Multi-source data analysis-based campus intelligent security and protection method

Abstract

The invention provides a campus intelligent security method based on multi-source data analysis, and relates to the technical field of multi-source data analysis, and the method comprises the steps: collecting a campus entrance image; performing image segmentation through a personnel image segmentation model to obtain a first target personnel image; performing personnel matching based on the campus personnel information, and when a personnel matching threshold is not met, determining that campus intrusion exists, and generating a first alarm instruction; receiving school structure information, executing an adaptive monitoring decision, configuring a monitoring time period, and virtualizing a monitoring fence; obtaining an infrared detection result; a signal difference threshold value is obtained, and when the signal difference reaches the signal difference threshold value, a second alarm instruction is generated; and according to the first alarm instruction and / or the second alarm instruction, an audible and visual alarm is started to carry out campus intrusion audible and visual alarm. The technical problems that a traditional intrusion detection method depends on a single data source, so that the adaptability to environment change and interference is poor, and the false alarm rate and the missing report rate are high are solved.

Description

Technical Field

[0001] The present invention relates to the technical field of multi-source data analysis, and in particular to a campus smart security method based on multi-source data analysis. Background Art

[0002] The campus smart security system is mainly used in campus security management of various educational institutions, such as universities, primary and secondary schools, vocational schools, etc. Traditional security systems usually rely on a single data source, such as camera monitoring. This reliance on a single data source makes the system prone to false alarms and missed alarms when facing complex environmental changes, such as light changes, occlusions, and bad weather, making it difficult to provide reliable security protection; in addition, the data processing and analysis efficiency of traditional systems is low, mainly relying on local processing, and it is difficult to achieve real-time response. When a security incident occurs, the system cannot quickly process and respond, resulting in security risks not being resolved in a timely manner. Summary of the invention

[0003] This application provides a campus smart security method based on multi-source data analysis, aiming to solve the technical problems that traditional intrusion detection methods rely on a single data source, resulting in poor adaptability to environmental changes and interference, and high false alarm and missed alarm rates.

[0004] The present application discloses a campus smart security method based on multi-source data analysis, the method is applied to the campus smart security system, the system is connected to the first monitoring unit, the second monitoring unit, the data processing center, and the sound and light alarm, the method includes: through the first monitoring unit, connecting the image collector to collect the campus entrance image, and synchronizing the campus entrance image to the data processing center; retrieving the campus personnel information through the data processing center, and through the personnel image segmentation model, performing image segmentation according to the first target person in the campus entrance image to obtain the first target person image; based on the campus personnel information, performing personnel matching on the first target person image, when the personnel matching result does not meet the personnel matching threshold, it is determined that there is a campus intrusion, and a first alarm instruction is generated; using the second monitoring unit to receive the campus structure information of the target campus, and based on the personnel matching ... image to obtain the first target person image; based on the campus personnel information, performing personnel matching on the first target person image, when the personnel matching result does not meet the personnel matching threshold, it is determined that there is a campus intrusion, and a first alarm instruction is generated; using the second monitoring unit to receive the campus structure information of the target campus, and based on the personnel matching model, performing image segmentation according to the first target person image to obtain the first target person image; based on the campus personnel information, performing personnel matching on the first target person image, when the personnel matching result does not meet the personnel matching threshold, it is determined that there is a campus intrusion, and a first alarm instruction is generated. Execute adaptive monitoring decisions according to the campus structure information, configure monitoring time periods and virtual monitoring fences; during the monitoring period, perform infrared detection on the virtual monitoring fence through multiple infrared detectors to obtain infrared detection results, and synchronize the infrared detection results to the data processing center, wherein the multiple infrared detectors have location identifiers, and the infrared detection results include infrared transmission signals emitted by an infrared signal transmission module and infrared reception signals received by an infrared signal reception module; obtain a signal difference threshold through the data processing center, and when the signal difference between the infrared transmission signal and the infrared reception signal reaches the signal difference threshold, determine that there is a campus intrusion, and generate a second alarm instruction with the location identifier; according to the first alarm instruction and / or the second alarm instruction, start the sound and light alarm to issue a campus intrusion sound and light alarm.

