Abnormal event prevention and control method based on image recognition

Through image recognition technology based on deep learning, a multi-layer convolutional neural network model is built to monitor and analyze abnormal events in real time, solving the problems of slow response speed and high missed rate of existing security systems, and efficient identification and rapid response to abnormal events are achieved.

CN120071219APending Publication Date: 2025-05-30UHV CONVERTER STATION BRANCH OF STATE GRID SHANGHAI ELECTRIC POWER CO +1
View PDF 0 Cites 1 Cited by

Patent Information

Application Number
CN202510154772.9
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2025-02-12
Publication Date
2025-05-30

AI Technical Summary

Technical Problem

The existing security systems have slow response speed, high missed rate and low processing efficiency. Image recognition technology is insufficient in real-time monitoring and early warning of abnormal events.

Method used

Using deep learning-based image recognition method, we collect image data in real time through high-definition cameras, build a multi-layer convolutional neural network model, perform image preprocessing and abnormal event recognition, real-time monitoring and analysis, automatically classify and evaluate the severity of abnormal events, and notify security personnel through mobile applications or alarm systems.

Benefits of technology

Automatic identification and rapid response to potential abnormal events is realized, the monitoring efficiency of abnormal events is greatly improved, and false alarms and missed reports are reduced. The system has the ability to learn and adapt to changes in different environments and scenarios. The rapid early warning mechanism can promptly notify security personnel to reduce potential risks.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN120071219A_ABST
    Figure CN120071219A_ABST
Patent Text Reader

Abstract

The invention relates to the technical field of abnormal event prevention and control, and discloses an abnormal event prevention and control method based on image recognition, and the method comprises the following steps: data collection; image preprocessing: processing the collected image to improve the accuracy of subsequent identification; an abnormal event recognition model is constructed, a multi-layer convolutional neural network model is constructed based on the deep learning technology, and training data comprises a labeling data set of normal behaviors and abnormal behaviors; real-time monitoring and analysis: carrying out image real-time analysis by utilizing the trained model, and setting a threshold value to judge whether an abnormal event exists or not; an early warning and response mechanism; and a feedback and learning mechanism: after the security personnel deal with the event, the result is fed back to the system, and self-learning and optimization of the model are promoted. Through intelligent image recognition and real-time analysis, automatic recognition and quick response to potential abnormal events are realized, the monitoring efficiency of the abnormal events is greatly improved, and various abnormal conditions can be effectively handled.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention relates to the technical field of prevention and control of abnormal events, and specifically to a method for preventing and controlling abnormal events based on image recognition. Background Art

[0003] With the rapid development of intelligent monitoring technology, traditional security systems can no longer meet the increasingly complex security needs. Existing technologies mainly rely on static surveillance cameras and manual monitoring, suffering from problems such as slow response speed, high false alarm rate, and low processing efficiency. Although some image recognition technologies have been applied in fields such as face recognition and license plate recognition, they are still insufficient in real-time monitoring and early warning of abnormal events. Therefore, corresponding technical solutions need to be designed to solve this problem. Summary of the Invention

[0004] (1) Technical Problems to be Solved

[0005] In view of the deficiencies of the prior art, the present invention provides a method for preventing and controlling abnormal events based on image recognition, which solves the technical problems thereof.

[0006] (2) Technical Solutions

[0007] To achieve the above objectives, the present invention is realized through the following technical solutions: A method for preventing and controlling abnormal events based on image recognition, the method steps are as follows:

[0008] S1. Data Acquisition: Use a high-definition camera to collect image data of the monitoring area in real-time and transmit it to the central processing system;

[0009] S2. Image Preprocessing: Process the collected images to improve the accuracy of subsequent recognition;

[0010] S3. Construction of Abnormal Event Recognition Model: Based on deep learning technology, construct a multi-layer convolutional neural network model, and the training data includes a labeled data set of normal behaviors and abnormal behaviors;

[0011] The model is used to identify various abnormal behaviors and update the model parameters in real-time to adapt to the changing environment;

[0012] S4. Real-time Monitoring and Analysis: Use the trained model to perform real-time image analysis, and set a threshold to determine whether there is an abnormal event;

[0013] For the identified abnormal events, the system will automatically classify and evaluate their severity;

[0014] S5. Early Warning and Response Mechanism: Once an abnormal event is identified, the system will notify the security personnel in a timely manner through a mobile application, SMS notification or alarm system;

[0015] Combined with location data, the system can guide security personnel to quickly reach the incident site;

[0016] S6. Feedback and learning mechanism: After security personnel handle an incident, they will feedback the results to the system to promote the self-learning and optimization of the model.

