Regional safety monitoring method based on infrared image

Abstract

The invention relates to the technical field of infrared safety monitoring, and discloses an infrared image-based area safety monitoring method, which comprises the following steps of: acquiring an infrared image of a target area through infrared camera equipment; performing denoising and detail enhancement processing on the acquired infrared image to obtain a denoised image; performing frequency domain conversion on the denoised image, and extracting high-frequency features of a target area; performing anomaly detection on the high-frequency features of the target region through a sparse representation model, and judging whether the target is abnormal or not; outputting an alarm signal or triggering linkage response equipment according to an abnormal detection result; and carrying out storage and analysis processing on the abnormal target and related monitoring data. The infrared radiation signals are accurately captured through quantization and digital conversion by adopting an acquisition technology that infrared camera equipment is combined with an active infrared sensor and a passive infrared sensor. The technical effect of acquiring high-quality infrared images in real time is achieved, and the problems of low acquisition efficiency and insufficient image signal stability are solved.

Description

Technical Field

[0001] The invention relates to the technical field of infrared safety monitoring, and in particular to a regional safety monitoring method based on infrared images. Background Art

[0002] With the continuous development of science and technology, the demand for regional security monitoring is growing, especially in unmanned or complex environments, where existing monitoring technologies are difficult to provide efficient and reliable monitoring results. In such scenarios, how to clearly capture target images, intelligently identify abnormal targets and respond in a timely manner in an all-weather environment has become a technical problem that needs to be solved in regional security monitoring; Existing regional monitoring technologies mainly include enhanced visible light cameras, low-light cameras and thermal imaging cameras, and their working principles have different focuses: enhanced visible light cameras achieve night vision effects through external light sources or self-filling light; low-light cameras use extremely weak light signals to form images; thermal imaging cameras generate images by detecting the temperature difference between the target and the background; However, the environmental adaptability of existing regional monitoring technologies is poor. Enhanced visible light cameras completely rely on external light sources or supplementary lighting equipment, and are difficult to adapt to nighttime or completely dark scenes. Low-light cameras have insufficient performance in dark environments, and image quality is unstable in dynamic scenes. Although thermal imaging cameras can work in dark conditions, their equipment costs are high and difficult to apply on a large scale. Secondly, the image processing methods of existing technologies are difficult to achieve detail retention and noise removal at the same time, resulting in loss of image details or residual noise, affecting the monitoring effect. In addition, these technologies have limited performance in target feature extraction and background separation, especially in long-distance or complex background scenes, and cannot effectively distinguish between targets and backgrounds. Summary of the invention

[0003] In view of the shortcomings of the prior art, the present invention provides a regional security monitoring method based on infrared images, which solves the problems that the existing regional monitoring technology has poor environmental adaptability, the enhanced visible light camera is completely dependent on external light sources or fill light equipment, and is difficult to adapt to scenes at night or in complete darkness; the low-light camera has insufficient performance in a dark environment and unstable image quality in dynamic scenes.

[0004] To achieve the above objectives, the present invention is implemented through the following technical solutions: A regional safety monitoring method based on infrared images comprises the following steps: Collect infrared images of the target area through infrared camera equipment; De-noising and detail enhancement processing are performed on the collected infrared image to obtain a denoised image; Perform frequency domain conversion on the denoised image to extract high-frequency features of the target area; The sparse representation model is used to detect anomalies in the high-frequency features of the target area to determine whether the target is abnormal; Output alarm signals or trigger linkage response equipment based on abnormal detection results; Store, analyze and process abnormal targets and related monitoring data.

[0005] Preferably, the infrared camera equipment acquisition includes the following steps: Use infrared camera equipment to collect data of the target area in real time. The infrared camera equipment includes passive infrared sensors and active infrared sensors. Quantify and digitally convert the collected original infrared signals, and convert the infrared radiation intensity into digital image data; The clarity and real-time performance of image acquisition can be adjusted by setting the sampling frequency and resolution parameters. The sampling frequency is adjusted according to the dynamics of the monitored area, and the resolution is achieved by adjusting the sensor pixel parameters. The collected data is preprocessed in real time, including removing background interference signals and noise during the collection process to ensure the effectiveness of the infrared image.

[0006] Preferably, the denoising and detail enhancement processing uses a nonlinear partial differential equation model to process the image, and the expression of the partial differential equation is:

[0007] in, Represents the image gradient, reflecting the rate of change of image pixels; is the diffusion control function; For the fidelity item; Represents the acquired original infrared image.

[0008] Preferably, the denoising and detail enhancement processing is achieved by minimizing the following energy functional:

[0009] in, is the total energy of image processing; Item 1 Represents the smoothing energy of the image; is the diffusion coefficient control function; Item 2 Represents the fidelity energy of the image; Represents the deviation between the pixel value of the processed image and the pixel value of the original image; is the weight parameter of the fidelity term.

