Campus safety information management method based on artificial intelligence

Abstract

The application discloses a kind of campus safety information management method based on artificial intelligence, it is related to campus safety information management technical field, the present application, real-time acquisition through the multi-modal identity characteristics of personnel, in combination with the pre-stored authorized personnel database, using the convolutional neural network CNN and space-time feature extraction model based on deep learning, high-precision comparison is carried out to face and gait features, first match face features, fail again trigger gait feature recognition, and in combination with dynamic comparison decision mechanism, for the situation classification generation suspicious person mark or check notice of matching failure, verification process is efficient and accurate;Integrate crowd density monitoring and congestion prediction model, using long short-term memory network LSTM to carry out trend prediction to the data collected in real time, calculate density growth rate and congestion state mark, identify peak congestion risk in advance, and automatically adjust traffic mode;Enable additional channel or switch fast verification mode, guarantee the traffic efficiency of peak period.

Description

Technical Field

[0001] The present invention relates to the technical field of campus safety information management, and in particular to an artificial intelligence-based campus safety information management method. Background Art

[0002] With the continuous expansion of campus scale and the increasing frequency of personnel flow, information management has gradually become an important support means for campus security management. Schools generally deploy surveillance cameras, access control systems and visitor registration systems to provide basic guarantees for campus safety.

[0003] During daily peak hours, such as when students go to and from school or when parents visit, school entrances and exits are often faced with heavy traffic. Although traditional access cards or manual verification methods can complete basic identity verification, and visitor registration systems can also record information on outsiders, such processes are cumbersome and inefficient, and can easily lead to queues and congestion at entrances and exits. In a chaotic environment, unauthorized personnel can easily take the opportunity to sneak into the campus, posing a potential safety hazard.

[0004] In addition, the role of traditional monitoring equipment is more limited to post-event tracing, and its ability to prevent real-time risks is limited. Faced with scenarios with rapid personnel flow, relying solely on manual review of monitoring images or verification of visitor information is not only difficult to detect problems in a timely manner, but is also prone to management loopholes due to human negligence, affecting the daily travel efficiency of teachers and students. Therefore, there is an urgent need for an artificial intelligence-based campus security information management method to solve such problems. Summary of the Invention

[0005] In view of the above existing problems, the present invention is proposed.

[0006] The present invention provides an artificial intelligence-based campus security information management method to solve the problem that during peak hours, such as morning and evening entrance times or during school opening and closing times, a large number of students and visitors pass through the school gate at the same time. Traditional access cards or manual verification methods are inefficient and easily lead to queues and congestion.

[0007] In order to solve the above technical problems, the present invention provides the following technical solutions:

[0008] The embodiment of the present invention provides a campus safety information management method based on artificial intelligence, which includes:

[0009] Step S1: deploying smart cameras that support face recognition and gait recognition at the school gate to collect the identity features of people passing by in real time;

[0010] Step S2: Build a database that stores authorized personnel data, and use artificial intelligence algorithms to compare the identity features collected in step S1 with the data in the database to confirm the identity of the personnel;

[0011] In step S3, the smart camera additionally monitors the crowd density data in the school gate area and uses the congestion prediction model to predict the collected crowd density data. When it is predicted that peak congestion is about to occur or it is detected that congestion has actually occurred, adjustment measures are automatically executed. The adjustment measures include enabling additional passages and switching to quick verification mode.

[0012] As a preferred solution of the campus security information management method based on artificial intelligence described in the present invention, the identity feature information includes facial images and gait features.

[0013] As a preferred solution of the campus security information management method based on artificial intelligence described in the present invention, the step of constructing a database for pre-storing authorized personnel data is as follows:

[0014] Define the dataset of authorized personnel as D:

[0015]

[0016] Among them, I i is the original face image of the i-th authorized person, G i is the original gait feature data of the i-th authorized person, L i is the identity label of the i-th authorized person, and n is the total number of authorized persons in the dataset;

[0017] For face image I i Perform feature extraction and generate feature vectors Expressed as:

[0018]

[0019] Among them, f face As a face feature extraction model, a CNN model based on convolutional neural network is used. is the embedded facial feature vector of the i-th authorized person;

[0020] Gait feature G i Perform feature extraction and generate feature vectors Expressed as:

[0021]

[0022] in,

[0023] g gait It is a gait feature extraction model based on spatiotemporal feature extraction method. is the embedded gait feature vector of the i-th authorized person;

[0024] The extracted features are stored in the database and a feature set is constructed, which is expressed as:

[0025]

[0026] Among them, DB is the final constructed database, are the feature vectors of face and gait respectively, L i For identity tags.

