing and complexity, thereby accelerating the processing speed of the entire algorithm.

[0036] Furthermore, for target semantic segmentation, this module makes lightweight adjustments to the DeepLabv3+ semantic segmentation network, and uses self-collected images to perform fine-tuning training on the open source model, with good segmentation effect, strong generalization, and accurate segmentation effect. By lightweight adjustment and fine-tuning training of the DeepLabv3+ semantic segmentation network, the network can be more adapted to the target cargo box segmentation task in specific application scenarios, which can improve the accuracy of the segmentation algorithm, reduce misjudgments and missed judgments, and ensure accurate segmentation of target cargo. Strong generalization performance can be improved by fine-tuning training with self-collected images. Generalization performance refers to the algorithm's ability to adapt to images in different scenes and under different lighting conditions. Through fine-tuning training, the algorithm can better adapt to various changes in actual applications, so that the segmentation results can show good results in different situations.

[0037] Furthermore, the overturning detection module uses the target cargo information obtained by target semantic segmentation to accurately calculate the inclination and displacement of the target cargo, thereby making an accurate judgment on the overturning situation. By using the target cargo information obtained by target semantic segmentation, the overturning detection module can analyze each target cargo point cloud and calculate its inclination and displacement. This can achieve accurate judgment of the overturning situation, instantly find out whether the cargo is tilted or overturned, and provide timely alarm and early warning functions.

[0038] The cargo transportation and prevention and control management system 102 plans the optimal transportation route for trucks in the industrial park based on the ACO-SA algorithm, including: obtaining industrial park map data; establishing a topological road network based on the industrial park map data to simulate the map environment, and setting the starting and ending points and loading and unloading points of the industrial park; establishing an industrial park truck path planning model based on the topological road network, constructing an improved ant colony algorithm based on the industrial park truck path planning model, and completing truck path planning in the topological road network based on the improved ant colony algorithm.

[0039] The specific steps include:

[0040] Step 1: Establish a topological road network to simulate the map environment, set the work shed points and the starting and ending points of the park, and initialize the various parameters of the improved algorithm; the values ​​on the lines connecting the points in the topological road network are the cost values ​​of the path, the lines represent the road sections, and the points represent the changes in the road sections, such as lane changes, traffic lights, road grade changes, and intersections.

[0041] Step 2: Establish a road transport truck path planning model and construct a truck trip cost function;

[0042]

[0043] p∈P,q∈Q,γi≥0 and θj≥0 and Among them, P and Q represent the set of industrial park road paths respectively; γi represents the weight of the influencing factors of the path selection of the industrial park roads; θj represents the weight of the influencing factors of the path selection of the roads; represents the value of the i-th influencing factor of the p-th section of industrial park road; represents the value of the jth influencing factor of the qth section of road; xp represents the path length of the pth section of industrial park road; xq represents the path length of the qth section of road.

[0044] The factors affecting the path selection of industrial park roads comprehensively consider transportation time, transportation cost and transportation convenience. The selected influencing factors are road section length, road section traffic volume, number of lanes and traffic jam incidence.

[0045] Step 3: Place all ants at the starting point and build a taboo table

[0046] Step 4: Introduce the Metropolis criterion and the truck travel cost function into the state probability transfer formula of the ant colony algorithm, and stipulate a new initial pheromone distribution rule to construct an improved ACO-SA algorithm. Based on the ACO-SA algorithm, complete the truck path planning in the topological road network.

[0047] Step 4.1: Connect the starting and ending points of the park and the work shed point, and draw a circle with the connecting straight line as the diameter and the end point of the straight line as the center. Figure 2 As shown, within the circle, pheromones are distributed according to the distance from the connected straight line. The closer to the straight line, the more initial pheromones are distributed. However, the maximum and minimum values ​​of pheromones shall not exceed 0.1 to avoid affecting the randomness of ant colony search.

