Campus safety management method and system

Abstract

The present invention discloses a campus security management method and system, the method comprising: obtaining a point cloud map of a target campus and a point cloud map of other campuses in the global jurisdiction of the target campus; determining multiple additional patrol areas based on historical security incidents occurring in the other campuses; calculating a set of point cloud map feature descriptors for the target campus and a set of point cloud map feature descriptors for the other campuses, and determining a transformation matrix of the two sets of point cloud map feature descriptors; aligning multiple similar areas in the target campus based on the multiple additional patrol areas and the transformation matrix; allocating security personnel in corresponding proportions based on the frequency of historical security incidents occurring in the multiple similar areas corresponding to the multiple additional patrol areas. The present invention achieves more efficient and intelligent campus security management through data-driven prediction, precise resource allocation, real-time behavior recognition, and a flexible dynamic adjustment mechanism.

Description

Technical Field

[0001] The present invention relates to the field of campus monitoring, and in particular to a campus safety management method and system. Background Art

[0002] In traditional campus safety management, identifying and predicting potential safety hazards often relies on historical data, human judgment, or simple statistical analysis. This approach has the following problems:

[0003] First, methods based on human judgment are easily influenced by personal experience and bias, resulting in less than objective predictions. Second, traditional methods often fail to accurately locate high-risk areas, leading to irrational resource allocation, potentially overprotecting some areas and underprotecting others. Furthermore, security personnel allocation is typically fixed and cannot be dynamically adjusted based on real-time conditions. This can result in understaffed security in certain areas during specific time periods, while wasting resources during other periods. Existing allocation mechanisms make it difficult to flexibly respond to emergencies or temporary increases in security needs. Summary of the Invention

[0004] The embodiment of the present invention provides a campus security management method and system, which uses automatic identification of high-risk areas based on historical security incident data from other campuses, and continuously obtains security personnel location information, and combines behavior recognition models to monitor and adjust security forces in real time.

[0005] To achieve the above objectives, a first aspect of an embodiment of the present application provides a campus safety management method, comprising:

[0006] Obtaining a point cloud map of the target campus and point cloud maps of other campuses in the global jurisdiction where the target campus is located;

[0007] Based on historical security incidents that occurred on other campuses, multiple additional patrol areas were identified;

[0008] Calculating a set of point cloud feature descriptors for the target campus and a set of point cloud feature descriptors for the other campuses, and determining a transformation matrix for the two sets of point cloud feature descriptors;

[0009] Aligning a plurality of similar areas in the target campus according to the plurality of additional patrol areas and the transformation matrix;

[0010] Allocate security personnel in a corresponding proportion according to the frequency of historical security incidents occurring in the plurality of similar areas corresponding to the plurality of additional patrol areas; each security personnel is responsible for one of the similar areas;

[0011] Continuously obtain the position vector corresponding to the location of the security personnel in the point cloud map of the target campus. When the preset recognition model identifies the person to be managed, calculate the distance between the position vector and the position vector corresponding to the position of the person to be managed in the point cloud map of the target campus. According to the distance and the safety level of the person to be managed, assign a preset number of the security personnel to the location of the person to be managed.

[0012] In a possible implementation of the first aspect, obtaining a point cloud map of the target campus and point cloud maps of other campuses in the global jurisdiction where the target campus is located specifically includes:

[0013] For the target campus and other campuses in the entire jurisdiction where the target campus is located, multiple scanning stations are set according to a preset overlap range percentage; the preset overlap range percentage is equal to the overlap percentage of the patrol ranges of two adjacent security personnel, the preset overlap range percentage is the ratio of the overlapping scanning area of ​​adjacent scanning stations to the scanning area of ​​each of the scanning stations, and the patrol range overlap percentage is the ratio of the overlapping patrol area of ​​two adjacent security personnel to the scanning area of ​​each of the security personnel.

[0014] In a possible implementation of the first aspect, determining multiple additional patrol areas based on historical security incidents occurring on other campuses specifically includes:

[0015] Performing K-means clustering on the historical security events that occurred on the other campuses, and using multiple locations corresponding to the most recent historical security events in multiple clusters as the centers of multiple additional patrol areas;

[0016] A plurality of additional patrol areas are determined according to a preset patrol range and the centers of the plurality of additional patrol areas.

[0017] In a possible implementation of the first aspect, calculating a set of point cloud feature descriptors for the target campus and a set of point cloud feature descriptors for the other campuses, and determining a transformation matrix of the two sets of point cloud feature descriptors specifically includes:

[0018] Calculating the FPFH and SHOT of the target campus point cloud image as a set of point cloud feature descriptors, and calculating the FPFH and SHOT of the other campus point cloud images as another set of point cloud feature descriptors;

[0019] Based on the two sets of point cloud feature descriptors, the nearest neighbor algorithm is used to find all valid nearest neighbor point pairs between the point cloud of the target campus and the point cloud of other campuses.

[0020] Based on all the valid nearest neighbor point pairs, the transformation matrices of the two sets of point cloud feature descriptors are calculated.

[0021] In a possible implementation of the first aspect, finding all valid nearest neighbor point pairs between the point cloud image of the target campus and the point cloud images of other campuses using a nearest neighbor algorithm based on the two sets of point cloud image feature descriptors specifically includes:

[0022] Calculate the mean projection distance of the FPFH feature descriptor of each point in the other campus point cloud map on the corresponding point fitness vector as the FPFH distance;

[0023] Calculate the average of the SHOT feature descriptor and the corresponding point fitness vector projection distance of each point in the other campus point cloud map as the SHOT distance;

[0024] Using the difference between the FPFH distance and the SHOT distance as a distance threshold;

[0025] Use the nearest neighbor algorithm to find all valid nearest neighbor point pairs between the point cloud of the target campus and the point cloud of other campuses;

[0026] The nearest neighbor point pairs whose distance is less than or equal to the distance threshold are regarded as valid nearest neighbor point pairs.