[0027] As a preferred solution of the campus safety information management method based on artificial intelligence described in the present invention, the comparison process of step S2 includes facial feature comparison and gait feature comparison, and the facial feature comparison is performed using a convolutional neural network CNN framework based on deep learning combined with an angle loss function, and the gait feature comparison is performed using a deep learning model based on spatiotemporal feature extraction.

[0028] As a preferred solution of the campus safety information management method based on artificial intelligence described in the present invention, in which: during the comparison process of step S2,

[0029] First, the collected facial features are matched. If the facial features are matched successfully, gait features are further identified. When gait features are successfully identified, the passenger is automatically released.

[0030] If the facial feature matching fails, gait feature recognition will also be performed. If gait feature recognition fails, the person will be marked as a suspicious person and the security guard will be notified for verification. If gait feature recognition is successful, only the security guard will be notified for verification and the suspicious status will not be marked.

[0031] As a preferred solution of the campus security information management method based on artificial intelligence described in the present invention, wherein: the authorized personnel data includes facial images and gait features of students and faculty members;

[0032] In the database, temporary reservation data is added as needed, including facial images of temporary personnel.

[0033] As a preferred solution of the campus safety information management method based on artificial intelligence described in the present invention, the steps of using a convolutional neural network (CNN) framework based on deep learning combined with an angle loss function to perform facial feature comparison and using a deep learning model based on spatiotemporal feature extraction to perform gait feature comparison are as follows:

[0034] The input features and reference features are defined as follows:

[0035] X f =f face (I), X g =g gait (G),

[0036] Among them, X fis the facial feature vector collected in real time, X g is the gait feature vector collected in real time, I and G are the face image and gait feature data collected in real time, and f face 、g gait They are face and gait feature extraction models respectively.

[0037] Based on the convolutional neural network CNN, facial feature comparison is performed and the similarity is optimized by combining the angle loss function. The angle loss function formula is:

[0038]

[0039] Among them, L face is the angle loss function for facial feature comparison, θ j is the angle between the jth collected feature and the feature vector in the database, θ k is the angle between the kth collected feature in the database and the feature vector in the database, m is the number of facial features collected in real time, and n is the number of facial features pre-stored in the database;

[0040] The temporal and spatial feature comparison of gait features is performed, and the similarity calculation formula is:

[0041]

[0042] Among them, S gait is the gait feature similarity score, α t is the gait timing weighting coefficient, is the temporal angle of the gait feature vector, T is the temporal frame number of the gait data,

[0043] Calculate the comprehensive comparison score using the following formula:

[0044] S total =λ f ·S face +λ g ·S gait ,

[0045] Among them, S total is the comprehensive similarity score, S face is the facial feature similarity score,

[0046] λ f ,λ g is the weight parameter,

[0047] Compare the results and determine:

[0048] If S total ≥τ, then Decision=Accept,

[0049] If S total<τ, then Decision=Reject,

[0050] Among them, Decision is the comparison result, Accept indicates authorized identity, Reject indicates unauthorized identity and needs further judgment, τ is the similarity threshold, which is set to a predefined constant.