[0048] Step 4.2: Deploy a group of virtual "ants" in the problem space, the number of ants is m, and initialize the pheromone concentration τij(t) = [a-0.1, a], Δτij(0) = 0;

[0049] Step 4.3: Calculate the probability of each road based on the state probability transfer formula of the ant colony algorithm and introduce the Metropolis criterion and the truck travel cost function; determine the next travel node of the ant based on the transfer probability. After improvement, the ant colony will not only choose the road with the highest probability, but will choose other paths in a certain proportion and put the travel nodes passed into the taboo table; until the ant reaches the end point, the current search is completed, and the total travel cost function value of each ant in the current iteration is obtained;

[0050]

[0051] Among them, pij(t) represents the probability of the ant colony moving from node i to node j at time t; τij(t) represents the pheromone concentration from node i to node j; α, β represent the pheromone factor and heuristic factor; allowedk represents the set of nodes not included in the taboo table; P is the probability of the ant colony going to each path;

[0052] Step 4.4: After the ant colony completes a path search, it will leave pheromones on the path. Due to the volatilization of pheromones and the different number of ants passing through each path, the pheromone concentration on each path is also different, and the pheromone needs to be updated;

[0053]

[0054] Where ρ represents the pheromone volatility coefficient.

[0055] Step 4.5: Record the route of the ant colony after completing a movement, update the pheromone, and clear the taboo table for the next path search;

[0056] Step 4.6: Determine whether the maximum number of iterations has been reached or a satisfactory solution has been found. If the conditions are not met, clear the taboo table and repeat steps 3.3 and 3.5 until the termination conditions are met.

[0057] Step 4.7: After the termination condition is met, output the optimal path.

[0058] The industrial park intelligent security system 100 also includes a personnel behavior intelligent analysis system 103, which is used to perform behavioral analysis on personnel in the park. It includes a face recognition module, which performs facial recognition on personnel in the park and determines whether the personnel is an intruder; a behavior monitoring module, which monitors the behavior of personnel in the park and determines whether the personnel has violated regulations; and a group behavior detection module, which monitors the density and flow speed of personnel and determines whether abnormal gathering occurs.

[0059] For further information, see Figure 3 The face recognition module of this embodiment first establishes a local face library before performing face recognition, collects recent face images of staff members in advance to establish the local face library, extracts features of the face images in the local face library, and then stores the local face library and the face features in the face library.

[0060] In the embodiment of the present invention, the face recognition process mainly includes six parts: loading database, face correction, feature characterization, data preprocessing, feature extraction, and similarity comparison.

[0061] Specifically:

[0062] Load database – load local face library labels and feature vectors into the face recognition model;

[0063] Face normalization: Input the captured face image into the model. The model first detects the face in the image, normalizes the face, and performs 3D rotation on the side face. The face image is normalized to an angle that meets the feature extraction of the face image.

[0064] Feature characterization: After face correction, the model characterizes the features on the face; the model marks the characteristic parts of the face used for face recognition, accurately calibrates the position and size of the face in the image, selects key features, and uses the features as a benchmark to extract and preprocess the features in a targeted manner;

[0065] Data preprocessing: Since the human face has quite complex details and the environment will have a great impact on face recognition, the light compensation, grayscale transformation, histogram equalization, and normalization of the face image are performed to obtain the same size, and geometric correction, median filtering, sharpening, and standardized face images with the same grayscale value range are used;

[0066] Feature extraction - using convolutional neural networks to learn face images into 128-dimensional feature vectors;

[0067] Similarity comparison - calculate the Euclidean distance between key features and feature vectors in the local face database, and make threshold decisions.

[0068] See also Figure 3 , compare the 128-dimensional feature vector of the face with the feature vector in the local database, calculate the Euclidean distance, and make a threshold judgment. When the Euclidean distance is less than the threshold, it is judged that the person is not an illegal intruder; if all Euclidean distances are greater than the threshold, the person is judged as an illegal intruder.