[0010] Preferably, the performing frequency domain conversion on the denoised image to extract the high-frequency features of the target area comprises the following steps: Perform Fourier transform on the image data after denoising and detail enhancement to convert the image from time domain representation to frequency domain representation; Apply a bandpass filter to the frequency domain image for processing. The bandpass filter retains the frequency range of the target features in the frequency domain and suppresses other frequency components; Perform inverse Fourier transform on the frequency domain data processed by the bandpass filter to restore the data from frequency domain representation to time domain representation; The image reconstructed by inverse Fourier transform is used as the target enhanced image for the subsequent anomaly detection step.

[0011] Preferably, the sparse representation model represents the features of the target area as a sparse linear combination of the feature dictionary, and the specific expression is:

[0012] in, Indicates the characteristics of the target area; Represents a feature dictionary; is a sparse coefficient vector; is the reconstruction error.

[0013] Preferably, the anomaly detection solves the sparse coefficients by the following optimization problem:

[0014] in, The observed feature vector representing the target area; represents the feature dictionary matrix; represents a sparse coefficient vector; represents the sum of squares of reconstruction errors; Represents the weight parameter of the sparse constraint term.

[0015] Preferably, the abnormal detection result outputs an alarm signal or triggers a linkage response device including: Trigger local sound and light alarm equipment and generate alarm information including target category, location and time; Sending the alarm information to the user terminal via the remote communication module; According to the alarm information, control the external equipment to perform linkage operations, including the camera direction adjustment device, the access control system control device and the regional lighting control device; The alarm information and linkage results are stored in the local storage module for subsequent data analysis and response optimization.

[0016] Preferably, the storage and analysis of abnormal targets and related monitoring data includes the following steps: The detected abnormal targets and their related data, including target category, location, time, reconstruction error and alarm information, are stored in the local storage module; Classify and manage the stored abnormal target data and normal target data to generate classified data sets; Utilize the stored abnormal target data to optimize the feature dictionary and enhance the accuracy of the sparse representation model; Generate historical detection record analysis reports through the data analysis module to evaluate the safety status of the monitored area; Output optimization suggestions and upload historical data to the remote management platform to support cross-regional security management and resource scheduling.

[0017] Preferably, the regional safety monitoring system based on infrared images includes: Infrared image acquisition module, used for acquiring infrared images of target areas; Image processing module, used for denoising and detail enhancement of infrared images; The frequency domain analysis module is used to perform frequency domain conversion on the denoised image and extract target features; Anomaly detection module, used to detect abnormal targets based on target features based on sparse representation model; An alarm response module is used to output an alarm signal or trigger a linkage device after detecting an abnormal target; The storage module is used to store anomaly detection results and related data.

[0018] The present invention provides a regional safety monitoring method based on infrared images. It has the following beneficial effects: 1. The present invention uses infrared camera equipment combined with active infrared sensors and passive infrared sensors to accurately capture infrared radiation signals through quantization and digital conversion, and enhances the adaptability of collected data through dynamic adjustment of sampling frequency and resolution. The technical effect of real-time acquisition of high-quality infrared images is achieved. Compared with the technical solution in the prior art that relies on a single sensor acquisition mode and results in poor dynamic adaptability of images and large signal noise, the problem of low acquisition efficiency and insufficient stability of image signals is solved.

[0019] 2. The present invention adopts a denoising and detail enhancement processing technology based on nonlinear partial differential equations, adjusts the difference in diffusion intensity between the edge area and the background area through a diffusion control function, and combines the optimization solution of the energy functional to remove noise and effectively retain the detail features of the target edge. The technical effect of improving the quality and detail clarity of infrared images is achieved. Compared with the problem of image detail loss and noise residual caused by relying on simple filtering denoising or edge enhancement processing in the prior art, the problem of image detail and noise suppression cannot be taken into account at the same time is solved.

[0020] 3. The present invention adopts a sparse representation model, constructs a feature dictionary consisting of normal targets and abnormal targets, reconstructs and analyzes the features of the target area based on sparse linear combinations, and combines reconstruction errors and sparsity constraints to achieve accurate detection of abnormal targets. The technical effect of fast and high-precision identification of abnormal targets is achieved. Compared with the technical solutions in the prior art that rely on template matching or simple threshold judgment methods, the problems of low accuracy and poor adaptability of abnormal detection are solved.

[0021] 4. The present invention classifies and manages the detected abnormal target data and historical monitoring records through the storage module, generates a detection analysis report in combination with the data analysis module, and optimizes the sparse representation feature dictionary using the stored data, thereby enhancing the accuracy and adaptability of the anomaly detection model. The technical effect of improving data utilization and the adaptive optimization capability of the anomaly detection model is achieved. Compared with the technical solutions in the prior art that lack effective data management and model self-optimization capabilities, the problem of model degradation and data resource waste in the long-term use of the anomaly detection system is solved. BRIEF DESCRIPTION OF THE DRAWINGS

[0022] Figure 1 is a flow chart of the method of the present invention; Figure 2 It is a system architecture diagram of the present invention. DETAILED DESCRIPTION

[0023] The technical solution of the present invention will be described clearly and completely below in conjunction with the accompanying drawings of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0024] The present invention provides a regional security monitoring method based on infrared images. By collecting, processing, analyzing and responding to abnormalities of infrared images, efficient detection, alarm and management of abnormal targets in the monitoring area are achieved, thereby improving the intelligent level of regional security monitoring.