[0017] Preferably, in step S1, the specific method steps of data collection include the following:

[0018] S101. Use the mobile platform of the drone to initially arrange the camera and select the optimal position according to the characteristics and requirements of the monitoring area;

[0019] Adopt lidar or ultrasonic sensors to perform 3D modeling of the environment to ensure that the camera can automatically adjust its position to the best angle for monitoring in a dynamic environment;

[0020] S102. Real-time monitor environmental light, temperature and humidity factors through sensors, and intelligently adjust the exposure, focus and gain settings of the camera to adapt to different environmental conditions;

[0021] Adopt an adaptive algorithm to real-time adjust the frame rate and resolution according to the dynamic changes in the monitoring area to ensure capturing high-definition images at critical moments;

[0022] S103. Adopt cloud computing technology to real-time upload the image data of various sensors to the central processing system and perform preliminary processing using edge computing to reduce latency;

[0023] S104. During the image acquisition process, use the built-in intelligent algorithm to real-time mark the dynamic targets in the picture and identify different types of objects;

[0024] By analyzing the movement trajectory of the target, predict the occurrence of abnormal behaviors and capture relevant image data in advance;

[0025] S105. During the data transmission process, real-time encrypt the collected image data to ensure data security;

[0026] Adopt a compression algorithm to reduce the data transmission volume and perform high-speed transmission through a dedicated network or 5G network to ensure real-time performance.

[0027] Preferably, the adaptive algorithm dynamically adjusts parameters according to real-time data and environmental changes;

[0028] The adaptive filter adopts the LMS algorithm: w(n + 1) = w(n) + μe(n)x(n);

[0029] Where:

[0030] w(n): Filter weight;

[0031] μ: Step size factor, which controls the convergence rate;

[0032] e(n): Error signal;

[0033] x(n): Input signal;

[0034] Adaptive gain control adopts the gain adjustment formula: G(n) = G(n - 1) + α·(D - P(n - 1));

[0035] Where:

[0036] G(n): Current gain;

[0037] D: Target value;

[0038] P(n - 1): Actual value at the previous moment;

[0039] α: Adjustment factor.

[0040] Preferably, the compression algorithm is used to reduce the storage and transmission space of data, and the compression algorithm is the JPEG compression algorithm;

[0041] JPEG compression adopts the discrete cosine transform (DCT) formula:

[0042]

[0043] Where:

[0044] f(x, y): Image pixel value;

[0045] C(u, v): DCT coefficient.

[0046] Preferably, in step S2, the method steps for preprocessing the acquired image include the following:

[0047] S201, Denoising processing: Mean filtering: Smoothing the image by calculating the mean of neighboring pixels to remove high-frequency noise;

[0048] Median filtering: Replacing the central pixel with the median of neighboring pixels to effectively remove salt-and-pepper noise;

[0049] Gaussian filtering: Weighting neighboring pixels using a Gaussian function to reduce random noise in the image;

[0050] S202, Image enhancement: Histogram equalization: Adjusting the contrast of the image so that pixels of different brightness levels are evenly distributed to improve the overall visibility of the image;

[0051] Adaptive histogram equalization: Performing histogram equalization in a local area to avoid distortion caused by over-enhancement;

[0052] Gamma correction: Adjust the brightness and contrast of the image to adapt to different display conditions;

[0053] S203, Image segmentation: Threshold segmentation: Divide the image into foreground and background according to pixel intensity values, using global threshold or adaptive threshold;

[0054] Edge detection: Use algorithms to extract edges in the image;

[0055] Region growing: Start from seed points and merge similar neighboring pixels to form complete regions;

[0056] S204, Feature extraction: HOG features: Extract the shape and orientation features of objects in the image, suitable for human detection;

[0057] SURF features: Extract key points and descriptors of the image for object recognition;

[0058] SIFT features: Extract key points and calculate their descriptors, with scale and rotation invariance, suitable for complex scenes;

[0059] S205, Image scaling and transformation: Image scaling: Resize the image to a specified size to meet the input requirements of the deep learning model;

[0060] Rotation and flipping: Rotate or flip the image to enhance the diversity of the dataset and help the model learn features from different angles;

[0061] S206, Color space conversion: RGB to HSV conversion: Convert the image from the RGB color space to the HSV color space for more intuitive color analysis;

[0062] Grayscale processing: Convert the color image to a grayscale image to reduce computational complexity while retaining important structural information;

[0063] S207, Data augmentation: Random cropping: Randomly crop small pieces from the original image for training to increase data diversity;

[0064] Adding noise: Add a small amount of noise to the original image to help the model better adapt to changes in the real environment;

[0065] Affine transformation: Translate, scale, and rotate the image to enhance the robustness of the model;

[0066] S208, Brightness and contrast adjustment: Brightness adjustment: Improve the visibility of the image by increasing or decreasing the brightness value;

[0067] Contrast adjustment: Increase or decrease the contrast between light and dark parts of the image to make important features more obvious.