[0005] One or more technical solutions provided in this application have at least the following technical effects or advantages:

[0006] The first monitoring unit and the second monitoring unit collect images and infrared signals to achieve multi-source data fusion and enhance the detection accuracy of the system; the image segmentation model is used to process the campus entrance image, accurately extract the target person image, and improve the recognition accuracy; the personnel matching image processor based on the twin network improves the accuracy of personnel matching through deep learning and reduces the false alarm and missed alarm rate; the data processing center centrally processes various monitoring data to improve the processing efficiency and response speed of the system and realize real-time intrusion detection and alarm; the integrated sound and light alarm triggers the real-time alarm through the first alarm instruction and the second alarm instruction to improve the speed and efficiency of emergency response; the monitoring period and virtual monitoring fence are dynamically configured according to the campus structure information, the monitoring strategy is flexibly adjusted, and the monitoring equipment and monitoring strategy are automatically configured through adaptive monitoring decision-making to achieve efficient and intelligent campus safety management. Through the above methods, the accuracy, real-time response capability and environmental adaptability of the campus smart security system are improved, the technical problems existing in the traditional security system are solved, and more efficient and intelligent campus safety management is achieved.

[0007] The above description is only an overview of the technical solution of the present application. In order to more clearly understand the technical means of the present application, it can be implemented in accordance with the contents of the specification. In order to make the above and other purposes, features and advantages of the present application more obvious and easy to understand, the specific implementation methods of the present application are listed below. BRIEF DESCRIPTION OF THE DRAWINGS

[0008] Figure 1 A schematic diagram of a campus smart security method flow based on multi-source data analysis is provided for the embodiment of the present application;

[0009] Figure 2 A schematic diagram of the process of obtaining the first target person image in the campus smart security method based on multi-source data analysis is provided for an embodiment of the present application. DETAILED DESCRIPTION

[0010] The embodiment of the present application solves the technical problem that the traditional intrusion detection method relies on a single data source, resulting in poor adaptability to environmental changes and interference, and high false alarm and missed alarm rates, by providing a campus smart security method based on multi-source data analysis.

[0011] After introducing the basic principles of the present application, various non-limiting implementation methods of the present application will be specifically described below in conjunction with the drawings in the specification.

[0012] like Figure 1 As shown, the embodiment of the present application provides a campus smart security method based on multi-source data analysis, the method is applied to a campus smart security system, the system is connected to a first monitoring unit, a second monitoring unit, a data processing center, and an audible and visual alarm, the method includes:

[0013] Through the first monitoring unit, connect the image collector to collect the campus entrance image, and synchronize the campus entrance image to the data processing center;

[0014] The campus smart security method based on multi-source data analysis provided in the embodiment of the present application is applied to the campus smart security system, and the system is communicatively connected with the first monitoring unit, the second monitoring unit, the data processing center, and the sound and light alarm. The campus smart security system refers to a system that uses modern information technology, data analysis and intelligent equipment to improve the efficiency and effectiveness of campus security management, and can monitor the security status within the campus in real time and promptly alarm when an abnormality occurs.

[0015] The first monitoring unit is a submodule in the system, which is used to monitor and collect image data of the campus entrance to monitor the situation of people entering and leaving. The second monitoring unit is another submodule in the system, which can configure monitoring time periods and virtual monitoring fences, and perform specific monitoring tasks through infrared detectors.

[0016] The data processing center is the core part of the system, responsible for receiving, storing, processing and analyzing the data transmitted from each monitoring unit to determine whether there is a security threat. The sound and light alarm is a device used to sound an alarm. When the system detects an intrusion, the sound and light alarm is activated to alert people on the campus through sound and light signals. These components work together to form a complete campus smart security system to achieve comprehensive and real-time security monitoring.