[0025] like Figure 1 As shown, the regional safety monitoring method of the present invention may include the following steps: S1. Collect infrared images of the target area through infrared camera equipment; S2. performing denoising and detail enhancement processing on the collected infrared image to obtain a denoised image; S3. Perform frequency domain conversion on the denoised image to extract high-frequency features of the target area; S4. Perform anomaly detection on the high-frequency features of the target area through a sparse representation model to determine whether the target is abnormal; S5. Output an alarm signal or trigger a linkage response device based on the abnormal detection results; S6. Store, analyze and process abnormal targets and related monitoring data.

[0026] For step S1, the infrared data acquisition capability of multiple scenes and multiple targets is realized by combining hardware optimization with algorithm processing, and high-quality input is provided for subsequent frequency domain processing and anomaly detection. The infrared image acquisition process includes the following: Infrared image acquisition is completed by high-precision infrared camera equipment, which includes two types of equipment: passive infrared sensors and active infrared sensors. Passive infrared sensors: Generate image data by receiving thermal radiation signals from the target area, suitable for most scene monitoring. Active infrared sensors: Generate image data by emitting infrared light of a specific band and receiving reflected signals, suitable for low-contrast scenes (such as outdoor monitoring at night).

[0027] As an option, a thermal imaging sensor with a wavelength range of 3μm to 14μm can be selected to cover the thermal radiation wavelengths of most targets. In one possible implementation, by optimizing the optical lens parameters (such as focal length and aperture) of the infrared camera equipment, higher image resolution and wider monitoring range can be achieved. For example, for long-distance monitoring scenarios, a telephoto lens with a focal length of 50mm can be used in combination with a sensor module with a resolution of 640×480 to obtain a clear target image.

[0028] The collected original infrared signal is an analog signal and needs to be converted into digital image data through an analog-to-digital converter (ADC).

[0029] Specifically, the bit depth of analog-to-digital conversion directly affects the quantization accuracy of data. In the present invention, some embodiments use a 16-bit analog-to-digital converter, so that the device can capture a temperature difference of 0.03°C, which is suitable for high-precision monitoring scenarios.

[0030] As a possible implementation method, the sampling frequency can be dynamically adjusted to meet the monitoring requirements of different target areas. For example, the sampling frequency is set to 30 frames per second in high-speed motion scenes, and reduced to 10 frames per second in low-dynamic scenes to save computing resources.

[0031] Generally, background thermal noise is eliminated by a high-pass filter, and interference signals in non-target bands are shielded by an optical filter.

[0032] Specifically, random noise can be eliminated in real time in combination with an image preprocessing algorithm. In a possible implementation, a mean filter or a median filter based on a pixel domain can be used to process the random noise introduced by the sensor.

[0033] To meet the needs of multi-scene monitoring, the working parameters of the infrared camera equipment, such as the field of view and sensitivity, are dynamically adjusted. In some embodiments, the focus and direction of the camera can be adjusted by the servo control system to ensure that the target is always in the monitoring field of view. Specifically, the servo control system can be linked with the motion trajectory of the monitored target to achieve dynamic monitoring.

[0034] In infrared image acquisition, the infrared radiation signal of the target area can be described by the following formula:

[0035] in, For the location Infrared signal strength at is the emissivity of the target surface; is the radiance of the target; is the noise introduced during the acquisition process.

[0036] For step S2, by adopting denoising and detail enhancement technology based on mathematical modeling, high-quality input data is provided for subsequent frequency domain conversion, target feature extraction and anomaly detection. In general, due to the influence of external interference such as thermal noise and sensor noise during the infrared image acquisition process, the image often has a high noise level and blurred edge details. By introducing technology based on nonlinear partial differential equations and energy functional optimization, the present invention can effectively suppress noise while retaining image edge features to the greatest extent, thereby improving the quality of infrared images and the accuracy of subsequent processing.

[0037] Specifically, denoising is performed by using a nonlinear partial differential equation based on diffusion theory, and edge detail features are further enhanced by optimizing the energy functional. The combination of the two allows denoising and detail enhancement to achieve a good balance in efficiency and accuracy. Image denoising and detail enhancement are implemented as follows: The infrared image is denoised by constructing a nonlinear partial differential equation. The specific equation is as follows:

[0038] in, Represents image gradient; For the fidelity item; Represents the collected original infrared image; is the diffusion control function, the specific form is:

[0039] in, is the threshold parameter; In one implementation, a numerical method is used to solve the above partial differential equation. Through discretization, the time derivative and the space derivative are discretized using the finite difference method respectively, and the iterative formula after discretization is:

[0040] Indicates The pixel value after iterations; is the iteration step length, is the discrete form of pixel space gradient.