[0068] Preferably, in step S3, the deep learning technology includes a multi-layer structure, automatic feature learning, and large-scale data processing;

[0069] Multi-layer structure: It includes an input layer, hidden layers, and an output layer. Each hidden layer maps the input to a higher-dimensional feature space through a non-linear activation function;

[0070] Automatic feature learning: Automatically learns useful feature representations from the original data;

[0071] Large-scale data processing: Effectively uses a large amount of labeled data for training;

[0072] The multi-layer neural network model includes a convolutional layer, an activation function, a pooling layer, a fully connected layer, and an output layer;

[0073] Convolutional layer: Extracts local features of the input image through convolution operations. Each convolutional kernel slides over the image to calculate the weighted sum of the local area and generates a feature map; It contains multiple filters to capture different features in the image;

[0074] Activation function: Applies an activation function after the convolutional layer to introduce non-linearity, enabling the network to learn complex feature representations;

[0075] Pooling layer: Performs downsampling on the feature map;

[0076] Fully connected layer: At the final stage of the network, the feature map is flattened into a one-dimensional vector and input into the fully connected layer for the final classification or regression task;

[0077] Output layer: According to the task of the model, the output layer maps the final features to specific class labels through an activation function.

[0078] Preferably, in step S3, the abnormal behaviors include personal safety behaviors, public order behaviors, suspicious behaviors, emergencies, traffic safety behaviors, natural disaster behaviors, behavior patterns, and emotional states;

[0079] Personal safety behaviors:

[0080] Fighting: Intense physical conflicts between two or more people;

[0081] Falling: An individual loses balance and falls to the ground due to an accident or other reasons, and may require emergency medical assistance;

[0082] Pushing and shoving: Strong physical contact between individuals in a crowd, which may trigger conflicts;

[0083] Public order behaviors:

[0084] Crowding: The crowd is dense and exceeds the safe capacity, which may lead to accidents such as stampedes;

[0085] Fights and brawls: Physical conflicts involving weapons or obvious threatening behavior;

[0086] Damage to public facilities: Acts of deliberately damaging or destroying public property;

[0087] Suspicious behavior:

[0088] Loitering: Staying or walking back and forth in a specific area for a long time, which may indicate potential criminal intent;

[0089] Theft: Intentional contact with others' property, especially when unnoticed;

[0090] Illegal intrusion: The act of entering a prohibited or restricted area;

[0091] Emergency incidents:

[0092] Fire: The appearance of smoke or flames, requiring immediate response;

[0093] Medical emergencies: Such as someone suddenly fainting or having other health problems;

[0094] Traffic safety behaviors:

[0095] Going against the traffic: Driving or walking in the opposite direction in a traffic area, posing a safety hazard;

[0096] Running a red light: Vehicles or pedestrians driving or passing through when the red light is on, which may cause traffic accidents;

[0097] Natural disaster behaviors:

[0098] Escape behaviors: The escape behaviors of the crowd during natural disasters such as earthquakes and fires;

[0099] Abnormal climate reactions: Such as objects flying up or the chaotic reactions of the crowd during strong wind weather;

[0100] Behavior patterns:

[0101] Abnormal behaviors: Individuals showing long-term immobility or abnormal activity patterns may require further investigation;

[0102] Gatherings of multiple people: In non-public activity areas, there are gatherings of multiple people without an obvious purpose;

[0103] Emotional states:

[0104] Extreme emotional expressions: Individuals showing extreme anger, anxiety or panic may lead to violent behaviors.