[0051] As a preferred solution of the campus safety information management method based on artificial intelligence described in the present invention, the comparison process of step S2 includes:

[0052] Input the real-time collected features and compare them with the pre-stored features in the database to calculate the similarity. The calculation formula is:

[0053]

[0054] in, The facial features X collected in real time f and the facial features of the i-th person in the database The similarity of is the gait feature X collected in real time g and the gait characteristics of the i-th person in the database Similarity, cos is the cosine similarity between two feature vectors;

[0055] To determine whether the facial feature comparison is successful, the judgment formula is:

[0056] like FaceMatch i =1,

[0057] like FaceMatch i =0,

[0058] Among them, FaceMatch i is the face comparison result flag, 1 for success and 0 for failure, τf is the similarity threshold of face feature comparison;

[0059] When facial feature comparison fails, gait feature comparison is triggered. The judgment formula is:

[0060] like GaitMatch i =1,

[0061] like GaitMatch i =0,

[0062] Among them, GaitMatch i is the gait comparison result flag, 1 is success, 0 is failure, τg is the similarity threshold for gait feature comparison;

[0063] Combined with the comparison results, the review status flag is generated, which is expressed as:

[0064] If FaceMatch i =1 and GaitMatch i =1, then AuditState=Pass,

[0065] If FaceMatch i =0 and GaitMatch i =1, then AuditState=Warning,

[0066] If FaceMatch i =0 and GaitMatch i =0, then AuditState=Alert,

[0067] AuditState indicates the audit status, Pass indicates a successful match and confirmation by personnel, Warning indicates a successful gait match but a failed face match, and Alert indicates a failed face and gait match.

[0068] Record tag information, including:

[0069] M={(X f , X g ,AuditState)},

[0070] Among them, M is the tag information set, and AuditState is the audit status corresponding to the current comparison.

[0071] As a preferred solution of the campus security information management method based on artificial intelligence described in the present invention, the comparison process of the quick verification mode is as follows: after the facial feature match is successful, the person is directly released without the need for gait feature recognition; if the facial feature match fails, gait feature recognition is performed, and the person whose gait feature recognition fails is automatically marked and the guard is notified to record his identity.

[0072] As a preferred solution of the campus safety information management method based on artificial intelligence described in the present invention, wherein: the congestion prediction model is used to predict the collected crowd density data, and when peak congestion is predicted to occur or congestion is detected to have actually occurred, the step of automatically executing adjustment measures is as follows:

[0073] Define the real-time collected crowd density data sequence as D t , expressed as:

[0074] Dt ={d t-k , d t-k+1 ,…,d t},

[0075] Among them, D t is the crowd density data sequence collected k+1 moments before time t, d t-k , d t-k+1 ,...,d t is the local crowd density data collected between time tk and t, where k is the time series length of the prediction window;

[0076] Input the crowd density data into the time series forecasting model to generate the future density forecast value, which is expressed as:

[0077]

[0078] in, is the predicted crowd density value at time t+1, f lstm It is a density prediction model based on the long short-term memory network LSTM;

[0079] Calculate the density growth rate and compare it with the threshold. The density growth rate calculation formula is:

[0080]

[0081] Among them, g t+1 is the density growth rate at time t+1, τ g is the density growth rate threshold, Δ is the state where the mark growth rate exceeds the threshold;

[0082] The congestion state is determined by the following formula:

[0083] If g t+1 <τ g , then CongestionState=Normal,

[0084] If g t+1 ≥τ g , then CongestionState=Warning,

[0085] CongestionState is the congestion state mark, Normal means the current predicted density is within the normal range, and Warning means the current predicted density exceeds the threshold range and congestion may occur;

[0086] Adjust the traffic mode according to the congestion status. The adjustment method is as follows:

[0087] If CongestionState=Normal, then Action=None,

[0088] If CongestionState = Warning, then Action = OpenExtraChannels,

[0089] Among them, Action is the response action, OpenExtraChannels is to enable additional channels,

[0090] Adjust the feature comparison process in the quick verification mode by:

[0091] If CongestionState=Warning, then Verification=FaceOnly,

[0092] If CongestionState=Normal, then Verification=FaceAndGait,

[0093] Verification is the feature comparison mode, FaceOnly is for facial feature comparison only, and FaceAndGait is for a comprehensive comparison of facial and gait features.

[0094] The beneficial effects of the present invention are as follows: the present invention collects multimodal identity features of passing people in real time, combines them with a pre-stored authorized personnel database, and uses a deep learning-based convolutional neural network (CNN) and a spatiotemporal feature extraction model to perform high-precision comparison of facial and gait features. In the verification process, facial features are matched first, and gait feature recognition is triggered if it fails. In combination with a dynamic comparison decision-making mechanism, suspicious person tags or verification notifications are generated for matching failures, making the verification process efficient and accurate.