[0041] In order to enhance the edge details of the image while removing noise, the present invention further introduces an optimization method based on energy functional. Specifically, the following energy functional model is constructed:

[0042] in, is the total energy of image processing; Item 1 Represents the smoothing energy of the image; is the diffusion coefficient control function; Item 2 Represents the fidelity energy of the image; Represents the deviation between the pixel value of the processed image and the pixel value of the original image; is the weight parameter of the fidelity term.

[0043] During the optimization process, by minimizing the energy functional , achieving a dynamic balance between noise suppression and edge protection. The gradient descent method is used to solve the optimization problem, and the update formula is as follows:

[0044] in, is the step length coefficient; is the gradient of the energy functional, and the calculation result is:

[0045] In some embodiments, by setting a larger fidelity term weight parameter , which can further protect the detailed features in the image.

[0046] In general, the parameters in the diffusion control function can be adjusted according to the actual scenario. For example, in low-contrast images, a smaller value can be selected to enhance edge protection; in a high-noise environment, the value can be appropriately increased to improve noise suppression. As an option, the weight of the fidelity term can also be dynamically adjusted to meet the needs of different application scenarios. For example, in a scenario with high dynamics in the monitored area, a lower value can be selected to enhance the denoising effect.

[0047] Step S3 is mainly used to perform frequency domain conversion on the infrared image data after denoising and detail enhancement, and extract the high-frequency features of the target area. This process is implemented through digital signal processing technologies such as Fourier transform, bandpass filter and inverse Fourier transform, and has a clear mathematical basis and reproducibility. It is the core step of signal feature extraction, and the specific implementation steps are as follows: After denoising and detail enhancement, the processed infrared image data Input to the frequency domain conversion module, first perform Fourier transform on the image data. Fourier transform is a mathematical tool that converts an image from time domain representation to frequency domain representation. Its expression is as follows:

[0048] in, is the denoised image represented in the time domain; is the image represented in the frequency domain.

[0049] In a possible implementation, in order to meet the needs of digital processing, discrete Fourier transform can be used. The expression of discrete Fourier transform is:

[0050] After Fourier transform, the characteristics of the frequency domain image can be clearly distinguished: the low-frequency components are concentrated in the center of the frequency domain image, reflecting the background or overall changes; the high-frequency components are distributed at the edge of the frequency domain image, reflecting the edge and detail features of the target.

[0051] Frequency domain image processed by bandpass filter It contains enhanced features of the target area while suppressing background interference and noise interference.

[0052] Then, the inverse Fourier transform is used to restore the filtered frequency domain image to the time domain. The formula for the inverse Fourier transform is:

[0053] In one digital implementation, a discrete inverse Fourier transform may be used, which is expressed as:

[0054] in, is the pixel value in the time domain image; is the frequency domain image after being processed by the bandpass filter; are the horizontal and vertical frequency variables in the frequency domain; ， are the horizontal and vertical spatial variables in the time domain; is an imaginary unit; and is the width and height of the image; and Respectively represent the phase contribution of the frequency components in the horizontal and vertical directions to the current pixel; is the normalization factor of the discrete inverse Fourier transform; Final time domain image It includes the following features: the high-frequency details of the target area are significantly enhanced; the background and noise interference are effectively suppressed.

[0055] Step S4 is described in detail for detecting abnormalities in the high-frequency features of the target area through a sparse representation model. This step is a key step for in-depth analysis of the processed image after denoising, frequency domain conversion, and high-frequency feature extraction. Its purpose is to model and analyze the characteristic information of the target area and detect whether there are abnormal targets. Through the sparse representation model, the present invention can achieve efficient detection of abnormal targets and provide reliable data support for subsequent alarms and responses.

[0056] In technical implementation, the sparse representation model is a mathematical tool widely used in the field of signal processing. Its core idea is to represent the target features as a sparse linear combination of a feature dictionary. The present invention combines the characteristics of infrared images, constructs a sparse dictionary for infrared image features, and achieves accurate and efficient anomaly detection by optimizing the sparse representation coefficients.

[0057] The high-frequency feature image of the target area is obtained through frequency domain conversion and high-frequency feature extraction steps. Target feature image It is the result of inverse Fourier transform, which retains the high-frequency detail features of the target area and removes low-frequency background interference and irrelevant noise. For these high-frequency features, this step constructs a mathematical model based on sparse representation to determine the abnormality of the target area.

[0058] In general, the sparse representation model uses a feature dictionary Target Features Perform sparse decomposition, expressed as follows;

[0059] in, Indicates the characteristics of the target area; Represents a feature dictionary; is a sparse coefficient vector; is the reconstruction error.