[0069] Further, the monitoring of the behavior of the personnel in the park and determining whether the personnel have violated the rules includes: converting the video stream acquired by the first camera into a video frame sequence, analyzing and detecting the video frame sequence images according to the target detection analysis training model, analyzing and detecting the coordinate lists of three target categories of human figures, safety helmets, and reflective vests, and recording the coordinate lists in the following format;

[0070] {(x1,y1,w1,h1),(x2,y2,w2,h2),......,(xn,yn,wn,hn)}, where x and y are the center coordinates of the detection target, and w and h are the width and height of the detection target; the target detection analysis training model detects whether the operator wears the safety helmet and reflective vest correctly based on the coordinate list of three target categories: human figure, safety helmet and reflective vest; the steps of detecting whether the operator wears the safety helmet correctly are as follows:

[0071] A1), according to the target detection alarm module, analyze the video to obtain the human figure coordinate list I human figure and the safety helmet coordinate list I safety helmet; A2), take a human figure coordinate (xj, yj, wj, hj), traverse the safety helmet coordinate list (xi, yi, wi, hi), and judge whether the worker corresponding to the human figure coordinate is wearing the safety helmet correctly. The judgment standard is:

[0072] If the above conditions are met at the same time, it is determined that the worker is wearing the safety helmet correctly, otherwise, it is considered that the worker is not wearing the safety helmet correctly;

[0073] The steps to detect whether the operator is wearing a reflective vest correctly are:

[0074] B1), analyzing the video according to the target detection alarm module to obtain a human coordinate list I human and a reflective vest coordinate list I reflective vest;

[0075] B2) Take a human figure coordinate (xj, yj, wj, hj), traverse the safety helmet coordinate list (xi, yi, wi, hi), and determine whether the worker corresponding to the human figure coordinate is wearing a reflective vest correctly. The judgment standard is:

[0076]

[0077] If the above conditions are met at the same time, it is determined that the worker is wearing the reflective vest correctly. Otherwise, it is considered that the worker is not wearing the reflective vest correctly.

[0078] In this embodiment, the target detection analysis training model is obtained by the following method:

[0079] C1) Collect sample sources and obtain monitoring video sources under various light and weather conditions in construction scenes at different time periods; C2) Convert video sources into video frame images

[0080] C3) Use the image labeling tool to input the defined categories: human figure, helmet, reflective vest, label the corresponding targets on the video frame image, and obtain the training sample set through this method;

[0081] C4) Based on the deep learning target detection algorithm training sample set, a target detection analysis training model is obtained;

[0082] C5) Use forward and reverse test cases to verify the target detection analysis training model and obtain the accuracy, false alarm rate, and missed alarm rate;

[0083] C6) Analyze the accuracy, false alarm rate, and missed alarm rate of the target detection analysis training model, and repeat steps C1 to C5 by increasing the number and diversity of samples until the accuracy, false alarm rate, and missed alarm rate of the target detection analysis training model meet the requirements.

[0084] Furthermore, the group behavior detection module obtains a real-time video stream through a first camera, parses the real-time video stream, obtains a picture to be analyzed, and the picture to be analyzed includes an intra-frame coded frame and a predictive coded frame, obtains the optical flow component value of the picture to be analyzed; if the optical flow component value of the picture to be analyzed is greater than a first preset threshold, the picture to be analyzed is selected as the detection picture. Human image samples and non-human image samples are collected, classified, labeled and trained to obtain a human feature model; k different rectangular frames are used to nominate on the convolution feature layer of the detection picture to screen out candidate areas where human bodies may exist, where k is a positive integer; human body features are extracted from the candidate areas respectively; the human body feature model is used to classify and regress the features to identify whether the detection picture has a human body; if it is identified that the detection picture has a human body, the human body is subjected to a border position regression process to obtain a human body frame. It is determined whether the number of human bodies in the detection picture is greater than a preset threshold. When the number of human bodies in the detection picture is greater than the preset threshold, it is determined that there is a possibility of abnormal gathering of people in the area.

[0085] Furthermore, the industrial park intelligent security system 100 also includes an alarm system 104. When the personnel behavior intelligent analysis system 103 determines that someone has illegally entered the park, employees have not correctly worn safety helmets and reflective vests, or there is an abnormal gathering of employees in the park, an alarm is broadcast and the corresponding security personnel are notified to arrive at the scene.