[0027] In a possible implementation of the first aspect, before using a nearest neighbor algorithm to find all valid nearest neighbor point pairs between the point cloud image of the target campus and the point cloud images of other campuses based on the two sets of point cloud image feature descriptors, the method further includes:

[0028] A KD tree index structure is established based on two sets of point cloud feature descriptors.

[0029] In a possible implementation of the first aspect, aligning the plurality of similar areas in the target campus according to the plurality of additional patrol areas and the transformation matrix specifically includes:

[0030] According to the transformation matrix, coordinate transformation is performed on each point of the plurality of additional patrol areas to obtain a plurality of similar areas.

[0031] In a possible implementation of the first aspect, allocating security personnel in corresponding proportions according to the frequencies of historical security incidents occurring in the multiple similar areas corresponding to the multiple additional patrol areas specifically includes:

[0032] According to a first ratio of the frequency of historical security incidents occurring in each of the similar areas corresponding to the additional patrol area to the frequency of historical security incidents occurring in all the similar areas, security personnel in the same ratio as the first ratio are allocated to the corresponding similar areas.

[0033] In a possible implementation of the first aspect, when a preset recognition model identifies a person to be managed, calculating a distance between a position vector and a position vector corresponding to the person to be managed in a point cloud map of the target campus, and assigning, based on the distance and a safety level of the person to be managed, a number of security personnel matching the safety level to the location of the person to be managed, specifically includes:

[0034] Obtaining a database of registered personnel of the target campus;

[0035] In the preset C3D model, the weight of each camera is evaluated according to the Taylor expansion method for structured pruning;

[0036] The C3D model is used to perform behavior recognition on persons in the registered person database. If a person is recognized to have performed a target action, the person is designated as a first person to be managed. A first distance between a position vector and a position vector corresponding to the first person to be managed in the point cloud map of the target campus is calculated. Based on ascending order of the first distances and the security level corresponding to the target action, a number of security personnel matching the security level is assigned to the location of the person to be managed.

[0037] Add context variables including regional road direction information and regional location information to the preset VAEs model;

[0038] The VAEs model is called to perform behavior recognition on personnel in the non-registered personnel database. If a person is recognized to perform a target action, the person is regarded as a second person to be managed. The second distance between the position vector and the corresponding position vector of the second person to be managed in the point cloud map of the target campus is calculated. According to the ascending order of the second distances and the security level corresponding to the target action, a number of security personnel matching the security level are assigned to the location of the person to be managed.

[0039] A second aspect of the embodiments of the present application provides a campus safety management system, including:

[0040] A point cloud acquisition module is used to obtain a point cloud map of the target campus and point cloud maps of other campuses in the global jurisdiction where the target campus is located;

[0041] An area determination module, configured to determine a plurality of additional patrol areas based on historical security incidents occurring on other campuses;

[0042] a matrix determination module, configured to calculate a set of point cloud feature descriptors for the target campus and a set of point cloud feature descriptors for the other campuses, and determine a transformation matrix for the two sets of point cloud feature descriptors;

[0043] An area alignment module, configured to align a plurality of similar areas in the target campus according to the plurality of patrol-added areas and the transformation matrix;

[0044] A personnel allocation module is used to allocate security personnel in a corresponding proportion according to the frequency of historical security incidents occurring in the plurality of similar areas corresponding to the plurality of additional patrol areas; each security personnel is responsible for one of the similar areas;

[0045] The real-time allocation module is used to continuously obtain the position vector corresponding to the location of the security personnel in the point cloud map of the target campus. When the preset recognition model identifies the person to be managed, the distance between the position vector and the position vector corresponding to the person to be managed in the point cloud map of the target campus is calculated. According to the distance and the safety level of the person to be managed, a preset number of the security personnel are allocated to the location of the person to be managed.

[0046] Compared with existing technologies, the present invention ensures the comprehensiveness and accuracy of data by obtaining point cloud maps of the target campus and other campuses in its entire jurisdiction. It then automatically identifies high-risk areas and uses them as the center of the additional patrol area, avoiding the subjectivity and bias of human judgment and making the prediction more scientific and objective. By calculating feature descriptors, the feature information of the point cloud image is accurately extracted. Then, the nearest neighbor algorithm is used to find corresponding point pairs, and the transformation matrix between the two sets of point clouds is calculated based on these point pairs. This process achieves accurate registration of point cloud images between different campuses, providing a reliable foundation for subsequent security management. Then, based on the transformation matrix, the additional patrol areas in other campuses are mapped to the target campus to form similar areas. In this way, the successful experience of other campuses can be directly applied to quickly locate high-risk areas in the target campus. Finally, security personnel are allocated based on the ratio of the historical security incident frequency of similar areas to the additional patrol areas, ensuring that each area receives security forces that match its risk. This method avoids the problems of resource waste or insufficient protection and improves the effectiveness of resource allocation.