[0105] (III) Beneficial effects

[0106] Compared with the prior art, the beneficial effects of the present invention are as follows: Through intelligent image recognition and real-time analysis, the automatic recognition and rapid response to potential abnormal events are realized, greatly improving the monitoring efficiency of abnormal events. The application of the deep learning model significantly improves the accuracy of abnormal behavior recognition, reduces false alarms and missed alarms. The system has the ability of self-learning, can adapt to changes in different environments and scenarios, and the rapid warning mechanism can notify security personnel in time, reducing potential risks. The organic combination of image recognition, deep learning and security response mechanism forms an integrated prevention and control system. The introduction of the self-learning mechanism improves the flexibility and adaptability of the system, is applicable to various complex environments such as public places, transportation hubs, and enterprise parks, and can effectively respond to various abnormal situations. BRIEF DESCRIPTION OF THE DRAWINGS

[0107] Figure 1 It is a schematic diagram of the overall method steps of the present invention;

[0108] Figure 2 It is a schematic diagram of the specific method steps of data collection of the present invention;

[0109] Figure 3 It is a schematic diagram of the method steps for preprocessing the collected images of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0110] Next, the technical solutions in the embodiments of the present invention will be clearly and completely described in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts shall fall within the protection scope of the present invention.

[0111] Please refer to Figures 1-3 , the embodiments of the present invention provide a technical solution: An abnormal event prevention and control method based on image recognition, and the method steps include the following:

[0112] S1. Data collection: Use high-definition cameras to collect image data of the monitoring area in real time and transmit it to the central processing system;

[0113] S2. Image preprocessing: Process the collected images to improve the accuracy of subsequent recognition;

[0114] S3. Construction of abnormal event recognition model: Based on deep learning technology, construct a multi-layer convolutional neural network model, and the training data includes a labeled data set of normal behavior and abnormal behavior;

[0115] The model is used to identify various abnormal behaviors and update the model parameters in real time to adapt to the changing environment;

[0116] S4, Real-time Monitoring and Analysis: Use the trained model to perform real-time image analysis, and set thresholds to determine whether abnormal events exist;

[0117] For the identified abnormal events, the system will automatically classify and evaluate their severity;

[0118] S5, Early Warning and Response Mechanism: Once an abnormal event is identified, the system will promptly notify security personnel through mobile applications, text messages, or alarm systems;

[0119] Combined with location data, the system can guide security personnel to quickly reach the scene of the incident;

[0120] S6, Feedback and Learning Mechanism: After security personnel handle the incident, they will feedback the results to the system to promote the self-learning and optimization of the model.

[0121] Further improved, in step S1, the specific method steps of data collection include the following:

[0122] S101, Use the mobile platform of the drone to initially arrange the camera, and select the optimal position according to the characteristics and requirements of the monitoring area;

[0123] Adopt lidar or ultrasonic sensors to perform 3D modeling of the environment to ensure that the camera can automatically adjust its position to the best angle for monitoring in a dynamic environment;

[0124] S102, Real-time monitor environmental light, temperature, and humidity factors through sensors, and intelligently adjust the exposure, focus, and gain settings of the camera to adapt to different environmental conditions;

[0125] Adopt an adaptive algorithm to real-time adjust the frame rate and resolution according to the dynamic changes in the monitoring area to ensure capturing high-definition images at critical moments;

[0126] S103, Adopt cloud computing technology to real-time upload the image data of various sensors to the central processing system, and use edge computing for preliminary processing to reduce latency;

[0127] S104, During the image acquisition process, use the built-in intelligent algorithm to real-time mark the dynamic targets in the picture and identify different types of objects;

[0128] By analyzing the movement trajectories of the targets, anticipate the occurrence of abnormal behaviors and capture relevant image data in advance;

[0129] S105, During the data transmission process, real-time encrypt the collected image data to ensure data security;

[0130] Adopt a compression algorithm to reduce the data transmission volume and perform high-speed transmission through a dedicated network or 5G network to ensure real-time performance.

[0131] Through the above steps, real-time acquisition of high-definition images can be ensured in various complex environments, the overall performance of the monitoring system can be improved, efficient and accurate identification of abnormal events can be achieved, and it has broad application prospects.

[0132] Further improved, the adaptive algorithm dynamically adjusts parameters according to real-time data and environmental changes;

[0133] The adaptive filter adopts the LMS algorithm: w(n + 1) = w(n) + μe(n)x(n);

[0134] Where:

[0135] w(n): filter weight;

[0136] μ: step factor (learning rate), controlling the convergence speed;

[0137] e(n): error signal (the difference between the desired output and the actual output);

[0138] x(n): input signal;

[0139] The adaptive gain control adopts the gain adjustment formula: G(n) = G(n - 1) + α·(D - P(n - 1));

[0140] Where:

[0141] G(n): current gain;

[0142] D: target value (such as the set light level);

[0143] P(n - 1): actual value at the previous moment;

[0144] α: adjustment factor.