[0086] Furthermore, the industrial park intelligent security system 100 also includes an intelligent patrol system 105, which includes: a drone, which patrols the park regularly along a designated route, and when the personnel behavior intelligent analysis system detects abnormal personnel behavior, the drone warns and expel the personnel; a storage module, which is used to store initial images of fire protection, water pipes, and infrastructure in the park. During the patrol process, the drone obtains image information of fire protection, water pipes, and infrastructure in the park and compares it with the initial image to check for fire hazards in the park.

[0087] Furthermore, the industrial park intelligent security system 100 also includes an access control system 106, which has a face recognition function, can quickly identify people with the required identity to enter and exit the park, and can register the identities of park visitors to facilitate personnel management.

[0088] Furthermore, the industrial park intelligent security system 100 also includes an intelligent security system 107, which is connected to the fire department in communication to perform security alarms and emergency commands when a fire emergency occurs.

[0089] Finally, it should be noted that the above embodiments are only used to illustrate the technical solution of the present invention rather than to limit it. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solution of the present invention can be modified or replaced by equivalents without departing from the spirit and scope of the technical solution of the present invention.

Claims

1. An intelligent security system for industrial parks, characterized in that: include: Video surveillance system to monitor the internal environment of the industrial park in real time; The cargo transportation and prevention and control management system monitors the operating parameters of trucks and cargo on trucks in the park in real time, and conducts prevention and control management when the operating parameters of trucks or cargo on trucks are abnormal, as well as dispatching and managing trucks; The personnel behavior intelligent analysis system analyzes the behavior of personnel in the park; the alarm system issues an alarm when the personnel behavior intelligent analysis system analyzes that there is an abnormal situation among the personnel in the park; the intelligent patrol system patrols the park regularly along the designated routes; the access control system controls the access control of the park entrances and exits; the intelligent security system issues security alarms and emergency commands in the event of a fire emergency.

2. The industrial park intelligent security system according to claim 1 is characterized in that: The video monitoring system includes a first camera, a display and a memory; a plurality of the first cameras are arranged in the industrial park to perform all-round monitoring of the park, and the display displays the monitoring images of the first cameras in real time; The memory stores the monitoring screen.

3. The intelligent security system for industrial parks according to claim 2 is characterized in that: The cargo transportation and prevention and control management system includes: a cargo monitoring module, which uses on-board sensors to obtain the driving data of the truck, uses GPS to obtain the location of the truck and tracks the driving trajectory of the truck in real time, and uses an on-board camera installed on the truck to obtain the status information of the cargo on the truck; an anomaly detection module, which processes and analyzes the driving data of the truck obtained by the on-board sensors and the images obtained by the on-board cameras, determines whether the driving data and the status information of the cargo on the truck are abnormal, and issues an alarm to the driver when an abnormality occurs; a route planning and scheduling module, which plans the optimal transportation route for the truck in the park according to the transportation demand of the cargo, road condition information, and the busyness of the loading and unloading locations.

4. The industrial park intelligent security system according to claim 3 is characterized in that: The state information of the cargo on the truck is analyzed by using a visual algorithm, wherein the visual algorithm includes data acquisition and preprocessing, target semantic segmentation, cross-border detection, overturning detection, and alarm and recording; the data acquisition and preprocessing collects the color and depth information of the cargo boxes to be loaded and in the target area through RGBD cameras installed at four positions of the truck, generates an RGB image and a depth map, and converts the depth map into a point cloud according to the calibrated camera internal and external parameters, and performs data enhancement on the RGB image and the point cloud; The target semantic segmentation inputs the enhanced RGB image and point cloud data into the improved version of the 3D target semantic segmentation network to obtain the point cloud bounding box of the cargo box, and uses the point cloud denoising algorithm to segment the cargo point cloud from the environment; the cross-boundary detection detects whether the cargo exceeds the boundary of the truck; the overturning detection detects whether the cargo on the truck is tilted or overturned; the alarm and recording, if it is judged that the cargo is tilted or overturned, the system will send an alarm to the main control through the CAN communication module and record relevant data for further analysis and processing.