[0017] First, according to the specific environment of the campus entrance, select an image collector, which can be a camera with high resolution and wide-angle lens to ensure that the entire entrance area can be covered. The camera is installed in a prominent position at the campus entrance, such as above the gate or on the wall next to it, so that the camera can clearly capture the facial features of all people entering and leaving. After the camera is started, it begins to collect image data of the campus entrance in real time, and transmits the collected campus entrance images to the data processing center in real time via a network connection.

[0018] The campus personnel information is retrieved through the data processing center, and the image segmentation model is used to perform image segmentation according to the first target person in the campus entrance image to obtain the first target person image.

[0019] The data processing center stores information about all registered personnel on campus, including photos, names, student numbers / work numbers, etc. The personnel information on campus is retrieved at the data processing center. The personnel image segmentation model is a pre-built semantic segmentation model, which is trained with a large amount of annotated data and is used to segment personnel from the background in an image.

[0020] In the campus entrance image, the target detection algorithm is used to identify the people in the image, and the first target person is randomly selected and the bounding box of the first target person is determined. The bounding box area image of the first target person is input into the person image segmentation model. The model processes the input image and outputs a binary mask image, in which the white area represents the target person and the black area represents the background. The mask image is combined with the original image to generate a segmented image containing only the first target person, that is, the first target person image.

[0021] Furthermore, if Figure 2 As shown, by using the personnel image segmentation model, image segmentation is performed according to the first target person in the campus entrance image to obtain the first target person image, including:

[0022] When a person enters the campus entrance of the target campus, an image of the campus entrance is collected, wherein the image of the campus entrance includes the first target person;

[0023] Obtain a set of sample campus entrance images and a set of sample personnel image segmentation results after image segmentation;

[0024] Based on semantic segmentation, an encoder and a decoder are constructed, and the encoder and the decoder are trained using the sample campus entrance image set and the sample personnel image segmentation result set to obtain a personnel image segmentation model;

[0025] The campus entrance image is input into the personnel image segmentation model to perform image segmentation of the first target person to obtain the first target person image.

[0026] A high-definition camera is installed at the campus entrance to ensure coverage of the entire entrance area. The camera monitors the entrance area in real time. When a person enters the campus entrance, the camera is triggered to capture images. The captured image includes a panoramic view of the entrance, which includes multiple people entering, from which one is randomly selected as the first target person.

[0027] Collect a large number of campus entrance images from historical records or specific collection activities. These images should cover scenes under different time, weather and lighting conditions. The sample image collection contains various features such as different people's postures, clothing, backpacks, etc.

[0028] With the help of professional image annotation tools, sample images are annotated to generate corresponding person segmentation masks to form a sample image segmentation result set. The sample image segmentation result set contains the precise boundaries of people in each image for model training.

[0029] Select the semantic segmentation model architecture, which includes two parts: encoder and decoder. The encoder is used to extract the features of the input image, gradually compress the spatial resolution through multi-layer convolution and pooling operations, and extract high-level semantic features; the decoder is used to gradually upsample the features extracted by the encoder, restore the spatial resolution, and generate a segmentation mask.

[0030] The sample campus entrance image set and the corresponding personnel image segmentation result set are input into the model for training. The model parameters are optimized using the cross entropy loss function. The optimization algorithms such as stochastic gradient descent are used for iterative training until the model shows good segmentation performance on the validation set, thus obtaining the personnel image segmentation model.

[0031] The campus entrance image collected in real time is input into the trained personnel image segmentation model. The encoder part of the model extracts the features of the input image, and the decoder part generates the corresponding segmentation mask. The segmentation mask output by the model separates the people in the image from the background. The personnel part in the mask is the foreground, and the other part is the background. According to the segmentation mask, the image of the first target person is extracted from the original image to generate image data containing only the target person, that is, the first target person image.

[0032] Based on the campus personnel information, personnel matching is performed on the first target person image. When the personnel matching result does not meet the personnel matching threshold, it is determined that there is campus intrusion and a first alarm instruction is generated.