[0145] Further improved, the compression algorithm is used to reduce the space for data storage and transmission, and the compression algorithm is the JPEG compression algorithm;

[0146] The JPEG compression adopts the discrete cosine transform (DCT) formula:

[0147]

[0148] Where:

[0149] f(x, y): image pixel value;

[0150] C(u, v): DCT coefficient.

[0151] The adaptive algorithm dynamically adjusts parameters according to environmental changes, while the compression algorithm reduces the data volume through a mathematical model to ensure efficient image acquisition and transmission.

[0152] Further improved, in step S2, the method steps for preprocessing the collected image include the following:

[0153] S201, Denoising processing: Mean filtering: Smooth the image by calculating the mean of neighboring pixels to remove high-frequency noise;

[0154] Median filtering: Replace the central pixel with the median of neighboring pixels to effectively remove salt-and-pepper noise;

[0155] Gaussian filtering: Weight neighboring pixels using a Gaussian function to reduce random noise in the image;

[0156] S202, Image enhancement: Histogram equalization: Adjust the contrast of the image so that pixels of different brightness levels are evenly distributed to improve the overall visibility of the image;

[0157] Adaptive histogram equalization (CLAHE): Perform histogram equalization in a local area to avoid distortion caused by over-enhancement;

[0158] Gamma correction: Adjust the brightness and contrast of the image to adapt to different display conditions;

[0159] S203, Image segmentation: Threshold segmentation: Divide the image into foreground and background based on pixel intensity values, using a global threshold or an adaptive threshold;

[0160] Edge detection: Use an algorithm to extract the edges in the image for subsequent analysis;

[0161] Region growing: Start from a seed point and merge similar neighboring pixels to form a complete region;

[0162] S204, Feature extraction: HOG feature (Histogram of Oriented Gradients): Extract the shape and orientation features of objects in the image, suitable for human detection;

[0163] SURF feature (Speeded Up Robust Features): Extract the key points and descriptors of the image for object recognition;

[0164] SIFT feature (Scale-Invariant Feature Transform): Extract key points and calculate their descriptors, with scale and rotation invariance, suitable for complex scenes;

[0165] S205, Image scaling and transformation: Image scaling: Resize the image to a specified size to meet the input requirements of the deep learning model;

[0166] Rotation and flipping: Rotate or flip the image to enhance the diversity of the dataset and help the model learn features at different angles;

[0167] S206, Color Space Conversion: RGB to HSV Conversion: Convert the image from the RGB color space to the HSV color space to make it more intuitive for color analysis;

[0168] Grayscale Processing: Convert the color image to a grayscale image to reduce computational complexity while retaining important structural information;

[0169] S207, Data Augmentation: Random Cropping: Randomly crop small patches from the original image for training to increase data diversity;

[0170] Adding Noise: Add a small amount of noise to the original image to help the model better adapt to changes in the real environment;

[0171] Affine Transformation: Perform translation, scaling, and rotation transformations on the image to enhance the robustness of the model;

[0172] S208, Brightness and Contrast Adjustment: Brightness Adjustment: Improve the visibility of the image by increasing or decreasing the brightness value;

[0173] Contrast Adjustment: Increase or decrease the contrast between the light and dark parts of the image to make important features more prominent.

[0174] Through the above processing methods, the quality and feature expression ability of the image can be effectively improved, providing a more accurate data basis for subsequent abnormal event recognition.

[0175] Further improved, in step S3, deep learning technology is a machine learning method that automatically extracts features and learns by constructing a multi-layer neural network model. Based on artificial neural networks, it is suitable for processing high-dimensional data such as images, speech, and text. Deep learning technology includes multi-layer structures, automatic feature learning, and large-scale data processing;

[0176] Multi-layer Structure: Includes an input layer, hidden layers, and an output layer. Each hidden layer maps the input to a higher-dimensional feature space through a non-linear activation function;

[0177] Automatic Feature Learning: Different from traditional machine learning methods that require manual feature extraction, it automatically learns useful feature representations from the original data;

[0178] Large-scale Data Processing: Effectively utilize a large amount of labeled data for training;

[0179] The multi-layer neural network model is a deep learning model widely used in image processing and computer vision tasks, including convolutional layers, activation functions, pooling layers, fully connected layers, and output layers;

[0180] Convolutional layer: Extract local features of the input image through convolution operations. Each convolutional kernel (filter) slides over the image to calculate the weighted sum of the local region and generate a feature map; It contains multiple filters to capture different features in the image (such as edges, textures, etc.);

[0181] Activation function: Apply an activation function (such as ReLU, Rectified Linear Unit) after the convolutional layer to introduce non-linearity, enabling the network to learn complex feature representations;

[0182] Pooling layer: Downsample the feature map (such as max pooling or average pooling) to reduce the feature dimension, reduce computational complexity, and at the same time retain important features for improving the robustness of the model;

[0183] Fully connected layer: At the final stage of the network, the feature map is flattened into a one-dimensional vector and input into the fully connected layer for the final classification or regression task;

[0184] Output layer: According to the task of the model (such as multi-classification, binary classification, etc.), the output layer maps the final features to specific class labels through an activation function (such as softmax).