[0095] The present invention integrates crowd density monitoring and congestion prediction models, uses long short-term memory (LSTM) networks to predict trends in real-time collected data, calculates density growth rates and congestion status markers, identifies peak congestion risks in advance, and automatically adjusts traffic patterns. It enables additional channels or switches to a fast verification mode to ensure efficiency and safety during peak hours. The fast verification mode simplifies the process, allowing passengers to pass directly after a successful facial feature match, avoiding unnecessary gait comparisons. Basic monitoring is retained for unmatched passengers, further balancing efficiency and safety.

[0096] The present invention adopts the angle loss function and the weighted similarity calculation method of gait spatiotemporal features in the feature comparison process, which enhances the robustness and versatility of the comparison and effectively reduces the false alarm rate and missed alarm rate in complex scenarios. BRIEF DESCRIPTION OF THE DRAWINGS

[0097] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the following briefly introduces the drawings required for use in the description of the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0098] Figure 1 This is a flow chart of the campus safety information management method based on artificial intelligence of the present invention. DETAILED DESCRIPTION

[0099] In order to make the above-mentioned objects, features and advantages of the present invention more obvious and easy to understand, the specific embodiments of the present invention are described in detail below with reference to the accompanying drawings.

[0100] In the following description, many specific details are set forth to facilitate a full understanding of the present invention. However, the present invention may also be implemented in other ways different from those described herein. Those skilled in the art may make similar generalizations without violating the connotation of the present invention. Therefore, the present invention is not limited to the specific embodiments disclosed below.

[0101] Secondly, the term "one embodiment" or "embodiment" herein refers to a specific feature, structure, or characteristic that may be included in at least one implementation of the present invention. The phrase "in one embodiment" appearing in various places throughout this specification does not necessarily refer to the same embodiment, nor does it refer to a separate or selective embodiment that is mutually exclusive of other embodiments.

[0102] Example 1, reference Figure 1 This embodiment provides a campus safety information management method based on artificial intelligence, including:

[0103] Step S1: deploying smart cameras that support face recognition and gait recognition at the school gate to collect the identity features of people passing by in real time;

[0104] The identity feature information includes facial image and gait features;

[0105] Step S2: Build a database that stores authorized personnel data, and use artificial intelligence algorithms to compare the identity features collected in step S1 with the data in the database to confirm the identity of the personnel;

[0106] The step of constructing a database for pre-storing authorized personnel data is:

[0107] Define the dataset of authorized personnel as D:

[0108]

[0109] Among them, I iis the original face image of the i-th authorized person, G i is the original gait feature data of the i-th authorized person, L i is the identity label of the i-th authorized person, and n is the total number of authorized persons in the dataset;

[0110] For face image I i Perform feature extraction and generate feature vectors Expressed as:

[0111]

[0112] Among them, f face As a face feature extraction model, a CNN model based on convolutional neural network is used. is the embedded facial feature vector of the i-th authorized person;

[0113] Gait feature G i Perform feature extraction and generate feature vectors Expressed as:

[0114]

[0115] in,

[0116] g gait It is a gait feature extraction model based on spatiotemporal feature extraction method. is the embedded gait feature vector of the i-th authorized person;

[0117] The extracted features are stored in the database and a feature set is constructed, which is expressed as:

[0118]

[0119] Among them, DB is the final constructed database, are the feature vectors of face and gait respectively, L i For identity tags;

[0120] Specifically, this step defines the original data set and extracts facial and gait features, storing the feature vectors in the database for subsequent comparison; by adopting convolutional neural networks and spatiotemporal feature extraction models, the data storage process and accuracy are improved, and feature extraction deeply encodes input data of different dimensions to enhance the versatility of the database.

[0121] The comparison process in step S2 includes facial feature comparison and gait feature comparison. The facial feature comparison is performed using a convolutional neural network (CNN) framework based on deep learning combined with an angle loss function, while the gait feature comparison is performed using a deep learning model based on spatiotemporal feature extraction.