[0033] Based on the campus personnel information, the image data of all registered personnel are extracted, and a personnel matching image processor is built based on the twin network. The twin network consists of two neural networks that share the same weights and can effectively calculate the similarity between two images. The sample personnel image segmentation result set and the campus personnel image set, as well as the corresponding similarity labels are used to train the twin network. Through a large amount of training data, the twin network learns how to measure the similarity between images.

[0034] The first target person image and each campus person image pair extracted from the database are input into the twin network. The twin network outputs the similarity score between each pair of images, indicating the similarity between the two images in visual features. All matching results are sorted according to the similarity score, and the result with the highest similarity is selected as the candidate matching person. A similarity threshold is set. If the highest similarity is lower than the threshold, the match is considered to have failed and it is determined that there is a campus intrusion. In this case, the first alarm instruction is generated, and the alarm information is transmitted to the sound and light alarm and related equipment for real-time alarm.

[0035] Furthermore, based on the campus personnel information, performing personnel matching on the first target personnel image includes:

[0036] Extracting a set of campus personnel images based on the campus personnel information;

[0037] Based on the twin network, a person matching image processor is constructed;

[0038] Based on the personnel matching image processor, similarity recognition is performed between the first target personnel image and the set of campus personnel images to obtain a maximum similarity as a personnel matching result.

[0039] The image data of all registered personnel on campus are extracted from the personnel information database of the data processing center to form a set of campus personnel images. The extracted personnel images are preprocessed, including size normalization, color adjustment, contrast enhancement, etc., to ensure image consistency.

[0040] The twin network consists of two convolutional neural networks (CNNs) with shared weights, which are used to calculate the similarity between two input images. The input images are subjected to feature extraction through these two branches respectively, and the output of each branch is a feature vector representing the high-level features of the input image.

[0041] The first target person image collected in real time is subjected to the same preprocessing steps as the campus personnel image to ensure consistency. The first target person image and each image in the campus personnel image set are respectively input into the two branches of the twin network, and the feature vector is extracted. The similarity between the feature vectors of the first target person image and each campus personnel image is calculated. Commonly used similarity measurement methods include Euclidean distance, cosine similarity, etc. From all the calculated similarities, the campus personnel image corresponding to the maximum similarity is selected as the personnel matching result, and the maximum similarity is compared with the preset matching threshold. If the maximum similarity value is lower than the threshold, it is considered that the match has failed, and there may be unauthorized intrusion.

[0042] Furthermore, based on the twin network, a person matching image processor is constructed, including:

[0043] Obtain a sample similarity set based on the sample personnel image segmentation result set and the campus personnel image set;

[0044] Based on the Siamese network, a person matching image processor including a first convolution channel and a second convolution channel is constructed;

[0045] The personnel matching image processor is trained using the sample personnel image segmentation result set, the campus personnel image set, and the sample similarity set until convergence.

[0046] Each image in the sample personnel image segmentation result set is paired with each image in the campus personnel image set to form an image pair. A similarity measurement method, such as Euclidean distance and cosine similarity, is used to calculate the similarity score of each pair of images to generate a sample similarity set.

[0047] Build a twin network architecture, including the first convolution channel and the second convolution channel. The two input layers receive two input images, that is, image pairs. The weights of the two channels are shared to ensure that the input image pairs use the same filter during feature extraction. The output of each channel is the feature vector of the image, which represents the representation of the image in the feature space. By calculating the similarity between the two feature vectors, the image similarity is output.

[0048] Using the sample personnel image segmentation result set and the campus personnel image set, we construct positive sample pairs (different images of the same person) and negative sample pairs (images of different people). We use the calculated similarity scores as training labels to guide the model to learn the similarity relationship between image pairs. We input the positive and negative sample pairs into the two channels of the twin network, extract the feature vectors respectively, and use the contrast loss function to calculate the loss between the model output and the true similarity label. We encourage the feature vectors of positive sample pairs to be close and the feature vectors of negative sample pairs to be far away. We use optimization algorithms such as stochastic gradient descent to update the model parameters, minimize the loss function, and repeat the training process until the loss function converges.