[0047] Furthermore, the present invention continuously acquires security personnel's location information and, in conjunction with a behavioral recognition model, monitors and adjusts security resources in real time. Once abnormal behavior is identified, the system rapidly calculates the location vector of the individual in question and assigns appropriate security personnel based on proximity and security level. This dynamic response mechanism enhances the system's flexibility and adaptability. BRIEF DESCRIPTION OF THE DRAWINGS

[0048] Figure 1 1 is a flow chart of a campus safety management method provided by one embodiment of the present invention;

[0049] Figure 2 The present invention provides a schematic structural diagram of a campus safety management system according to an embodiment of the present invention. DETAILED DESCRIPTION

[0050] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. All other embodiments obtained by ordinary technicians in this field based on the embodiments of the present invention without making any creative efforts shall fall within the scope of protection of the present invention.

[0051] To solve the above problem, see Figure 1 An embodiment of the present invention provides a campus safety management method, comprising:

[0052] S10. Obtain a point cloud map of the target campus and point cloud maps of other campuses in the entire jurisdiction where the target campus is located.

[0053] S11. Determine multiple additional patrol areas based on historical security incidents that occurred on other campuses.

[0054] S12. Calculate a set of point cloud feature descriptors for the target campus and a set of point cloud feature descriptors for the other campuses, and determine a transformation matrix of the two sets of point cloud feature descriptors.

[0055] S13. Aligning multiple similar areas in the target campus according to the multiple additional patrol areas and the transformation matrix.

[0056] S14. Allocate security personnel in corresponding proportions according to the frequencies of historical security incidents occurring in the plurality of similar areas corresponding to the plurality of increased patrol areas; each security personnel is responsible for one of the similar areas.

[0057] S15. Continuously obtain the position vector corresponding to the location of the security personnel in the point cloud map of the target campus. When the preset recognition model identifies the person to be managed, calculate the distance between the position vector and the position vector corresponding to the position of the person to be managed in the point cloud map of the target campus. According to the distance and the safety level of the person to be managed, assign a preset number of the security personnel to the location of the person to be managed.

[0058] S10 ensures data comprehensiveness and accuracy by acquiring point cloud images of the target campus and other campuses within its jurisdiction. This not only covers all key areas within the campus but also enables cross-campus analysis using data from other campuses, increasing the richness of the overall data. S11 automatically identifies high-risk areas by analyzing historical security incidents on other campuses and uses these as the center of additional patrol zones. Based on the actual patrol range and the center location of the additional patrol zone, the additional patrol zone can be flexibly adjusted to accommodate different campus layouts and security requirements. S12 accurately extracts feature information from the point cloud images by calculating feature descriptors. These feature descriptors not only incorporate spatial geometry but also consider local shape features, improving matching accuracy. A nearest neighbor algorithm is then used to identify corresponding point pairs, and a transformation matrix is ​​calculated based on these point pairs between the two point cloud images. This process enables precise registration of point cloud images across different campuses, providing a reliable foundation for subsequent security management. S13, based on the transformation matrix, maps the additional patrol zones on other campuses to the target campus, creating similarity zones. This allows for direct application of successful experiences from other campuses to quickly locate high-risk areas on the target campus. S14 allocates security personnel based on the historical frequency of security incidents in similar areas relative to the increased patrol areas, ensuring that each area receives security personnel commensurate with its risk. This approach avoids wasted resources or inadequate protection, improving the effectiveness of resource allocation. S15 continuously acquires security personnel's location information and, in conjunction with a behavioral recognition model, monitors and adjusts security personnel in real time. Once abnormal behavior is identified, the system quickly calculates the location vector of the individual being monitored and assigns appropriate security personnel based on proximity and security level. This dynamic response mechanism enhances the flexibility and adaptability of security management.

[0059] In summary, this method leverages historical security incident data and advanced point cloud processing technology to ensure accurate and scientific predictions, avoiding the subjectivity and uncertainty inherent in traditional methods. Through precise point cloud registration and alignment of similar regions, it enables data sharing and application across different campuses, improving overall management efficiency. Security personnel are allocated based on the frequency of historical incidents, combined with real-time monitoring and dynamic adjustment mechanisms to ensure the rational allocation and efficient use of resources. These combined advantages make this method significantly effective in improving the overall effectiveness of campus security management.

[0060] Exemplarily, obtaining a point cloud map of a target campus and point cloud maps of other campuses in the entire jurisdiction of the target campus specifically includes:

[0061] For the target campus and other campuses in the entire jurisdiction where the target campus is located, multiple scanning stations are set according to a preset overlap range percentage; the preset overlap range percentage is equal to the overlap percentage of the patrol ranges of two adjacent security personnel, the preset overlap range percentage is the ratio of the overlapping scanning area of ​​adjacent scanning stations to the scanning area of ​​each of the scanning stations, and the patrol range overlap percentage is the ratio of the overlapping patrol area of ​​two adjacent security personnel to the scanning area of ​​each of the security personnel.

[0062] In order to set up multiple scanning stations on the target campus and other campuses within its jurisdiction to ensure comprehensive coverage and obtain high-quality point cloud data, it is necessary to determine the preset overlap range percentage, select compliant scanning equipment, and plan the scanning station locations.

[0063] Let's assume that after analysis and testing, a 30% overlap percentage is chosen. This means that the overlap between two adjacent scanning stations should account for 30% of each station's total area. This percentage also applies to the overlap area when security personnel patrol, ensuring there are no blind spots.

[0064] Then, according to the specific environment of the campus (such as building density, vegetation cover, etc.), select a suitable laser scanner or other 3D scanning equipment. Considering cost and efficiency, it may be necessary to use portable or vehicle-mounted scanning equipment.

[0065] Finally, use GIS software or a professional scanning station planning tool to plan the optimal location of each scanning station based on the campus map and the preset overlap range percentage. This process needs to ensure sufficient overlap between scanning stations, but also avoid excessive overlap that will cause data redundancy.