[0060] Specifically, the feature dictionary It consists of two parts One part is for the standard features of normal target areas, collected from historical data; The other part is the standard features for abnormal target areas, which are obtained through experimental data and model training.

[0061] In one possible implementation, the solution of the sparse representation model is achieved by optimizing the following objective function:

[0062] in, is the reconstruction error term; is the sparse regularization term; is the regularization parameter.

[0063] As an option, the sparse coefficients in this embodiment The solution can be solved by optimization methods, such as Lagrange multiplier method or alternating direction multiplier method. These methods converge quickly through iterative optimization and can effectively solve the sparse representation problem under large-scale data.

[0064] In another possible implementation, in order to improve the generalization ability of the model, the feature dictionary It can be dynamically updated based on the detected abnormal data. Specifically, when a new abnormal target is detected, its features are added to the abnormal part of the feature dictionary, and the dictionary is retrained through the dictionary update algorithm to improve the robustness and accuracy of anomaly detection.

[0065] For the abnormal detection results obtained by using the sparse representation model in step S5, real-time response and data feedback are further realized. In general, in order to ensure the timely handling of security incidents, a variety of alarm mechanisms and linkage control strategies are designed to enhance the security and practicality of the system. Specifically, it includes the processing of detection results and the triggering of alarm mechanisms, and also involves the control of linkage response equipment, storage and transmission of alarm data.

[0066] In this embodiment, through the generation and transmission of alarm signals, linkage devices (such as camera direction adjustment devices, lighting systems, and access control systems) can achieve multi-level responses based on abnormal detection results. Specifically, the linkage control from detection results to response devices involves four main links: abnormality determination, alarm signal generation, response device control, and data storage. The logic between each link is clear and the connection is tight.

[0067] In one possible implementation, when the reconstruction error value calculated by the sparse representation model is Exceeding the set threshold , the system determines that the target is an abnormal target. Specifically, the abnormal detection module outputs the determination result to the alarm response module. At this time, the system generates an alarm signal, including the location coordinates of the abnormal target, the detection time, the abnormal category (such as human intrusion, fire source), and the determination basis (such as reconstruction error value and sparse coefficient distribution).

[0068] As an option, after the alarm signal is generated, the system first triggers the local sound and light alarm device. Specifically, the local alarm device can be a buzzer, flashing light or other sound and light equipment installed in the monitoring area. When the alarm is triggered, the alarm device sounds an alarm at a preset frequency and flashes the light to attract the attention of people in the area. This sound and light alarm method is suitable for scenes where there are people on duty in the monitoring area.

[0069] In another implementation, the system of the present invention sends the alarm information to the user terminal through a remote communication module. The user terminal may include a mobile device (such as a smart phone, a tablet computer) or a computer terminal. Specifically, the alarm information can be transmitted to the remote terminal through a wireless communication network (such as 4G, 5G) or a wired network. Generally, the alarm information is delivered to the user in real time in the form of a push message or an email. In some embodiments, the alarm information will be accompanied by the collected infrared image of the abnormal target so that the user can further confirm the abnormal situation.

[0070] Specifically, the transmission of alarm information can also be combined with cloud storage and analysis functions. Alarm data is uploaded to the cloud security monitoring platform through the transmission module. The cloud platform can analyze historical alarm data and generate security reports based on the alarm frequency of different areas. For example, when the alarm frequency of a certain monitoring area increases significantly within a specific time period, the cloud platform can automatically generate early warning information to prompt administrators to strengthen patrols in related areas.

[0071] As an option, after detecting an abnormal target, the system of the present invention can also trigger a response device linked to the alarm module. Specifically, the response device may include the following: Camera direction adjustment device: The system adjusts the camera direction through the pan / tilt control module according to the position coordinates of the abnormal target, so that the camera can track and enlarge the abnormal target area in real time.

[0072] Lighting control device: At night or in low light conditions, the system can enhance the visibility of abnormal targets by turning on fill lights in the monitoring area, while also acting as a deterrent.

[0073] Access control system control device: When an abnormal target approaches the boundary of the monitoring area or the access control area, the system can trigger the locking function of the access control system to prevent the abnormal target from further entering the sensitive area.

[0074] In one possible implementation, the specific parameters of the linkage response can be dynamically adjusted according to the category of the abnormal target. For example, when the abnormal target is a fire source, the system will give priority to triggering the linked fire warning device and send an alarm message to the fire center; when the abnormal target is an intruder, the system will give priority to controlling the access control system to isolate the area and notify security personnel to come to the scene to handle it.

[0075] Furthermore, in this embodiment, in order to achieve efficient storage and management of abnormal target information, the system stores the alarm signal, linkage response results and abnormal target images in a unified manner in the local storage module. Generally, the stored data format is structured data to facilitate subsequent retrieval and analysis. For example, the stored content may include the following fields: Abnormal target position: indicates the relative coordinates of the target in the monitoring area; Detection time: record the timestamp of the abnormal situation; Reconstruction error value: records the reconstruction error value calculated by the sparse representation model ; Response device status: record the working status of the linked device after the alarm is triggered, such as camera direction and light switch status.