[0049] The second monitoring unit receives campus structure information of the target campus, and performs adaptive monitoring decisions according to the campus structure information, and configures monitoring time periods and virtual monitoring fences.

[0050] The second monitoring unit receives detailed structural information of the campus from the data processing center, including campus maps, building layouts, road distribution, etc. It analyzes the received campus structural information and identifies key areas and potential safety hazard areas, such as entrances, exits, surrounding walls, and areas around important buildings. Based on the analysis results, it formulates specific monitoring strategies, including priority monitoring areas, monitoring frequencies, and the layout of monitoring equipment.

[0051] The monitoring period is planned according to the daily activities of the campus, such as class time, get out of class time, night security period, etc., to ensure high-frequency monitoring during key periods.

[0052] Based on the campus structure information, virtual monitoring fences are defined, that is, virtual boundaries are set in specific areas. These fences can be closed polygonal areas, covering the key entrances and exits, surrounding walls and important internal areas of the campus. Attributes are set for each virtual monitoring fence, including alarm priority and monitoring equipment, such as the deployment plan of infrared sensors. Multiple infrared detectors are arranged inside and outside the virtual monitoring fence to ensure full coverage, and each detector has a location identifier.

[0053] During the monitoring period, infrared detection is performed on the virtual monitoring fence through multiple infrared detectors to obtain infrared detection results, and the infrared detection results are synchronized to the data processing center, wherein the multiple infrared detectors have position identifications, and the infrared detection results include infrared transmission signals emitted by infrared signal transmission modules and infrared reception signals received by infrared signal reception modules.

[0054] During the predetermined monitoring period, the infrared detector is automatically activated to start infrared detection of the virtual monitoring fence. The infrared detector consists of an infrared signal transmitting module and a receiving module. The infrared signal transmitting module sends infrared signals according to the set frequency and power. These signals are invisible light and can penetrate some obstacles, but will be reflected or blocked when encountering human bodies or other objects. The infrared signal receiving module receives the infrared signals from the transmitting module in real time and measures parameters such as signal strength and arrival time. The infrared transmitting signal and the infrared receiving signal constitute the infrared detection result, which is transmitted to the data processing center in real time via a wireless or wired network.

[0055] A signal difference threshold is obtained through the data processing center. When the signal difference between the infrared transmitting signal and the infrared receiving signal reaches the signal difference threshold, it is determined that there is a campus intrusion and a second alarm instruction with the location identifier is generated.

[0056] The data processing center retrieves the historical detection records of the infrared detectors, including the historical infrared transmission signal timing and the historical infrared reception signal timing, calculates the difference between the historical infrared transmission signal and the historical reception signal, and counts these differences to determine the signal fluctuation range under normal circumstances. Based on the statistical results, a reasonable signal difference threshold is set to distinguish normal situations from abnormal intrusion situations.

[0057] After receiving the real-time signal data, the data processing center calculates the difference between the infrared transmission signal and the received signal at the current moment, and compares the calculated real-time signal difference with the pre-set signal difference threshold. When the real-time signal difference reaches or exceeds the signal difference threshold, it is determined that an abnormal situation exists, that is, unauthorized personnel or objects may have passed through the infrared monitoring area. According to the position identifier of the infrared detector that triggers the signal difference, the specific location of the abnormal situation is determined, and a second alarm instruction with the position identifier is generated, indicating the specific abnormal area and detector position.

[0058] According to the first alarm instruction and / or the second alarm instruction, the sound and light alarm device is activated to issue a sound and light alarm for campus intrusion.