[0122] During the comparison process of step S2,

[0123] First, the collected facial features are matched. If the facial features are matched successfully, gait features are further identified. When gait features are successfully identified, the passenger is automatically released.

[0124] If facial feature matching fails, gait feature recognition is also performed. If gait feature recognition fails, the person is marked as suspicious and the security guard is notified for verification. If gait feature recognition is successful, only the security guard is notified for verification and the suspicious status is not marked.

[0125] The authorized personnel data includes facial images and gait characteristics of students and faculty members;

[0126] In the database, temporary reservation data is added as needed, including facial images of temporary personnel;

[0127] The steps of using a deep learning-based convolutional neural network (CNN) framework combined with an angle loss function to perform facial feature comparison and a deep learning model based on spatiotemporal feature extraction to perform gait feature comparison are as follows:

[0128] The input features and reference features are defined as follows:

[0129] X f =f face (I), X g =g gait (G),

[0130] Among them, X f is the facial feature vector collected in real time, X g is the gait feature vector collected in real time, I and G are the face image and gait feature data collected in real time, and f face 、g gait They are face and gait feature extraction models respectively.

[0131] Based on the convolutional neural network CNN, facial feature comparison is performed and the similarity is optimized by combining the angle loss function. The angle loss function formula is:

[0132]

[0133] Among them, L face is the angle loss function for facial feature comparison, θ j is the angle between the jth collected feature and the feature vector in the database, θ k is the angle between the kth collected feature in the database and the feature vector in the database, m is the number of facial features collected in real time, and n is the number of facial features pre-stored in the database;

[0134] The temporal and spatial feature comparison of gait features is performed, and the similarity calculation formula is:

[0135]

[0136] Among them, S gait is the gait feature similarity score, α t is the gait timing weighting coefficient, is the temporal angle of the gait feature vector, T is the temporal frame number of the gait data,

[0137] Calculate the comprehensive comparison score using the following formula:

[0138] S total =λ f ·S face +λ g ·S gait ,

[0139] Among them, S total is the comprehensive similarity score, S face is the facial feature similarity score,

[0140] λ f ,λ g is the weight parameter,

[0141] Compare the results and determine:

[0142] If S total ≥τ, then Decision=Accept,

[0143] If S total <τ, then Decision=Reject,

[0144] Among them, Decision is the comparison result, Accept indicates authorized identity, Reject indicates unauthorized identity and requires further judgment, T is the similarity threshold, which is set to a predefined constant;

[0145] Specifically, this step performs multi-level feature comparison, adopts angle loss function and spatiotemporal weighted similarity calculation to improve the robustness of the comparison; in addition, comprehensive similarity calculation is combined with weight adjustment to dynamically balance facial and gait features.

[0146] The comparison process of step S2 includes:

[0147] Input the real-time collected features and compare them with the pre-stored features in the database to calculate the similarity. The calculation formula is:

[0148]

[0149] in, The facial features X collected in real time fand the facial features of the i-th person in the database The similarity of is the gait feature X collected in real time g and the gait characteristics of the i-th person in the database Similarity, cos is the cosine similarity between two feature vectors;

[0150] To determine whether the facial feature comparison is successful, the judgment formula is:

[0151] like FaceMatch i =1,

[0152] like FaceMatch i =0,

[0153] Among them, FaceMatch i is the face comparison result flag, 1 for success, 0 for failure, τ f is the similarity threshold for facial feature comparison;

[0154] When facial feature comparison fails, gait feature comparison is triggered. The judgment formula is:

[0155] like GaitMatch i =1,

[0156] like GaitMatch i =0,

[0157] Among them, GaitMatch i is the gait comparison result flag, 1 is success, 0 is failure, τ g is the similarity threshold for gait feature comparison;

[0158] Combined with the comparison results, the review status flag is generated, which is expressed as:

[0159] If FaceMatch i =1 and GaitMatch i =1, then AuditState=Pass,

[0160] If FaceMatch i =0 and GaitMatch i =1, then AuditState=Warning,

[0161] If FaceMatch i =0 and GaitMatch i=0, then AuditState=Alert,

[0162] AuditState indicates the audit status, Pass indicates a successful match and confirmation by personnel, Warning indicates a successful gait match but a failed face match, and Alert indicates a failed face and gait match.