[0185] Automatically identify abnormal behaviors: By training on a labeled dataset of normal and abnormal behaviors, the CNN model can learn the features of different types of abnormal events; Update model parameters in real time: Perform online learning based on real-time data, adapt to environmental changes, and improve the recognition accuracy.

[0186] Specifically, in step S3, abnormal behaviors include personal safety behaviors, public order behaviors, suspicious behaviors, emergencies, traffic safety behaviors, natural disaster behaviors, behavior patterns, and emotional states;

[0187] Personal safety behaviors:

[0188] Fighting: Intense physical conflicts between two or more people;

[0189] Falling: An individual loses balance and falls to the ground due to an accident or other reasons, and may require emergency medical assistance;

[0190] Pushing and shoving: Strong physical contact between individuals in a crowd, which may trigger conflicts;

[0191] Public order behaviors:

[0192] Crowding: The crowd is dense and exceeds the safety capacity, which may lead to accidents such as stampedes;

[0193] Fighting with weapons: Physical conflicts involving weapons or obvious threatening behaviors;

[0194] Damage to public facilities: The act of deliberately damaging or destroying public property;

[0195] Suspicious behavior:

[0196] Loitering: Staying in a specific area for a long time or walking back and forth, which may indicate potential criminal intent;

[0197] Theft: Intentional contact with someone else's property, especially when no one is paying attention;

[0198] Illegal entry: The act of entering a prohibited or restricted area;

[0199] Emergency incidents:

[0200] Fire: The appearance of smoke or flames, which requires an immediate response;

[0201] Medical emergencies: Such as someone suddenly fainting or having other health problems;

[0202] Traffic safety behaviors:

[0203] Going against the traffic: Driving or walking in the opposite direction in a traffic area, which poses a safety hazard;

[0204] Running a red light: Vehicles or pedestrians driving or passing through when the red light is on, which may cause traffic accidents;

[0205] Natural disaster behaviors:

[0206] Escape behaviors: The escape behaviors of the crowd during natural disasters such as earthquakes and fires;

[0207] Abnormal climate reactions: Such as objects flying up or the chaotic reactions of the crowd during strong wind weather;

[0208] Behavior patterns:

[0209] Abnormal behavior: Individuals who remain stationary for a long time or have abnormal activity patterns may require further investigation;

[0210] Multiple people gathering: Multiple people gathering in a non-public activity area without an obvious purpose;

[0211] Emotional states:

[0212] Extreme emotional expressions: Individuals showing extreme anger, anxiety, or panic may lead to violent behavior.

[0213] The identification of these abnormal behaviors not only relies on visual features but can also be combined with other sensor data (such as audio, temperature, etc.) for multimodal analysis, thereby improving the accuracy and reliability of identification.

[0214] In summary, through intelligent image recognition and real-time analysis, the automatic recognition and rapid response to potential abnormal events are realized, significantly improving the monitoring efficiency of abnormal events. The application of deep learning models has significantly improved the accuracy of abnormal behavior recognition, reducing false alarms and missed detections. The system has the ability of self-learning, can adapt to changes in different environments and scenarios, and the rapid warning mechanism can notify security personnel in time, reducing potential risks. By organically combining image recognition, deep learning and security response mechanisms, an integrated prevention and control system is formed. Introducing a self-learning mechanism improves the flexibility and adaptability of the system, which is applicable to various complex environments such as public places, transportation hubs, and enterprise parks, and can effectively handle various abnormal situations.

[0215] The foregoing has shown and described the basic principles, main features and advantages of the present invention. For a person skilled in the art, it is obvious that the present invention is not limited to the details of the above-described exemplary embodiments, and without departing from the spirit or basic characteristics of the present invention, the present invention can be implemented in other specific forms. Therefore, in any aspect, the embodiments should be regarded as exemplary and non-limiting. The scope of the present invention is defined by the appended claims rather than the above description. Therefore, it is intended to embrace all changes within the meaning and scope of the equivalent elements of the claims in the present invention. Any reference signs in the claims should not be regarded as limiting the claims involved.