[0076] In a specific implementation, the stored data can also be uploaded to a remote management platform via the network. The management platform can collect and analyze the alarm data, such as calculating the alarm frequency in a certain area, analyzing the behavior patterns of abnormal targets, etc. These analysis results can provide a decision-making basis for the security management of the monitored area. For example, the analysis results may prompt the administrator to increase the density of cameras or strengthen the patrol frequency in high-frequency alarm areas.

[0077] For step S6, the systematic storage, intelligent management and in-depth analysis of abnormal targets and related data provide support for long-term optimization and resource scheduling of security monitoring. Combining the results of abnormal target detection and historical data analysis, the modular data storage architecture and optimization model are implemented to ensure efficient recording and intelligent management of abnormal events.

[0078] The storage of abnormal targets and related monitoring data is based on a modular storage architecture design. The storage content of the data includes but is not limited to the following: Target category: the type of target anomaly detected, such as human, vehicle, or fire source, obtained through the output of the sparse representation model; Target location: the specific pixel coordinates of the abnormal target in the infrared image; Detection time: including the timestamp of the event to support historical backtracking and multi-target event management; Reconstruction error: The reconstruction error value detected in the sparse representation model is used to further analyze the abnormality of the target; Alarm information: includes relevant information about the triggering alarm, such as alarm category and linkage device status.

[0079] As an implementation method, these data are stored in a local storage module in a structured format. Generally, a relational database is used for storage, and the data can be organized in a table form, where each record contains complete fields of target category, location, timestamp, reconstruction error, and alarm information.

[0080] Specifically, in some embodiments, a distributed file storage system may be used to expand data to meet the storage requirements of large-scale data volumes. In one possible implementation, a distributed storage system, such as HDFS, may be used to manage abnormal data generated by multiple monitoring devices to ensure efficient data writing and reading performance.

[0081] In order to facilitate subsequent data analysis and model optimization, abnormal target data and normal target data are classified and managed. The classification is based on the detection results of the sparse representation model, and is mainly based on the comparison between the target's reconstruction error value and the threshold.

[0082] In general, when the reconstruction error When the preset threshold is exceeded, the target is judged as an abnormal target, otherwise it is a normal target.

[0083] As an option, the abnormal target data and the normal target data are stored in different classification folders, or different label fields are set in the database to identify the category of the data. In some embodiments, a machine learning classification algorithm can be used to further optimize the classification results to improve the accuracy and efficiency of data classification.

[0084] In one possible implementation, the stored abnormal target data is used to identify the sparse feature dictionary. Optimize to improve the detection accuracy of sparse representation models.

[0085] Specifically, the optimization of the sparse feature dictionary is based on the following steps: Dictionary update: Based on the stored abnormal target data, the feature dictionary is dynamically updated through the sparse representation learning algorithm to introduce new abnormal target features. Sparsity constraint: During the update process, the sparsity of the dictionary is maintained to ensure that the newly added features do not affect the representation ability of the existing features.

[0086] The stored historical data is analyzed and processed through the data analysis module to evaluate the safety status of the monitored area and generate a historical detection record analysis report.

[0087] As an implementation method, the following technical steps can be adopted: cluster analysis of historical abnormal target data, generate abnormal heat maps based on target category, location and time distribution, and use them to evaluate high-risk areas; determine the frequency and duration of abnormal events based on time series analysis of reconstruction errors, and infer potential safety hazards; generate regular inspection reports, which include the type distribution of high-frequency abnormal targets, the changing trend of risk areas, and the response efficiency of alarm equipment.

[0088] Specifically, in some embodiments, the stored historical data of multiple devices can be jointly analyzed through a big data analysis platform to support regional security optimization strategies.

[0089] Generally, the analyzed data and optimization suggestions are uploaded to the security management platform through a remote interface to support cross-regional resource scheduling and security management. In one possible implementation, the data output includes the following parts: historical detection record report; distribution map of high-risk areas; and update results of the optimization feature dictionary.

[0090] As an option, in this embodiment, an encrypted transmission protocol is used to protect the uploaded data to ensure the security and privacy of the data.

[0091] The regional safety monitoring system based on infrared images described below and the regional safety monitoring method based on infrared images described above may refer to each other.

[0092] Please see attached Figure 2 The present invention also provides a regional security monitoring system based on infrared images, which combines infrared image acquisition, data processing, frequency domain analysis, sparse representation anomaly detection, response linkage and data storage analysis. By organically integrating the above technical modules, the system can work stably in complex environments, achieve high-precision detection of abnormal targets, real-time alarm and efficient management, and provide complete technical support for regional security monitoring.