[0066] This approach not only enables the acquisition of high-quality point cloud data, but also provides a solid foundation for campus safety management. The advantage of this approach is that it combines scientific planning with advanced technology to ensure data integrity and accuracy, thereby supporting more precise and effective safety management and decision-making.

[0067] For example, the method of determining multiple additional patrol areas based on historical security incidents occurring on other campuses specifically includes:

[0068] Performing K-means clustering on the historical security events that occurred on the other campuses, and using multiple locations corresponding to the most recent historical security events in multiple clusters as the centers of multiple additional patrol areas;

[0069] A plurality of additional patrol areas are determined according to a preset patrol range and the centers of the plurality of additional patrol areas.

[0070] The K-means clustering algorithm is used to perform cluster analysis on the location coordinates of all historical security events. Each cluster represents a group of security events with similar geographical locations.

[0071] The centroid (i.e., geometric center) of each cluster can be considered the center of a potential high-risk area. For each cluster, the nearest historical security incident location or locations are identified and used as the center of the additional patrol area corresponding to that cluster. These centers will serve as reference points for subsequent determination of additional patrol areas.

[0072] Then, based on the actual patrol capabilities of the security personnel and the campus layout, set a reasonable patrol range radius (for example, 100 meters). This range should ensure coverage of high-risk areas and their surrounding areas. Circular areas can be drawn with the center of each additional patrol area as the center of the circle using the preset patrol range radius. These circular areas are the additional patrol areas. If there is overlap between certain areas, the boundaries can be adjusted according to actual conditions to avoid repeated patrols or missing key areas. All additional patrol areas can be reviewed to ensure that they cover all high-risk points and there are no obvious blank areas. If necessary, the regional division can be further optimized in combination with geographic information system (GIS) tools, taking into account factors such as terrain and building distribution.

[0073] This method not only scientifically and rationally identifies additional patrol areas within the campus but also provides a solid foundation for subsequent security resource allocation. The advantage of this approach lies in its analysis based on actual historical data, which reduces the subjectivity and bias inherent in human judgment, improving prediction accuracy and the scientific nature of management decisions.

[0074] Exemplarily, calculating a set of point cloud feature descriptors for the target campus and a set of point cloud feature descriptors for the other campuses, and determining a transformation matrix of the two sets of point cloud feature descriptors, specifically includes:

[0075] Calculating the FPFH and SHOT of the target campus point cloud image as a set of point cloud feature descriptors, and calculating the FPFH and SHOT of the other campus point cloud images as another set of point cloud feature descriptors;

[0076] Based on the two sets of point cloud feature descriptors, the nearest neighbor algorithm is used to find all valid nearest neighbor point pairs between the point cloud of the target campus and the point cloud of other campuses.

[0077] Based on all the valid nearest neighbor point pairs, the transformation matrices of the two sets of point cloud feature descriptors are calculated.

[0078] First, for each reference campus' point cloud data, the FPFH algorithm is used to compute the local geometric features of each point. FPFH considers the distance and angular relationships between a point and its neighbors, effectively capturing local structural information in the point cloud. Similarly, for each reference campus' point cloud data, the SHOT algorithm is used to compute a feature descriptor for each point. SHOT not only considers geometric information but also includes information about the surface normal direction, making it suitable for point cloud registration tasks.

[0079] Next, a nearest neighbor search algorithm (such as the K-nearest neighbor algorithm (KNN)) is used to find the nearest neighbor points of each target campus point in the KD tree of the reference campus feature descriptors. This step will find one or more most similar reference campus points for each target campus point. Valid nearest neighbor pairs are screened based on a predefined distance threshold or other criteria. For example, a distance threshold can be set, and only when the distance between the nearest neighbors is less than or equal to the threshold are they considered a valid matching pair.

[0080] The coordinates of the target campus points and the corresponding reference campus points for all valid nearest neighbor pairs can be organized into two matrices. These two matrices represent the source point set (target campus points) and the target point set (reference campus points), respectively. Using algorithms such as SVD (Singular Value Decomposition) or RANSAC (Random Sample Consensus), the optimal rigid body transformation matrix (including rotation and translation) is calculated based on the source and target point sets. This transformation matrix can align the target campus point cloud with the reference campus point cloud.

[0081] This method achieves precise registration of point cloud images across different campuses, providing an important spatial analysis foundation for subsequent safety management strategies. The advantage of this approach lies in its combination of point cloud processing technology and data analysis methods, ensuring high accuracy and reliability of the registration results.

[0082] Exemplarily, the method of finding all valid nearest neighbor point pairs between the point cloud image of the target campus and the point cloud images of other campuses using a nearest neighbor algorithm based on the two sets of point cloud image feature descriptors specifically includes:

[0083] Calculate the mean projection distance of the FPFH feature descriptor of each point in the other campus point cloud map on the corresponding point fitness vector as the FPFH distance;

[0084] Calculate the average of the SHOT feature descriptor and the corresponding point fitness vector projection distance of each point in the other campus point cloud map as the SHOT distance;

[0085] Using the difference between the FPFH distance and the SHOT distance as a distance threshold;

[0086] Use the nearest neighbor algorithm to find all valid nearest neighbor point pairs between the point cloud of the target campus and the point cloud of other campuses;

[0087] The nearest neighbor point pairs whose distance is less than or equal to the distance threshold are regarded as valid nearest neighbor point pairs.

[0088] For each point in the cloud, calculate the difference between its FPFH distance and SHOT distance. This difference reflects the degree of difference between different feature descriptors. A global distance threshold can be determined by performing statistical analysis on the distance differences of all points (such as taking the average or setting a fixed ratio). For example, the average of the distance differences of all points can be selected as the threshold.