[0216] In addition, it should be understood that although this specification is described according to the embodiments, not every embodiment only contains an independent technical solution. This narrative way of the specification is only for clarity. Those skilled in the art should regard the specification as a whole, and the technical solutions in each embodiment can also be appropriately combined to form other embodiments that can be understood by those skilled in the art.

Claims

1. A method for preventing and controlling abnormal events based on image recognition, characterized in that: The method steps include the following: S1, data acquisition: use high-definition cameras to collect image data of the monitored area in real time and transmit it to the central processing system; S2, image preprocessing: processing the collected images to improve the accuracy of subsequent recognition; S3. Construction of abnormal event recognition model: Based on deep learning technology, a multi-layer convolutional neural network model is constructed. The training data includes annotated data sets of normal and abnormal behaviors. The model is used to identify a variety of abnormal behaviors and update model parameters in real time to adapt to the changing environment; S4, Real-time monitoring and analysis: Use the trained model to perform real-time image analysis and set thresholds to determine whether there are abnormal events; For the identified abnormal events, the system will automatically classify and assess their severity; S5. Early warning and response mechanism: Once an abnormal event is identified, the system will promptly notify security personnel through mobile applications, SMS notifications or alarm systems; Combined with location data, the system can guide security personnel to the scene of the incident quickly; S6. Feedback and learning mechanism: After handling an incident, security personnel will feed back the results to the system to promote self-learning and optimization of the model.

2. The abnormal event prevention and control method based on image recognition according to claim 1 is characterized in that: In step S1, the specific method steps for data collection include the following: S101, using the mobile platform of the UAV to carry out preliminary arrangement of cameras and select the optimal location according to the characteristics and needs of the monitoring area; Use lidar or ultrasonic sensors to model the environment in three dimensions, ensuring that the camera can automatically adjust its position to monitor at the best angle in a dynamic environment; S102, which uses sensors to monitor ambient light, temperature and humidity factors in real time, and intelligently adjusts the camera's exposure, focus and gain settings to adapt to different environmental conditions; Adopting adaptive algorithms, the frame rate and resolution are adjusted in real time according to dynamic changes in the monitoring area to ensure that high-definition images are captured at critical moments; S103, using cloud computing technology, uploads image data from various sensors to the central processing system in real time, and uses edge computing for preliminary processing to reduce latency; S104, during the image acquisition process, using a built-in intelligent algorithm to mark dynamic targets in the image in real time and identify different types of objects; By analyzing the target's moving trajectory, the occurrence of abnormal behavior can be predicted and relevant image data can be captured in advance; S105, during the data transmission process, the collected image data is encrypted in real time to ensure data security; Compression algorithms are used to reduce the amount of data transmission, and high-speed transmission is performed through dedicated networks or 5G networks to ensure real-time performance.

3. The abnormal event prevention and control method based on image recognition according to claim 2 is characterized in that: The adaptive algorithm dynamically adjusts parameters according to real-time data and environmental changes; The adaptive filter uses the LMS algorithm: w(n+1)=w(n)+μe(n)x(n); in: w(n): filter weight; μ: step size factor, controlling the convergence speed; e(n): error signal; x(n): input signal; Adaptive gain control uses the gain adjustment formula: G(n) = G(n-1) + α·(DP(n-1)); in: G(n): current gain; D: target value; P(n-1): the actual value at the previous moment; α: adjustment factor.

4. The method for preventing and controlling abnormal events based on image recognition according to claim 2 is characterized in that: The compression algorithm is used to reduce the space for data storage and transmission, and the compression algorithm is a JPEG compression algorithm; JPEG compression uses the discrete cosine transform (DCT) formula: in: f(x,y): image pixel value; C(u,v): DCT coefficient.