5. The industrial park intelligent security system according to claim 3 is characterized in that: The method of planning the optimal transportation route for trucks in the industrial park based on the transportation demand, road condition information, and the busyness of the loading and unloading locations includes: obtaining industrial park map data; establishing a topological road network based on the industrial park map data to simulate the map environment, and setting the starting and ending points and loading and unloading points of the industrial park; establishing an industrial park truck path planning model based on the topological road network, constructing an improved ant colony algorithm based on the industrial park truck path planning model, and completing truck path planning in the topological road network based on the improved ant colony algorithm.

6. The industrial park intelligent security system according to claim 2, characterized in that: The personnel behavior intelligent analysis system includes: a face recognition module, which performs face recognition on personnel in the park and determines whether the personnel are intruders; a behavior monitoring module, which monitors the behavior of personnel in the park and determines whether the personnel have violated the rules; and a group behavior detection module, which monitors the density and flow speed of personnel and determines whether abnormal gathering occurs.

7. The industrial park intelligent security system according to claim 6, characterized in that: The method of performing face recognition on people in the park and judging whether the people are intruders includes: loading a local face library and feature vectors; normalizing the face image to an angle that meets the feature extraction of the face image; marking the feature parts of the face used for face recognition, accurately calibrating the position and size of the face in the image, and screening key features, and using the features as a benchmark to extract and preprocess the features in a targeted manner; using geometric correction, median filtering, sharpening and standardized face images with the same grayscale value range; calculating the Euclidean distance between the key features and the feature vectors in the local face library, and performing threshold judgment; when all Euclidean distances are greater than the threshold, the human body is judged to be an illegal intruder.

8. The industrial park intelligent security system according to claim 7, characterized in that: The behavior monitoring of personnel in the park and determining whether the personnel have violated the rules include: converting the video stream obtained by the first camera into a video frame sequence, analyzing and detecting the video frame sequence images according to the target detection analysis training model, analyzing and detecting the coordinate lists of three target categories: human figure, safety helmet, and reflective vest, and the coordinate lists are recorded in the following format; {(x1, y1, w1, h1), (x2, y2, w2, h2), ..., (xn, yn, wn, hn)}, wherein x and y are the center coordinates of the detection target, and w and h are the width and height of the detection target; the target detection analysis training model detects whether the operator wears the safety helmet and reflective vest correctly based on the coordinate lists of the three target categories: human figure, safety helmet, and reflective vest; wherein the step of detecting whether the operator wears the safety helmet correctly is: A1), according to the target detection alarm module, analyze the video to obtain the human figure coordinate list I human figure and the safety helmet coordinate list I safety helmet; A2), take a human figure coordinate (xj, yj, wj, hj), traverse the safety helmet coordinate list (xi, yi, wi, hi), and judge whether the worker corresponding to the human figure coordinate is wearing the safety helmet correctly. The judgment standard is: If the above conditions are met at the same time, it is determined that the worker is wearing the helmet correctly, otherwise, it is considered that the worker is not wearing the helmet correctly; The steps to detect whether the operator is wearing a reflective vest correctly are: B1), analyzing the video according to the target detection alarm module to obtain a human coordinate list I human and a reflective vest coordinate list I reflective vest; B2) Take a human figure coordinate (xj, yj, wj, hj), traverse the safety helmet coordinate list (xi, yi, wi, hi), and determine whether the worker corresponding to the human figure coordinate is wearing a reflective vest correctly. The judgment standard is: If the above conditions are met at the same time, it is determined that the worker is wearing the reflective vest correctly. Otherwise, it is considered that the worker is not wearing the reflective vest correctly.

9. The industrial park intelligent security system according to claim 8, characterized in that: The intelligent patrol system includes: a drone, which patrols the park regularly along a designated route, and when the intelligent personnel behavior analysis system detects abnormal personnel behavior, the drone is used to warn and expel the personnel; a storage module, which is used to store initial images of fire protection, water pipes, and infrastructure in the park. During the patrol process, the drone obtains image information of fire protection, water pipes, and infrastructure in the park and compares it with the initial image to check for fire hazards in the park.

10. An industrial park intelligent security system according to claim 9, characterized in that: The alarm system includes a broadcast alarm and a signal transmission module. When abnormal behavior of personnel or fire protection occurs in the park, a broadcast alarm is issued through the broadcast alarm and security personnel are notified through the signal transmission module.

Images