[0093] The regional safety monitoring system based on infrared images in this implementation includes the following functional modules: infrared image acquisition module, image processing module, frequency domain analysis module, anomaly detection module, alarm response module and storage module. The modules work together to form a closed-loop system, which can efficiently complete the whole process from data acquisition to anomaly detection to alarm response and data storage, and realize comprehensive, intelligent and accurate regional safety monitoring functions.

[0094] It is used to collect infrared images of the target area, capture infrared radiation signals in the area through infrared sensors, and convert them into digital image data. The module supports passive infrared sensors and active infrared sensors, and can work stably at night, in low light conditions or in complex environments.

[0095] In some embodiments, the acquisition module supports real-time sampling and multi-resolution adjustment functions, and can dynamically adjust the resolution and frame rate according to monitoring requirements to optimize acquisition performance. The collected data is pre-processed to remove background noise and interference to ensure the quality of the original image and provide high-quality input for subsequent processing.

[0096] It can collect thermal signal data of the monitored area around the clock without being affected by changes in light. It is particularly suitable for nighttime monitoring and target detection in low visible light environments, improving the system's applicability and monitoring coverage.

[0097] The image processing module is used to perform denoising and detail enhancement on infrared images, improve image quality and feature retention capabilities, and ensure the accuracy and stability of subsequent analysis.

[0098] In some embodiments, a nonlinear partial differential equation model is used to denoise the image, and a diffusion function is used to adaptively adjust the smoothness of the image, thereby effectively removing noise while retaining edge features. At the same time, an energy functional optimization method is used to achieve image fidelity and efficient processing, ensuring that the processed image is both clear and realistic.

[0099] Through denoising and detail enhancement processing, the signal-to-noise ratio and edge feature clarity of the image are significantly improved, providing high-quality input data for subsequent frequency domain analysis and anomaly detection, ensuring the reliability of system detection.

[0100] The frequency domain analysis module extracts the high-frequency features of the target by performing Fourier transform and bandpass filtering on the processed infrared image, while suppressing background noise and other interference signals.

[0101] In some embodiments, the frequency range of the bandpass filter is dynamically adjusted to optimize the high-frequency feature extraction effect according to the frequency distribution of the target feature. The filtered image is reconstructed into a time domain image through an inverse Fourier transform to generate a target enhanced image, ensuring that the input features of subsequent anomaly detection are more prominent.

[0102] Through frequency domain analysis, low-frequency background signals and high-frequency noise are effectively eliminated, highlighting the edge features and detail features of the target, providing more accurate and significant target information for anomaly detection.

[0103] The anomaly detection module analyzes the high-frequency features of the target area based on the sparse representation model, determines whether the target is abnormal, and outputs the corresponding detection results.

[0104] In this implementation, a sparse representation model is used to represent the target image as a sparse linear combination of feature dictionaries, and the reconstruction and classification of target features are achieved by minimizing the sparse optimization problem. The feature dictionary in the model is continuously optimized and can dynamically adapt to newly added abnormal targets, improving the accuracy and flexibility of detection.

[0105] The anomaly detection module can accurately detect and classify different types of abnormal targets, significantly improving the system's detection capabilities in multiple targets and scenarios, and reducing false positives and missed positives.

[0106] This module is used to output an alarm signal after detecting an abnormal target and link external response equipment to achieve real-time response.

[0107] In some embodiments, the alarm module supports local sound and light alarm and remote alarm functions, and can also control external devices (such as adjusting the direction of the camera, locking the door or turning on the area lighting) for linkage response. The alarm information includes detailed information such as the abnormal target category, location, time, etc.

[0108] The alarm response module ensures that the system can issue an alarm as soon as an abnormality is detected, and interact with external devices to quickly handle abnormal events, greatly improving the immediacy and reliability of regional security.

[0109] The storage module is used to record, classify and analyze abnormal targets and related monitoring data, and supports historical backtracking and system optimization.

[0110] In this implementation, the storage module adopts a modular structure design, and the storage content includes target category, location, time, reconstruction error, and alarm information. At the same time, the abnormal target and normal target data are stored separately through classification management technology to facilitate subsequent analysis and model optimization. The sparse feature dictionary is dynamically updated in the storage module to adapt to the newly added abnormal targets and improve the system detection performance.

[0111] The storage module supports long-term recording and classification management of data, generates risk reports and optimization suggestions through historical data analysis, and supports continuous improvement of regional safety monitoring and resource scheduling.

[0112] The system of this embodiment can be used to execute the above method embodiments, and its principles and technical effects are similar, which will not be repeated here.

[0113] Although embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions and variations may be made to the embodiments without departing from the principles and spirit of the present invention, and that the scope of the present invention is defined by the appended claims and their equivalents.