5. The method for preventing and controlling abnormal events based on image recognition according to claim 1, characterized in that: In step S2, the method steps for preprocessing the collected image include the following: S201, denoising processing: mean filtering: smoothing the image by calculating the mean of the neighborhood pixels to remove high-frequency noise; Median filtering: replace the central pixel with the median of the neighboring pixels to effectively remove salt and pepper noise; Gaussian filtering: Use Gaussian function to weight neighborhood pixels to reduce random noise in the image; S202, image enhancement: histogram equalization: by adjusting the contrast of the image, pixels of different brightness are evenly distributed to improve the overall visibility of the image; Adaptive histogram equalization: Perform histogram equalization in local areas to avoid distortion caused by over-enhancement; Gamma correction: adjusts the brightness and contrast of the image to suit different display conditions; S203, image segmentation: Threshold segmentation: the image is divided into foreground and background according to the pixel intensity value, using a global threshold or an adaptive threshold; Edge detection: using algorithms to extract edges in an image; Region growing: Starting from a seed point, similar neighboring pixels are merged to form a complete region; S204, feature extraction: HOG feature: extracts the shape and direction features of objects in the image, suitable for human detection; SURF features: extract key points and descriptors of images for object recognition; SIFT feature: extracts key points and calculates their descriptors, which are scale and rotation invariant and suitable for complex scenes; S205, image scaling and transformation: Image scaling: adjusting the image to a specified size to meet the input requirements of the deep learning model; Rotation and flipping: Rotate or flip the image to increase the diversity of the dataset and help the model learn features from different angles. S206, color space conversion: RGB to HSV conversion: converting the image from RGB color space to HSV color space to facilitate more intuitive color analysis; Grayscale processing: Convert color images into grayscale images to reduce computational complexity while retaining important structural information; S207, data enhancement: random cropping: randomly crop small pieces from the original image for training to increase data diversity; Add noise: Add a small amount of noise to the original image to help the model better adapt to changes in the real environment; Affine transformation: translate, scale, and rotate the image to enhance the robustness of the model; S208, brightness and contrast adjustment: Brightness adjustment: improving the visibility of the image by increasing or decreasing the brightness value; Contrast adjustment: Increase or decrease the contrast between light and dark parts of an image to make important features more visible.

6. The abnormal event prevention and control method based on image recognition according to claim 1 is characterized in that: In step S3, deep learning technology includes multi-layer structure, automatic feature learning and large-scale data processing; Multi-layer structure: including input layer, hidden layer and output layer. Each hidden layer maps the input to a higher-dimensional feature space through a nonlinear activation function. Automatic feature learning: Automatically learn useful feature representations from raw data; Large-scale data processing: effectively use a large amount of labeled data for training; The multi-layer neural network model includes convolutional layers, activation functions, pooling layers, fully connected layers, and output layers; Convolution layer: extracts local features of the input image through convolution operation. Each convolution kernel slides on the image, calculates the weighted sum of the local area, and generates a feature map. It contains multiple filters to capture different features in the image. Activation function: An activation function is applied after the convolutional layer to introduce nonlinearity, allowing the network to learn complex feature representations; Pooling layer: downsample the feature map; Fully connected layer: At the final stage of the network, the feature map is flattened into a one-dimensional vector and input into the fully connected layer for the final classification or regression task; Output layer: According to the task of the model, the output layer maps the final features to specific category labels through an activation function.

7. The method for preventing and controlling abnormal events based on image recognition according to claim 1, characterized in that: In step S3, abnormal behaviors include personal safety behaviors, public order behaviors, suspicious behaviors, emergency events, traffic safety behaviors, natural disaster behaviors, behavioral patterns, and emotional states; Personal safety behavior: Fight: A violent physical altercation between two or more people; Fall: An individual loses balance and falls to the ground due to an accident or other reason and may require emergency medical assistance; Pushing and shoving: strong physical contact between individuals in a crowd, which may lead to conflict; Public Order Acts: Crowding: The crowd is dense and exceeds the safety capacity, which may cause accidents such as stampedes; Fighting: A physical altercation involving a weapon or overtly threatening behavior; Vandalism: The act of intentionally damaging or destroying public property; Suspicious behavior: Loitering: Staying or moving around in a specific area for a long time, which may indicate potential criminal intent; Theft: intentionally accessing another person's property, especially when no one is watching; Trespass: The act of entering a prohibited or restricted area; Emergency: Fire: The presence of smoke or flames requires immediate response; Medical emergencies: if someone suddenly faints or has other health problems; Traffic safety behavior: Going against the flow: Driving or walking against the flow in a traffic area poses a safety hazard; Running a red light: Vehicles or pedestrians driving or passing under a red light signal may cause traffic accidents; Natural Disaster Behavior: Escape behavior: the escape behavior of people in the event of natural disasters such as earthquakes and fires; Abnormal weather reactions: such as objects flying or crowds reacting in a chaotic manner in windy weather; Behavior pattern: Abnormal behavior: individuals who remain motionless for long periods of time or have unusual activity patterns may require further investigation; Gathering of multiple people: In non-public activity areas, multiple people gather without obvious purpose; Emotional state: Extreme emotional displays: The individual displays extreme anger, anxiety, or panic, which may lead to violent behavior.

Citation Information

Cited By

  • Integrated optical cable wiring cabinet and use method

    CN120276104A