[0089] The K-nearest neighbor (KNN) algorithm can be used to find the nearest neighbor of each target campus point in the KD tree of the reference campus feature descriptor. The distance between each target campus point and its nearest neighbor is recorded. Based on a pre-set distance threshold, valid nearest neighbor pairs that meet the criteria are selected. Specifically, a pair of nearest neighbors is considered a valid match only if the distance between them is less than or equal to the global distance threshold.

[0090] This method not only achieves precise registration of point cloud images between different campuses, but also ensures the validity and accuracy of the matching point pairs. The advantage of this method lies in its combination of advanced point cloud processing technology and scientific data analysis methods, ensuring the high accuracy and reliability of the registration results, providing an important spatial analysis foundation for subsequent safety management strategies.

[0091] Exemplarily, before finding all valid nearest neighbor point pairs between the point cloud image of the target campus and the point cloud images of other campuses using a nearest neighbor algorithm based on the two sets of point cloud image feature descriptors, the method further includes:

[0092] A KD tree index structure is established based on two sets of point cloud feature descriptors.

[0093] A KD tree index structure can be constructed for the point cloud feature descriptors of the target campus and the reference campus (other campuses) respectively to accelerate the nearest neighbor search process.

[0094] Exemplarily, the method for aligning a plurality of similar areas in the target campus according to the plurality of additional patrol areas and the transformation matrix specifically includes:

[0095] According to the transformation matrix, coordinate transformation is performed on each point of the plurality of additional patrol areas to obtain a plurality of similar areas.

[0096] For each point in each additional patrol area, the transformation matrix is ​​used to transform its coordinates. The transformed coordinate points are then reassembled into new areas, which are similar areas on the target campus. For example, after the transformation, the additional patrol areas on the old campus are transformed into similar areas on the new campus.

[0097] This approach not only achieves precise alignment of additional patrol areas across different campuses but also provides a scientific basis for security management within the new campus. The advantage of this approach lies in its combination of advanced point cloud processing technology and spatial transformation methods, ensuring high accuracy and reliability in similar areas. This facilitates more precise deployment of security resources and improves the overall effectiveness of campus security management.

[0098] For example, allocating security personnel in corresponding proportions according to the frequencies of historical security incidents occurring in the plurality of similar areas corresponding to the plurality of additional patrol areas specifically includes:

[0099] According to a first ratio of the frequency of historical security incidents occurring in each of the similar areas corresponding to the additional patrol area to the frequency of historical security incidents occurring in all the similar areas, security personnel in the same ratio as the first ratio are allocated to the corresponding similar areas.

[0100] Assume that, through the aforementioned method, three similar areas have been identified on the target campus: Area A, Area B, and Area C. These areas correspond to the additional patrol areas on other campuses: Area D, Area E, and Area F. Assume that the following number of security incidents have occurred in the additional patrol areas on other campuses:

[0101] There were 50 incidents in area D, 30 incidents in area E, and 20 incidents in area F. The incident frequencies of each additional patrol area (area D, area E, and area F) accounted for 0.5, 0.3, and 0.2 of the total incidents, respectively.

[0102] Assume that the target campus plans to deploy a total of 100 security personnel for security management. Based on the first ratio for each similar area, the corresponding number of security personnel is allocated as follows: 50 to Area A, 30 to Area B, and 20 to Area C.

[0103] This approach not only enables precise alignment of additional patrol areas across campuses and effectively measures the frequency of security incidents, but also ensures the rational allocation of security resources. The advantage of this approach lies in its combination of historical data analysis and a scientific resource allocation strategy, ensuring that each similar area receives security personnel commensurate with its risk, thereby improving the efficiency and effectiveness of overall security management. If actual demand exceeds the allocated number, the shortfall can be compensated by dynamically adjusting patrol frequency or introducing temporary security personnel.

[0104] Exemplarily, when the preset recognition model identifies a person to be managed, the distance between the position vector and the position vector corresponding to the person to be managed in the point cloud map of the target campus is calculated, and according to the distance and the safety level of the person to be managed, a number of security personnel matching the safety level are assigned to the location of the person to be managed, specifically including:

[0105] Obtaining a database of registered personnel of the target campus;

[0106] In the preset C3D model, the weight of each camera is evaluated according to the Taylor expansion method for structured pruning;

[0107] The C3D model is used to perform behavior recognition on persons in the registered person database. If a person is recognized to have performed a target action, the person is designated as a first person to be managed. A first distance between a position vector and a position vector corresponding to the first person to be managed in the point cloud map of the target campus is calculated. Based on ascending order of the first distances and the security level corresponding to the target action, a number of security personnel matching the security level is assigned to the location of the person to be managed.

[0108] Add context variables including regional road direction information and regional location information to the preset VAEs model;

[0109] The VAEs model is called to perform behavior recognition on personnel in the non-registered personnel database. If a person is recognized to perform a target action, the person is regarded as a second person to be managed. The second distance between the position vector and the corresponding position vector of the second person to be managed in the point cloud map of the target campus is calculated. According to the ascending order of the second distances and the security level corresponding to the target action, a number of security personnel matching the security level are assigned to the location of the person to be managed.

[0110] The Taylor expansion method can be used to evaluate the weight of each camera and perform structured pruning accordingly. This step aims to reduce unnecessary computation and improve the efficiency of the model on edge devices. The optimized C3D model is then called to perform real-time behavior recognition on the personnel in the registered personnel database. If a person is detected to perform a predefined target action (such as running, climbing, etc.), they are marked as the first person to be managed. Based on the location information of the surveillance camera, the corresponding position vector of the first person to be managed in the target campus point cloud map is calculated. Finally, the distance between the current location of the security personnel and the first person to be managed is calculated and sorted in ascending order according to the distance.