Claims

1. A regional safety monitoring method based on infrared images, characterized in that: The following steps are involved: Collect infrared images of the target area through infrared camera equipment; De-noising and detail enhancement processing are performed on the collected infrared image to obtain a denoised image; Perform frequency domain conversion on the denoised image to extract high-frequency features of the target area; The sparse representation model is used to detect anomalies in the high-frequency features of the target area to determine whether the target is abnormal; Output alarm signals or trigger linkage response equipment based on abnormal detection results; Store, analyze and process abnormal targets and related monitoring data.

2. The regional safety monitoring method based on infrared images according to claim 1 is characterized in that: The infrared camera equipment acquisition includes the following steps: Use infrared camera equipment to collect data of the target area in real time. The infrared camera equipment includes passive infrared sensors and active infrared sensors. Quantify and digitally convert the collected original infrared signals, and convert the infrared radiation intensity into digital image data; The clarity and real-time performance of image acquisition can be adjusted by setting the sampling frequency and resolution parameters. The sampling frequency is adjusted according to the dynamics of the monitored area, and the resolution is achieved by adjusting the sensor pixel parameters. The collected data is preprocessed in real time, including removing background interference signals and noise during the collection process to ensure the effectiveness of the infrared image.

3. The regional safety monitoring method based on infrared images according to claim 1 is characterized in that: The denoising and detail enhancement process uses a nonlinear partial differential equation model to process the image. The expression of the partial differential equation is: ； in, Represents the image gradient, reflecting the rate of change of image pixels; is the diffusion control function; For the fidelity item; Represents the acquired original infrared image.

4. The regional safety monitoring method based on infrared images according to claim 3 is characterized in that: The denoising and detail enhancement process is achieved by minimizing the following energy functional: ； in, is the total energy of image processing; Item 1 Represents the smoothing energy of the image; is the diffusion coefficient control function; Item 2 Represents the fidelity energy of the image; Represents the deviation between the pixel value of the processed image and the pixel value of the original image; is the weight parameter of the fidelity term.

5. The regional safety monitoring method based on infrared images according to claim 1 is characterized in that: The frequency domain conversion of the denoised image to extract the high-frequency features of the target area comprises the following steps: Perform Fourier transform on the image data after denoising and detail enhancement to convert the image from time domain representation to frequency domain representation; Apply a bandpass filter to the frequency domain image for processing. The bandpass filter retains the frequency range of the target features in the frequency domain and suppresses other frequency components; Perform inverse Fourier transform on the frequency domain data processed by the bandpass filter to restore the data from frequency domain representation to time domain representation; The image reconstructed by inverse Fourier transform is used as the target enhanced image for the subsequent anomaly detection step.

6. The regional safety monitoring method based on infrared images according to claim 1 is characterized in that: The sparse representation model represents the features of the target area as a sparse linear combination of feature dictionaries. The specific expression is: ； in, Indicates the characteristics of the target area; Represents a feature dictionary; is a sparse coefficient vector; is the reconstruction error.

7. The regional safety monitoring method based on infrared images according to claim 1 is characterized in that: The anomaly detection solves the sparse coefficients through the following optimization problem: ； in, The observed feature vector representing the target area; represents the feature dictionary matrix; represents a sparse coefficient vector; represents the sum of squares of reconstruction errors; Represents the weight parameter of the sparse constraint term.

8. The regional safety monitoring method based on infrared images according to claim 1 is characterized in that: The abnormal detection result outputs an alarm signal or triggers a linkage response device including: Trigger local sound and light alarm equipment and generate alarm information including target category, location and time; Sending the alarm information to the user terminal via the remote communication module; According to the alarm information, control the external equipment to perform linkage operations, including the camera direction adjustment device, the access control system control device and the regional lighting control device; The alarm information and linkage results are stored in the local storage module for subsequent data analysis and response optimization.

9. The regional safety monitoring method based on infrared images according to claim 1 is characterized in that: The storage and analysis of abnormal targets and related monitoring data includes the following steps: The detected abnormal targets and their related data, including target category, location, time, reconstruction error and alarm information, are stored in the local storage module; Classify and manage the stored abnormal target data and normal target data to generate classified data sets; Utilize the stored abnormal target data to optimize the feature dictionary and enhance the accuracy of the sparse representation model; Generate historical detection record analysis reports through the data analysis module to evaluate the safety status of the monitored area; Output optimization suggestions and upload historical data to the remote management platform to support cross-regional security management and resource scheduling.

10. A regional safety monitoring system based on infrared images, applied to the regional safety monitoring method based on infrared images according to any one of claims 1 to 9, characterized in that: The regional safety monitoring system based on infrared images includes: Infrared image acquisition module, used for acquiring infrared images of target areas; Image processing module, used for denoising and detail enhancement of infrared images; The frequency domain analysis module is used to perform frequency domain conversion on the denoised image and extract target features; Anomaly detection module, used to detect abnormal targets based on target features based on sparse representation model; An alarm response module is used to output an alarm signal or trigger a linkage device after detecting an abnormal target; The storage module is used to store anomaly detection results and related data.

Images