[0059] The sound and light alarm is usually installed in multiple locations on the campus to ensure that the sound and light signals can cover all important areas. The configuration includes the volume of the sound, the brightness and frequency of the light signal, etc. The first alarm command and / or the second alarm command generated by the data processing center is transmitted to the sound and light alarm through the internal network. After receiving the alarm command, the sound and light alarm immediately sends out a high-decibel alarm and a strong flash signal to alert the personnel on the campus to the abnormal situation. The sound and light alarm lasts for a period of time or until the security personnel confirm and manually cancel the alarm.

[0060] Further, obtaining the signal difference threshold includes:

[0061] Retrieving historical detection records of similar detectors of the infrared detector from the data processing center, and extracting historical campus intrusion records;

[0062] The historical campus intrusion records include historical infrared transmission signal timing and historical infrared reception signal timing;

[0063] Compare the historical infrared transmission signal timing and the historical infrared reception signal timing to calculate and obtain the historical signal difference;

[0064] The signal difference threshold is set according to the historical signal difference value.

[0065] The historical detection records of similar detectors of infrared detectors are queried from the database of the data processing center, known campus intrusion events are marked in the historical detection records, and the historical campus intrusion record data is extracted. The extracted data includes the timing records of the infrared detector's transmission signal and reception signal. The extracted raw data is cleaned to remove invalid data and noise to ensure the accuracy of the data.

[0066] The historical infrared transmission signal timing and the reception signal timing are aligned to ensure that the transmission signal and the reception signal at the same time point can accurately correspond to each other, and the difference between the infrared transmission signal and the reception signal at each time point is calculated to generate a historical signal difference sequence.

[0067] Perform statistical analysis on the historical signal difference sequence, including calculating the average difference, variance, maximum value, minimum value and other statistics, to understand the distribution of the signal difference. According to the results of the statistical analysis, set a signal difference threshold, such as a fixed multiple, such as 3 times the standard deviation. In order to adapt to changes in different environments and conditions, the threshold can be dynamically adjusted according to real-time data.

[0068] Furthermore, the system is also in communication connection with the first electronic device and the second electronic device, and the method further includes:

[0069] When it is determined that there is a campus intrusion, push campus intrusion alarm information with a location identifier to the first electronic device and the second electronic device, wherein the first electronic device is the electronic device of the target monitoring user and the second electronic device is the electronic device of the target protection user;

[0070] The target supervision user performs dispatch management of security personnel according to the location identifier, and the target protection user performs autonomous evacuation according to the location identifier.

[0071] When the system determines that there is a campus intrusion, it immediately generates campus intrusion alarm information, which includes the specific location identifier, event type, timestamp and necessary detailed description of the intrusion event. The alarm information is pushed to the first electronic device and the second electronic device in real time. The first electronic device is the target supervision user, such as the security supervisor's device, which can be a mobile phone, tablet computer or computer; the second electronic device is the target protection user, such as the device of a teacher or student at school, usually their mobile phone or emergency contact device provided by the school.

[0072] The target monitoring user receives the alarm information on the first electronic device, checks the specific intrusion location and detailed information, and according to the intrusion location identifier, dispatches the nearest security personnel to the scene for processing through the internal communication system, such as walkie-talkie or dedicated security communication software.

[0073] The target protection user receives the alarm information on the second electronic device, understands the intrusion location and emergency situation, and the target protection user evacuates autonomously according to the intrusion location, avoids going to the dangerous area, and reports his or her safety status to the safe area in time.

[0074] Furthermore, the system is also in communication connection with a third electronic device, and the method further includes:

[0075] Constructing a first feedback constraint time;

[0076] After the campus intrusion alarm information is pushed to the first electronic device, a timing is started. When the first feedback constraint time is reached and no feedback information from the first electronic device is received, the campus intrusion alarm information is pushed to the third electronic device, wherein the third electronic device is an electronic device of an associated user of the target supervised user.

[0077] According to the campus safety policy and emergency handling process, set a reasonable feedback time as the first feedback constraint time. This time can be from a few minutes to more than ten minutes. The specific length of time is determined by the size of the campus and the response speed of the security personnel.

