Intelligent park safety risk sensing system and method with prediction capability

Abstract

The invention discloses an intelligent park security risk perception system and method with prediction capability. The method comprises the following steps: acquiring face feature data registered at a park entrance guard; collecting video data of T monitoring areas of the park in real time; real-time face feature data in the video data appearing in the tth monitoring area is extracted, the real-time face feature data is compared with registered face feature data, the comparison result comprises the existence of a garden entry record and the absence of the garden entry record, and the real-time face feature data without the garden entry record is marked as abnormal personnel features; setting a deliberate avoidance area of each monitoring area in advance; using a tracking algorithm to track the abnormal personnel with the abnormal personnel characteristics, recording deliberate avoidance areas where the abnormal personnel characteristics appear in the q monitoring areas, marking the deliberate avoidance areas as passing avoidance areas, and sequentially connecting all the recorded passing avoidance areas to form an action track of the abnormal personnel; and the safety risk perception capability of the park is improved.

Description

Technical Field

[0001] The present invention relates to the technical field of park security management, and in particular to an intelligent park security risk perception system and method with predictive capability. Background Art

[0002] With the continuous development and expansion of smart parks, security risk management within the parks has become increasingly complex and important. Traditional security monitoring systems usually rely on single behavioral features, such as personnel entry and exit records, simple abnormal behavior detection, etc., to assess potential security risks. However, this method has obvious limitations and cannot fully capture the behavior patterns of personnel, resulting in inaccurate identification of abnormal behaviors and a high false alarm rate.

[0003] In the existing technology, various behavioral characteristics are often analyzed in isolation, failing to fully consider the complex and close correlations between them. This lack of in-depth analysis makes it impossible for the security system to effectively identify potential risks and to provide timely warnings of possible security incidents. In addition, frequent false alarms not only increase the workload of security personnel, but may also cause managers to become passive and lazy, further weakening the effectiveness of security management.

[0004] How to deeply explore the correlation between multiple behavioral characteristics and comprehensively and accurately evaluate the behavior of abnormal personnel has become a technical problem that needs to be urgently solved in the current field of smart park security management. Summary of the invention

[0005] The purpose of the present invention is to provide an intelligent park security risk perception system and method with predictive capabilities to address the deficiencies in the background technology.

[0006] In order to achieve the above-mentioned purpose, the present invention provides the following technical solution: a method for sensing security risks in an intelligent park with predictive capability, comprising:

[0007] Collect facial feature data of people registered at the park access control;

[0008] Collect video data of T monitoring areas in the park in real time;

[0009] Extract the real-time facial feature data appearing in the video data of the t-th monitoring area, t∈T, and compare the real-time facial feature data with the registered facial feature data to obtain a comparison result; the comparison result includes the presence of an entry record and the absence of an entry record, and mark the real-time facial feature data of the person without an entry record as an abnormal person feature;

[0010] Pre-set deliberate avoidance areas for each monitoring area;

[0011] Use the tracking algorithm to track the abnormal person with abnormal personnel characteristics, and record the deliberate avoidance area where the abnormal personnel characteristics appear in q monitoring areas, mark it as the passing avoidance area, and connect all the recorded passing avoidance areas in sequence to form the action trajectory of the abnormal person, q is less than or equal to T;

[0012] Intercept the image frames where abnormal persons appear in the video data, input them into the action recognition model, obtain the avoidance actions of the abnormal persons, and count the number of avoidance actions per unit time;

[0013] Collect the light intensity and humidity at the current moment, combine them with the number of avoidance actions, and generate the avoidance action coefficient;

[0014] Obtain movement feature data and height data of abnormal persons based on video data;

[0015] Input the action trajectory, avoidance action coefficient, movement feature data and height data into the behavior feature analysis model to obtain the behavior analysis results;

[0016] If the behavior analysis result is abnormal behavior, an early warning will be issued.

[0017] Furthermore, the avoidance action includes lowering the head and covering the face, and the number of times of lowering the head and covering the face in a unit time is counted; and the avoidance action coefficient generation includes:

[0018] The avoidance action value is obtained by weighted summing up the number of times of lowering the head and covering the face in unit time, the environmental impact value is obtained by weighted summing up the light intensity and humidity at the current moment, and the avoidance action value is divided by the environmental impact value to obtain the avoidance action coefficient.

[0019] Furthermore, the method for extracting the movement characteristic data of abnormal persons includes:

[0020] The moving characteristic data include the moving speed mean and moving speed standard deviation;

[0021] Mapping the position coordinates of the abnormal person in the continuous images in the video data to the electronic map of the park to obtain a continuous position coordinate sequence of the abnormal person in the unit interval time;

[0022] Calculate the moving speed S within the unit time interval:

[0023]

[0024] X i+1 Indicates the horizontal coordinate of the i+1th position, X i Represents the horizontal coordinate of the i-th position, Y i+1 Indicates the ordinate of the i+1th position, Y i represents the ordinate of the i-th position coordinate;

[0025] The calculated moving speeds are used to establish a moving speed set, and the moving speed average and moving speed standard deviation in the moving speed set are calculated.

[0026] Furthermore, the method of obtaining the height data of abnormal persons includes:

[0027] Using video processing technology, obtain the top and bottom pixel coordinates of the reference object and the abnormal person in the video data image;

[0028] According to the top and bottom pixel coordinates of the reference object, the height of the reference object in the video data image is obtained; similarly, according to the top and bottom pixel coordinates of the abnormal person, the height of the abnormal person in the video data image is obtained; the ratio of the height of the reference object in the video data image to the actual height of the reference object is marked as a scale factor; the height of the abnormal person in the video data image is multiplied by the scale factor to obtain the height data of the abnormal person.