[0111] Based on the safety level of the target action (for example, running might be marked as medium risk, while climbing is high risk), the corresponding number of security personnel are assigned to handle it. For example, for high-risk actions, more or higher-quality security personnel can be dispatched first.

[0112] When using the VAEs model, context variables including regional road direction information and regional location information can be added to the preset VAEs model. This additional information helps improve the recognition accuracy of the model in complex environments. The optimized VAEs model is then called to perform real-time recognition of the behavior of non-registered persons. If a person is detected making a predefined target action, they are marked as the second person to be managed. Similarly, based on the location information of the surveillance camera, the corresponding position vector of the second person to be managed in the target campus point cloud map is calculated. The distance between the current location of the security personnel and the second person to be managed is calculated and sorted in ascending order according to the distance. Combined with the security level corresponding to the target action, the corresponding number of security personnel are assigned to handle it. Since non-registered persons may have higher uncertainty, the response level can be appropriately increased.

[0113] This approach not only enables efficient identification and rapid response to on-campus human behavior, but also ensures the rational allocation of security resources. The advantage of this approach lies in its combination of behavioral recognition technology and resource scheduling strategies, significantly improving the overall effectiveness of campus security management.

[0114] Compared to existing technologies, the campus security management method provided by the present invention ensures comprehensiveness and accuracy by acquiring a point cloud map of the target campus and other campuses within its jurisdiction. High-risk areas are then automatically identified and used as the center of additional patrol areas, eliminating the subjectivity and bias of human judgment and making predictions more scientific and objective. Feature descriptors are calculated to accurately extract feature information from the point cloud map. A nearest neighbor algorithm is then used to find corresponding point pairs, and a transformation matrix is ​​calculated based on these point pairs. This process enables precise registration of point cloud maps between different campuses, providing a reliable foundation for subsequent security management. The additional patrol areas on other campuses are then mapped onto the target campus based on the transformation matrix, forming similar areas. This allows for direct application of successful experiences from other campuses to quickly locate high-risk areas on the target campus. Finally, security personnel are allocated based on the ratio of historical security incident frequencies between similar areas and the additional patrol areas, ensuring that each area receives security resources commensurate with its risk. This method avoids resource waste or inadequate protection, improving the effectiveness of resource allocation.

[0115] Furthermore, the present invention continuously acquires security personnel's location information and, in conjunction with a behavioral recognition model, monitors and adjusts security resources in real time. Once abnormal behavior is identified, the system rapidly calculates the location vector of the individual in question and assigns appropriate security personnel based on proximity and security level. This dynamic response mechanism enhances the system's flexibility and adaptability.

[0116] An embodiment of the present application provides a campus safety management system, including a point cloud acquisition module 20, an area determination module 21, a matrix determination module 22, an area alignment module 23, a personnel allocation module 24 and a real-time allocation module 25.

[0117] The point cloud acquisition module 20 is used to obtain a point cloud map of the target campus and point cloud maps of other campuses in the global jurisdiction where the target campus is located.

[0118] The area determination module 21 is used to determine multiple additional patrol areas based on historical security incidents that occurred in other campuses.

[0119] The matrix determination module 22 is used to calculate a set of point cloud feature descriptors of the target campus and a set of point cloud feature descriptors of the other campuses, and determine the transformation matrix of the two sets of point cloud feature descriptors.

[0120] The area alignment module 23 is used to align multiple similar areas in the target campus according to the multiple patrol areas and the transformation matrix.

[0121] The personnel allocation module 24 is used to allocate security personnel in a corresponding proportion according to the frequency of historical security incidents occurring in the multiple similar areas corresponding to the multiple increased patrol areas; each security personnel is responsible for one of the similar areas.

[0122] The real-time allocation module 25 is used to continuously obtain the position vector corresponding to the location of the security personnel in the point cloud map of the target campus. When the preset recognition model identifies the person to be managed, the distance between the position vector and the position vector corresponding to the person to be managed in the point cloud map of the target campus is calculated. According to the distance and the safety level of the person to be managed, a preset number of the security personnel are allocated to the location of the person to be managed.

[0123] Those skilled in the art will clearly understand that, for the convenience and brevity of description, the specific working process of the campus security management system described above can refer to the corresponding process in the aforementioned method embodiment and will not be repeated here.

[0124] Compared to existing technologies, the campus security management system provided by the present invention ensures comprehensiveness and accuracy by acquiring point cloud maps of the target campus and other campuses within its jurisdiction. It then automatically identifies high-risk areas and uses them as the center of additional patrol zones, eliminating the subjectivity and bias of human judgment and making predictions more scientific and objective. By calculating feature descriptors, the feature information of the point cloud is accurately extracted. A nearest neighbor algorithm is then used to find corresponding point pairs, and a transformation matrix is ​​calculated based on these point pairs between the two point cloud maps. This process enables precise registration of point cloud maps between different campuses, providing a reliable foundation for subsequent security management. The additional patrol zones on other campuses are then mapped onto the target campus based on the transformation matrix, forming similar zones. This allows for direct application of successful experiences from other campuses to quickly locate high-risk areas on the target campus. Finally, security personnel are allocated based on the ratio of historical security incident frequencies between similar zones and the additional patrol zones, ensuring that each zone receives security resources commensurate with its risk. This method avoids resource waste or inadequate protection, improving the effectiveness of resource allocation.