[0163] Record tag information, including:

[0164] M={(X f ,X g ,AuditState)},

[0165] Among them, M is the set of mark information, AuditState is the audit status corresponding to the current comparison;

[0166] Specifically, a multi-layer comparison decision tree is used to improve identity accuracy. In the event of facial feature comparison failure, gait comparison is initiated in a timely manner to minimize the false alarm rate. At the same time, a hierarchical management mechanism based on the audit status is used to clearly distinguish between normal, suspicious and high-risk individuals.

[0167] In step S3, the smart camera additionally monitors the crowd density data in the school gate area and uses the congestion prediction model to predict the collected crowd density data. When a peak congestion is predicted or detected, adjustment measures are automatically implemented. The adjustment measures include enabling additional access channels and switching to a quick verification mode.

[0168] The comparison process of the quick verification mode is as follows: if the facial feature match is successful, the person is directly released without gait feature recognition; if the facial feature match fails, gait feature recognition is performed, and the person whose gait feature recognition fails is automatically marked and the security guard is notified to record his / her identity;

[0169] The step of using the congestion prediction model to predict the collected crowd density data and automatically executing adjustment measures when it is predicted that peak congestion is about to occur or congestion is detected to have actually occurred is:

[0170] Define the real-time collected crowd density data sequence as D t , expressed as:

[0171] D t ={d t-k , d t-k+1 ,…,d t},

[0172] Among them, D t is the crowd density data sequence collected k+1 moments before time t, d t-k , d t-k+1 ,...,dt is the local crowd density data collected between time tk and t, where k is the time series length of the prediction window;

[0173] Input the crowd density data into the time series forecasting model to generate the future density forecast value, which is expressed as:

[0174]

[0175] in, is the predicted crowd density value at time t+1, f lstm It is a density prediction model based on the long short-term memory network LSTM;

[0176] Calculate the density growth rate and compare it with the threshold. The density growth rate calculation formula is:

[0177]

[0178] Among them, g t+1 is the density growth rate at time t+1, τ g is the density growth rate threshold, Δ is the state where the mark growth rate exceeds the threshold;

[0179] The congestion state is determined by the following formula:

[0180] If g t+1 <τ g , then CongestionState=Normal,

[0181] If g t+1 ≥τ g , then CongestionState=Warning,

[0182] CongestionState is the congestion state mark, Normal means the current predicted density is within the normal range, and Warning means the current predicted density exceeds the threshold range and congestion may occur;

[0183] Adjust the traffic mode according to the congestion status. The adjustment method is as follows:

[0184] If CongestionState=Normal, then Action=None,

[0185] If CongestionState = Warning, then Action = OpenExtraChannels,

[0186] Among them, Action is the response action, OpenExtraChannels is to enable additional channels,

[0187] Adjust the feature comparison process in the quick verification mode by:

[0188] If CongestionState=Warning, then Verification=FaceOnly,

[0189] If CongestionState=Normal, then Verification=FaceAndGait,

[0190] Verification is the feature comparison mode, FaceOnly is for facial feature comparison only, and FaceAndGait is for comprehensive comparison of facial and gait features.

[0191] Specifically, the pedestrian density is predicted through a time series prediction model, and combined with the growth rate threshold to achieve early warning of potential congestion; the dynamic adjustment mechanism based on the congestion status effectively reduces the verification load during peak periods while ensuring traffic efficiency and safety; and the rapid verification mode provides a balance between real-time and safety, further improving the adaptability of the present invention.

[0192] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit the present invention. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solutions of the present invention may be modified or replaced by equivalents without departing from the spirit and scope of the technical solutions of the present invention, which should all be included in the scope of the claims of the present invention.