[0078] A timer function is embedded in the system. When the alarm information is pushed to the first electronic device, the timer starts automatically, starts calculating the feedback time, and continuously monitors the feedback status of the first electronic device. If feedback is received from the target supervisory user within the first feedback constraint time, the feedback time is recorded and the timing is stopped. If the system still has not received feedback information from the target supervisory user after the first feedback constraint time is reached, it is determined that the first alarm was not successfully confirmed or the target supervisory user did not respond in time, and the alarm information is pushed to the third electronic device, that is, the electronic device of an associated user of the target supervisory user, such as a superior supervisor or other security person in charge, to ensure that the alarm information can be processed in time.

[0079] Furthermore, it also includes:

[0080] Obtaining intrusion feedback information of the campus intrusion sound and light alarm;

[0081] The detection feature of the campus intrusion sound and light alarm is optimized through the intrusion feedback information, and the personnel matching threshold and the signal difference threshold are optimized and corrected through the detection feature optimization result.

[0082] After responding to the alarm, the security personnel will feedback the processing results through mobile devices or monitoring center systems, including the on-site situation, whether it is confirmed as an intrusion, and the processing measures, etc., obtain intrusion feedback information, and store it in the database of the data processing center.

[0083] Detection features are extracted from the feedback information, such as the frequency of alarm triggering, false alarm rate, response time, conditions for confirming intrusion, etc. Machine learning algorithms, such as decision trees, random forests, and support vector machines, are used to optimize the detection features and identify the key features that can best distinguish normal situations from intrusion events. Based on the optimized features, the personnel matching threshold and signal difference threshold are adjusted, and the personnel matching accuracy and false alarm rate under different thresholds are compared. The optimal threshold is selected to reduce false alarms and missed alarms, ensuring that the system can maintain efficient detection performance under different circumstances.

[0084] In summary, the campus smart security method based on multi-source data analysis provided by the embodiment of the present application has the following technical effects:

[0085] The first monitoring unit and the second monitoring unit collect images and infrared signals to achieve multi-source data fusion and enhance the detection accuracy of the system; the image segmentation model is used to process the campus entrance image, accurately extract the target person image, and improve the recognition accuracy; the personnel matching image processor based on the twin network improves the accuracy of personnel matching through deep learning and reduces the false alarm and missed alarm rate; the data processing center centrally processes various monitoring data to improve the processing efficiency and response speed of the system and realize real-time intrusion detection and alarm; the integrated sound and light alarm triggers the real-time alarm through the first alarm instruction and the second alarm instruction to improve the speed and efficiency of emergency response; the monitoring period and virtual monitoring fence are dynamically configured according to the campus structure information, the monitoring strategy is flexibly adjusted, and the monitoring equipment and monitoring strategy are automatically configured through adaptive monitoring decision-making to achieve efficient and intelligent campus safety management. Through the above methods, the accuracy, real-time response capability and environmental adaptability of the campus smart security system are improved, the technical problems existing in the traditional security system are solved, and more efficient and intelligent campus safety management is achieved.

[0086] The above description of the disclosed embodiments enables those skilled in the art to implement or use the present application. Various modifications to these embodiments will be apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the present application. Therefore, the present application will not be limited to the embodiments shown herein, but will conform to the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A campus smart security method based on multi-source data analysis, characterized in that: The method is applied to a campus smart security system, the system is connected to a first monitoring unit, a second monitoring unit, a data processing center, and an audible and visual alarm, and the method includes: Through the first monitoring unit, connect the image collector to collect the campus entrance image, and synchronize the campus entrance image to the data processing center; Retrieving campus personnel information through the data processing center, and performing image segmentation according to the first target person in the campus entrance image through the personnel image segmentation model to obtain the first target person image; Based on the campus personnel information, performing personnel matching on the first target person image, when the personnel matching result does not meet the personnel matching threshold, determining that there is campus intrusion, and generating a first alarm instruction; The second monitoring unit receives campus structure information of the target campus, and performs adaptive monitoring decision according to the campus structure information, and configures monitoring time periods and virtual monitoring fences; During the monitoring period, the virtual monitoring fence is infrared detected by multiple infrared detectors to obtain infrared detection results, and the infrared detection results are synchronized to the data processing center, wherein the multiple infrared detectors have position identifiers, and the infrared detection results include infrared transmission signals emitted by infrared signal transmission modules and infrared reception signals received by infrared signal reception modules; A signal difference threshold is obtained through the data processing center, and when the signal difference between the infrared transmission signal and the infrared reception signal reaches the signal difference threshold, it is determined that there is a campus intrusion, and a second alarm instruction with the location identifier is generated; According to the first alarm instruction and / or the second alarm instruction, the sound and light alarm device is activated to issue a sound and light alarm for campus intrusion.