[0125] Furthermore, the present invention continuously acquires security personnel's location information and, in conjunction with a behavioral recognition model, monitors and adjusts security resources in real time. Once abnormal behavior is identified, the system rapidly calculates the location vector of the individual in question and assigns appropriate security personnel based on proximity and security level. This dynamic response mechanism enhances the system's flexibility and adaptability.

[0126] An embodiment of the present application provides a computer-readable storage medium storing a computer program, which implements the above-mentioned campus safety management method when executed by a processor.

[0127] The computer device may be a computing device such as a smartphone, tablet computer, desktop computer, or cloud server. The computer device may include, but is not limited to, a processor and memory. Those skilled in the art will appreciate that the diagrams are merely examples of computer devices and do not limit the computer device. The computer device may include more or fewer components than shown, or a combination of certain components, or different components. For example, the computer device may also include input / output devices, network access devices, etc.

[0128] The processor may be a central processing unit (CPU), other general-purpose processors, digital signal processors (DSP), application-specific integrated circuits (ASIC), field-programmable gate arrays (FPGA) or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. A general-purpose processor may be a microprocessor or any conventional processor.

[0129] In some embodiments, the memory may be an internal storage unit of the computer device, such as a hard disk or memory of the computer device. In other embodiments, the memory may also be an external storage device of the computer device, such as a plug-in hard disk equipped on the computer device, a smart memory card (Smart Media Card, SMC), a secure digital (Secure Digital, SD) card, a flash card, etc. Furthermore, the memory may also include both an internal storage unit of the computer device and an external storage device. The memory is used to store an operating system, an application program, a boot loader (BootLoader), data, and other programs, such as the program code of the computer program. The memory may also be used to temporarily store data that has been output or is to be output.

[0130] An embodiment of the present application provides a computer program product. When the computer program product is run on a computer device, the computer device implements the steps in the above-mentioned various method embodiments when executing the computer program product.

[0131] In several embodiments provided in the present application, it is understood that each box in the flow chart or block diagram can represent a part of a module, program segment or code, and the part of the module, program segment or code contains one or more executable instructions for realizing the specified logical function. It should also be noted that in some alternative implementations, the functions marked in the box can also occur in an order different from that marked in the accompanying drawings. For example, two consecutive boxes can actually be executed substantially in parallel, and they can sometimes be executed in the opposite order, which depends on the functions involved.

[0132] If the functions are implemented in the form of software function modules and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present application, or the part that contributes to the prior art, or the part of the technical solution, can be embodied in the form of a software product. The computer software product is stored in a storage medium and includes several instructions for enabling a computer device to execute all or part of the steps of the methods described in each embodiment of the present application. The aforementioned storage medium includes various media that can store program codes, such as a USB flash drive, a mobile hard disk, a read-only memory (ROM), a random access memory (RAM), a magnetic disk, or an optical disk.

[0133] The above is a preferred embodiment of the present invention. It should be pointed out that for ordinary technicians in this technical field, several improvements and modifications can be made without departing from the principles of the present invention. These improvements and modifications are also considered to be within the scope of protection of the present invention.

Claims

1. A campus safety management method, characterized in that: include: Obtaining a point cloud map of the target campus and point cloud maps of other campuses in the global jurisdiction where the target campus is located; Based on historical security incidents that occurred on other campuses, multiple additional patrol areas were identified; Calculating a set of point cloud feature descriptors of the target campus and a set of point cloud feature descriptors of the other campuses, and determining the transformation matrix of the two sets of point cloud feature descriptors, specifically including: calculating FPFH and SHOT of the point cloud image of the target campus as one set of point cloud feature descriptors, and calculating FPFH and SHOT of the point cloud images of the other campuses as another set of point cloud feature descriptors; finding all valid nearest neighbor point pairs between the point cloud image of the target campus and the point cloud images of the other campuses by a nearest neighbor algorithm based on the two sets of point cloud feature descriptors; and calculating the transformation matrix of the two sets of point cloud feature descriptors based on all the valid nearest neighbor point pairs; Aligning a plurality of similar areas in the target campus according to the plurality of additional patrol areas and the transformation matrix; Allocate security personnel in a corresponding proportion according to the frequency of historical security incidents occurring in the plurality of similar areas corresponding to the plurality of additional patrol areas; each security personnel is responsible for one of the similar areas; Continuously obtain the position vector corresponding to the position of the security personnel in the point cloud map of the target campus. When the preset recognition model identifies the person to be managed, calculate the distance between the position vector corresponding to the position of the security personnel in the point cloud map of the target campus and the position vector corresponding to the position of the person to be managed in the point cloud map of the target campus. According to the distance and the safety level of the person to be managed, allocate a preset number of the security personnel to the location of the person to be managed.

2. A campus safety management method according to claim 1, characterized in that: The step of obtaining a point cloud map of the target campus and point cloud maps of other campuses in the entire jurisdiction of the target campus specifically includes: For the target campus and other campuses in the entire jurisdiction where the target campus is located, multiple scanning stations are set according to a preset overlap range percentage; the preset overlap range percentage is equal to the overlap percentage of the patrol ranges of two adjacent security personnel, the preset overlap range percentage is the ratio of the overlapping scanning area of ​​adjacent scanning stations to the scanning area of ​​each of the scanning stations, and the patrol range overlap percentage is the ratio of the overlapping patrol area of ​​two adjacent security personnel to the scanning area of ​​each of the security personnel.