Claims

1. A campus safety information management method based on artificial intelligence, characterized by: include, Step S1: deploying smart cameras that support face recognition and gait recognition at the school gate to collect the identity features of people passing by in real time; Step S2: Build a database that stores authorized personnel data, and use artificial intelligence algorithms to compare the identity features collected in step S1 with the data in the database to confirm the identity of the personnel; In step S3, the smart camera additionally monitors the crowd density data in the school gate area and uses the congestion prediction model to predict the collected crowd density data. When a peak congestion is predicted or detected, adjustment measures are automatically implemented. The adjustment measures include enabling additional access channels and switching to a quick verification mode. During the comparison process of step S2, First, the collected facial features are matched. If the facial features are matched successfully, gait features are further identified. When gait features are successfully identified, the passenger is automatically released. If facial feature matching fails, gait feature recognition is also performed. If gait feature recognition fails, the person is marked as suspicious and the security guard is notified for verification. If gait feature recognition is successful, only the security guard is notified for verification and the suspicious status is not marked. The comparison process of the fast verification mode is: after the facial features are successfully matched, the person is directly released without the need for gait feature recognition; If facial feature matching fails, gait feature recognition will be performed. For those whose gait feature recognition fails, they will be automatically marked and the security guard will be notified to record their identity. The steps of using the congestion prediction model to predict the collected crowd density data and automatically executing adjustment measures when peak congestion is predicted to occur or congestion is detected to have actually occurred are as follows: Define the real-time collected crowd density data sequence as D t , expressed as: D t ={d t-k ,d t-k+1 ,…,d t }, Among them, D t is the crowd density data sequence collected k+1 moments before time t, d t-k ,d t-k+1 ,…,d t is the local crowd density data collected between time tk and t, where k is the time series length of the prediction window; Input the crowd density data sequence into the time series forecasting model to generate the future density forecast value, which is expressed as: in, is the predicted crowd density value at time t+1, f lstm It is a density prediction model based on the long short-term memory network LSTM; Calculate the density growth rate and compare it with the threshold. The density growth rate calculation formula is: D=ξ g , Among them, g t+1 is the density growth rate at time t+1, ξ g is the density growth rate threshold, Δ is the state where the mark growth rate exceeds the threshold; The congestion state is determined by the following formula: If g t+1 <ξ g , then CongestionState=Normal, If g t+1 ≥ξ g , then CongestionState=Warning, CongestionState is the congestion status mark, Normal means the current predicted density is within the normal range, and Warning means the current predicted density exceeds the threshold range and congestion may occur; Adjust the traffic mode according to the congestion status. The adjustment method is as follows: If CongestionState=Normal, then Action=None, If CongestionState = Warning, then Action = OpenExtraChannels, Among them, Action is the response action, OpenExtraChannels is to enable additional channels, Adjust the feature comparison process in the quick verification mode by: If CongestionState=Warning, then Verification=FaceOnly, If CongestionState=Normal, then Verification=FaceAndGait, Verification is the feature comparison mode, FaceOnly is for facial feature comparison only, and FaceAndGait is for a comprehensive comparison of facial and gait features.

2. The campus safety information management method based on artificial intelligence according to claim 1, characterized in that: The identity features include facial images and gait features.

3. The campus safety information management method based on artificial intelligence according to claim 2, characterized in that: The step of constructing a database for pre-storing authorized personnel data is: Define the dataset of authorized personnel as D: Among them, I i is the original face image of the i-th authorized person, G i is the original gait feature data of the i-th authorized person, L i is the identity label of the i-th authorized person, and n is the total number of authorized persons in the dataset; The original face image I of the i-th authorized person i Perform feature extraction to generate the embedded face feature vector of the i-th authorized person Expressed as: Among them, f face As a face feature extraction model, a CNN model based on convolutional neural network is used. is the embedded facial feature vector of the i-th authorized person; The original gait feature data G of the i-th authorized person i Perform feature extraction to generate the embedded gait feature vector of the i-th authorized person Expressed as: in, g gait It is a gait feature extraction model based on spatiotemporal feature extraction method. is the embedded gait feature vector of the i-th authorized person; The generated embedded face feature vector and embedded gait feature vector are stored in the database, and a feature set is constructed, which is expressed as: Among them, DB is the final constructed database, are the embedded face feature vector and embedded gait feature vector of the i-th authorized person, L i is the identity tag of the i-th authorized person.