2. The method according to claim 1, characterized in that By using a personnel image segmentation model, image segmentation is performed according to a first target person in the campus entrance image to obtain a first target person image, including: When a person enters the campus entrance of the target campus, an image of the campus entrance is collected, wherein the image of the campus entrance includes the first target person; Obtain a set of sample campus entrance images and a set of sample personnel image segmentation results after image segmentation; Based on semantic segmentation, an encoder and a decoder are constructed, and the encoder and the decoder are trained using the sample campus entrance image set and the sample personnel image segmentation result set to obtain a personnel image segmentation model; The campus entrance image is input into the personnel image segmentation model to perform image segmentation of the first target person to obtain the first target person image.

3. The method according to claim 1, characterized in that Based on the campus personnel information, performing personnel matching on the first target personnel image includes: Extracting a set of campus personnel images based on the campus personnel information; Based on the twin network, a person matching image processor is constructed; Based on the personnel matching image processor, similarity recognition is performed between the first target personnel image and the set of campus personnel images to obtain a maximum similarity as a personnel matching result.

4. The method according to claim 3, characterized in that Based on the twin network, a person matching image processor is built, including: Obtain a sample similarity set based on the sample personnel image segmentation result set and the campus personnel image set; Based on the Siamese network, a person matching image processor including a first convolution channel and a second convolution channel is constructed; The personnel matching image processor is trained using the sample personnel image segmentation result set, the campus personnel image set, and the sample similarity set until convergence.

5. The method according to claim 1, characterized in that Get the signal difference threshold, including: Retrieving historical detection records of similar detectors of the infrared detector from the data processing center, and extracting historical campus intrusion records; The historical campus intrusion records include historical infrared transmission signal timing and historical infrared reception signal timing; Compare the historical infrared transmission signal timing and the historical infrared reception signal timing to calculate and obtain the historical signal difference; The signal difference threshold is set according to the historical signal difference value.

6. The method according to claim 1, characterized in that The system is also connected to the first electronic device and the second electronic device for communication, and the method further includes: When it is determined that there is a campus intrusion, push campus intrusion alarm information with a location identifier to the first electronic device and the second electronic device, wherein the first electronic device is the electronic device of the target monitoring user and the second electronic device is the electronic device of the target protection user; The target supervision user performs dispatch management of security personnel according to the location identifier, and the target protection user performs autonomous evacuation according to the location identifier.

7. The method according to claim 6, characterized in that The system is also connected to a third electronic device for communication, and the method further includes: Constructing a first feedback constraint time; After the campus intrusion alarm information is pushed to the first electronic device, a timing is started. When the first feedback constraint time is reached and no feedback information from the first electronic device is received, the campus intrusion alarm information is pushed to the third electronic device, wherein the third electronic device is an electronic device of an associated user of the target supervised user.

8. The method according to claim 1, characterized in that Also includes: Obtaining intrusion feedback information of the campus intrusion sound and light alarm; The detection feature of the campus intrusion sound and light alarm is optimized through the intrusion feedback information, and the personnel matching threshold and the signal difference threshold are optimized and corrected through the detection feature optimization result.

Images