3. A campus safety management method according to claim 1, characterized in that: Based on the historical security incidents that occurred on other campuses, multiple additional patrol areas are determined, including: Performing K-means clustering on the historical security events that occurred on the other campuses, and using multiple locations corresponding to the most recent historical security events in multiple clusters as the centers of multiple additional patrol areas; A plurality of additional patrol areas are determined according to a preset patrol range and the centers of the plurality of additional patrol areas.

4. A campus safety management method according to claim 1, characterized in that: The method of finding all valid nearest neighbor point pairs between the point cloud image of the target campus and the point cloud images of other campuses using the nearest neighbor algorithm based on the two sets of point cloud image feature descriptors specifically includes: Calculate the mean projection distance of the FPFH feature descriptor of each point in the other campus point cloud map on the corresponding point fitness vector as the FPFH distance; Calculate the average of the SHOT feature descriptor and the corresponding point fitness vector projection distance of each point in the other campus point cloud map as the SHOT distance; Using the difference between the FPFH distance and the SHOT distance as a distance threshold; Use the nearest neighbor algorithm to find all nearest neighbor point pairs between the point cloud of the target campus and the point cloud of other campuses; All the nearest neighbor point pairs whose distances between the nearest neighbor point pairs are less than or equal to the distance threshold are regarded as all valid nearest neighbor point pairs.

5. A campus safety management method according to claim 1, characterized in that: Before finding all valid nearest neighbor point pairs between the point cloud image of the target campus and the point cloud images of other campuses using the nearest neighbor algorithm based on the two sets of point cloud image feature descriptors, the method further includes: A KD tree index structure is established based on two sets of point cloud feature descriptors.

6. A campus safety management method according to claim 1, characterized in that: The method for aligning a plurality of similar areas in the target campus according to the plurality of additional patrol areas and the transformation matrix specifically includes: According to the transformation matrix, coordinate transformation is performed on each point of the plurality of additional patrol areas to obtain a plurality of similar areas.

7. A campus safety management method according to claim 1, characterized in that: The allocating of security personnel in a corresponding proportion according to the frequency of historical security incidents occurring in the plurality of similar areas corresponding to the plurality of additional patrol areas specifically includes: According to a first ratio of the frequency of historical security incidents occurring in each of the similar areas corresponding to the additional patrol area to the frequency of historical security incidents occurring in all the similar areas, security personnel in the same ratio as the first ratio are allocated to the corresponding similar areas.

8. A campus safety management method according to claim 1, characterized in that: When the preset recognition model identifies the person to be managed, calculating the distance between the position vector corresponding to the position of the security personnel in the point cloud map of the target campus and the position vector corresponding to the position of the person to be managed in the point cloud map of the target campus, and assigning, based on the distance and the safety level of the person to be managed, a number of the security personnel matching the safety level to the location of the person to be managed, specifically including: Obtaining a database of registered personnel of the target campus; In the preset C3D model, the weight of each camera is evaluated according to the Taylor expansion method for structured pruning; The C3D model is called to perform behavior recognition on persons in the registered person database. If a person is recognized to have performed a target action, the person is designated as a first person to be managed. A first distance is calculated between a position vector corresponding to the position of the security personnel in the point cloud map of the target campus and a position vector corresponding to the first person to be managed in the point cloud map of the target campus. Based on ascending order of the first distances and the security level corresponding to the target action, a number of security personnel matching the security level is assigned to the location of the person to be managed. Add context variables including regional road direction information and regional location information to the preset VAEs model; The VAEs model is called to perform behavior recognition on personnel in the non-registered personnel database. If a person is recognized to perform a target action, the person is regarded as the second person to be managed. The second distance between the position vector corresponding to the position of the security personnel in the point cloud map of the target campus and the position vector corresponding to the second person to be managed in the point cloud map of the target campus is calculated. According to the ascending order of the second distances and the security level corresponding to the target action, security personnel with a number matching the security level are assigned to the location of the person to be managed.

9. A campus safety management system, characterized in that: include: A point cloud acquisition module is used to obtain a point cloud map of the target campus and point cloud maps of other campuses in the global jurisdiction where the target campus is located; An area determination module, configured to determine a plurality of additional patrol areas based on historical security incidents occurring on other campuses; The matrix determination module is used to calculate a set of point cloud feature descriptors of the target campus and a set of point cloud feature descriptors of the other campuses, and determine the transformation matrix of the two sets of point cloud feature descriptors, specifically including: calculating FPFH and SHOT of the point cloud image of the target campus as one set of point cloud feature descriptors, and calculating FPFH and SHOT of the point cloud images of the other campuses as another set of point cloud feature descriptors; based on the two sets of point cloud feature descriptors, finding all valid nearest neighbor point pairs between the point cloud image of the target campus and the point cloud images of the other campuses through the nearest neighbor algorithm; and calculating the transformation matrix of the two sets of point cloud feature descriptors based on all the valid nearest neighbor point pairs; An area alignment module, configured to align a plurality of similar areas in the target campus according to the plurality of patrol-added areas and the transformation matrix; A personnel allocation module is used to allocate security personnel in a corresponding proportion according to the frequency of historical security incidents occurring in the plurality of similar areas corresponding to the plurality of additional patrol areas; each security personnel is responsible for one of the similar areas; The real-time allocation module is used to continuously obtain the position vector corresponding to the location of the security personnel in the point cloud map of the target campus. When the preset recognition model identifies the person to be managed, the distance between the position vector corresponding to the location of the security personnel in the point cloud map of the target campus and the position vector corresponding to the location of the person to be managed in the point cloud map of the target campus is calculated. According to the distance and the safety level of the person to be managed, a preset number of the security personnel are allocated to the location of the person to be managed.

Images