4. The campus safety information management method based on artificial intelligence according to claim 3, characterized in that: The comparison process of step S2 includes facial feature comparison and gait feature comparison. The convolutional neural network (CNN) framework based on deep learning is combined with the angle loss function for facial feature comparison, and the deep learning model based on spatiotemporal feature extraction is used for gait feature comparison.

5. The campus safety information management method based on artificial intelligence according to claim 4, characterized in that: The authorized personnel data includes facial images and gait characteristics of students and faculty members; In the database, temporary reservation data is added as needed, including facial images of temporary personnel.

6. The campus safety information management method based on artificial intelligence according to claim 5, characterized in that: The steps of using a deep learning-based convolutional neural network (CNN) framework combined with an angle loss function to perform facial feature comparison and a deep learning model based on spatiotemporal feature extraction to perform gait feature comparison are as follows: The input features and reference features are defined as follows: X f =f face (I),X g =g gait (G), Among them, X f is the facial feature vector collected in real time, X g is the gait feature vector collected in real time, I and G are the face image and gait feature data collected in real time, and f face 、g gait They are face and gait feature extraction models respectively. Based on the convolutional neural network CNN, facial feature comparison is performed and the similarity is optimized by combining the angle loss function. The angle loss function formula is: Among them, L face is the angle loss function for facial feature comparison, θ j is the angle between the jth collected feature and the feature vector in the database, θ k is the angle between the kth collected feature in the database and the feature vector in the database, m is the number of facial features collected in real time, and n is the number of facial features pre-stored in the database; The temporal and spatial feature comparison of gait features is performed, and the similarity calculation formula is: Among them, S gait is the gait feature similarity score, α p is the gait timing weighting coefficient, is the temporal angle of the gait feature vector, T is the temporal frame number of the gait data, Calculate the comprehensive comparison score using the following formula: S total =λ f ·S face +λ g ·S gait , Among them, S total is the comprehensive similarity score, S face is the facial feature similarity score, λ f ,λ g is the weight parameter, Compare the results and determine: If S total ≥τ, then Decision=Accept, If S total <τ, then Decision=Reject, Among them, Decision is the comparison result, Accept indicates authorized identity, Reject indicates unauthorized identity and needs further judgment, τ is the similarity threshold, which is set to a predefined constant.

7. The campus safety information management method based on artificial intelligence according to claim 6, characterized in that: The comparison process of step S2 includes: Input the real-time collected features and compare them with the pre-stored features in the database to calculate the similarity. The calculation formula is: in, is the real-time collected facial feature vector X f and the embedded facial feature vector of the i-th authorized person in the database The similarity of is the real-time collected gait feature vector X g and the embedded gait feature vector of the i-th authorized person in the database Similarity, cos is the cosine similarity between two feature vectors; To determine whether the facial feature comparison is successful, the judgment formula is: like FaceMatch i =1, like FaceMatch i =0, Among them, FaceMatch i is the face comparison result flag, 1 is success, 0 is failure, τ f is the similarity threshold for facial feature comparison; When facial feature comparison fails, gait feature comparison is triggered. The judgment formula is: like GaitMatch i =1, like GaitMatch i =0, Among them, GaitMatch i is the gait comparison result flag, 1 is success, 0 is failure, τ g is the similarity threshold for gait feature comparison; Combined with the comparison results, the review status flag is generated, which is expressed as: If FaceMatch i =1 and GaitMatch i =1, then AuditState=Pass, If FaceMatch i =0 and GaitMatch i =1, then AuditState=Warning, If FaceMatch i =0 and GaitMatch i =0, then AuditState=Alert, AuditState indicates the audit status, Pass indicates a successful match and confirmation by personnel, Warning indicates a successful gait match but a failed face match, and Alert indicates a failed face and gait match. Record tag information, including: M={(X f ,X g ,AuditState)}, Among them, M is the tag information set, and AuditState is the audit status corresponding to the current comparison.

Images