[0029] Furthermore, the training method of the feature analysis model includes:

[0030] K groups of action trajectories, avoidance action coefficients, abnormal personnel heights, and mobile feature data are collected in advance and combined into K groups of feature vectors, where K is an integer greater than 1. A supervised learning method is used to train a behavioral feature analysis model using a support vector machine, random forest, or deep neural network. The normal behaviors and abnormal behaviors pre-labeled in the feature vectors form a data set of known behaviors. The data set is divided into a training set and a test set for feature analysis model training and verification, and a feature analysis model that meets the preset accuracy is output.

[0031] Furthermore, it also includes:

[0032] Collect wind speed and direction data for each monitoring area;

[0033] According to the wind speed and direction data or the light intensity, the video data of each monitoring area is analyzed to see if it is abnormal. If it is abnormal, an enable instruction is generated. If it is normal, no enable instruction is generated, and the enable instruction is sent to the backup server and the data switching switch. The backup server can complete the real-time analysis of the video data. The backup server is turned on according to the enable instruction, and the data switching switch transmits the video data to the backup server for analysis according to the enable instruction.

[0034] Furthermore, the method of analyzing whether the video data of each monitoring area is abnormal according to the wind speed and direction data or the light intensity includes:

[0035] When the wind force level of the r-th monitoring area is greater than the preset wind force level, the flexible ribbon offset direction in the video data of the r-th monitoring area is analyzed to see whether it is consistent with the wind direction. If not, an activation instruction is generated; if consistent, no activation instruction is generated;

[0036] If the wind force level is less than or equal to the preset wind force level, the real-time collected light intensity and the real-time image frame in the video data of the rth monitoring area are input into the monitoring anomaly analysis model to determine whether the image frame is abnormal. If the image frame is abnormal, an enable instruction is generated; if the image frame is normal, no enable instruction is generated.

[0037] Further, the method for analyzing the offset direction of the flexible ribbon in the video data of the r-th monitoring area includes:

[0038] In the display screen of the video data, a plane coordinate system is preset, and the fixed end of the flexible ribbon is used as the origin of the plane coordinate system. The coordinates of each pixel point of the area occupied by the flexible ribbon in the display screen at the current moment are obtained, and the quadrant of the plane coordinate system where the flexible ribbon is located is determined according to the coordinates of each pixel point. According to the preset indication direction of the quadrant, the offset direction of the flexible ribbon is determined.

[0039] Furthermore, the training method of the monitoring anomaly analysis model includes:

[0040] E groups of training data are collected in advance, each group of training data includes light intensity and image frames corresponding to the light intensity, and E is an integer greater than 1; the training data is used as the input of a monitoring anomaly analysis model, and the monitoring anomaly analysis model uses the light intensity corresponding to each group of image frames as output, and the actual light intensity corresponding to each group of image frames as a prediction target; minimizing the sum of prediction errors of all light intensities is used as a training target; the monitoring anomaly analysis model is trained until the sum of prediction errors converges and the training is stopped, and the monitoring anomaly analysis model adopts a convolutional neural network.

[0041] The intelligent park security risk perception system with predictive capability is used for the intelligent park security risk perception method with predictive capability, comprising:

[0042] The first collection module is used to collect facial feature data of people registered for the park access control system;

[0043] The second acquisition module is used to collect video data of T monitoring areas in the park in real time;

[0044] The first analysis module extracts the real-time facial feature data appearing in the video data of the t-th monitoring area, t∈T, and compares the real-time facial feature data with the registered facial feature data to obtain a comparison result; the comparison result includes the presence of an entry record and the absence of an entry record, and the real-time facial feature data of the person without an entry record is marked as an abnormal person feature;

[0045] Setting module, pre-set the deliberate avoidance area for each monitoring area;

[0046] The second analysis module uses a tracking algorithm to track abnormal persons with abnormal personnel characteristics, and records the deliberate avoidance areas where the abnormal personnel characteristics appear in q monitoring areas, marking them as passing avoidance areas, and connects all the recorded passing avoidance areas in sequence to form the movement trajectory of the abnormal person, where q is less than or equal to T;

[0047] The third analysis module intercepts the image frames where abnormal persons appear in the video data, inputs them into the action recognition model, obtains the avoidance actions of the abnormal persons, and counts the number of avoidance actions per unit time;

[0048] The fourth analysis module collects the light intensity and humidity at the current moment, combines them with the number of avoidance actions, and generates an avoidance action coefficient;

[0049] A fifth analysis module, used to obtain movement feature data and height data of abnormal persons based on video data;

[0050] A comprehensive analysis module is used to input the action trajectory, avoidance action coefficient, movement feature data and height data into the behavior feature analysis model to obtain the behavior analysis results;

[0051] The early warning module issues an early warning if the behavior analysis result is abnormal behavior.

[0052] The present invention provides a smart park security risk perception system and method with predictive capabilities, and has the following technical effects and advantages:

[0053] 1. By inputting multiple behavioral characteristics into the behavioral characteristic analysis model, the behavior of abnormal personnel can be comprehensively and accurately evaluated; each characteristic reflects the behavioral pattern and potential risks of personnel from different angles, and its connection with the behavioral analysis results is close and complex; through in-depth analysis of these connections, the model can more effectively identify abnormal behavior, improve the park's security risk perception ability, and effectively reduce false alarms. High-accuracy early warnings not only improve the efficiency of security prevention, but also reduce the impact of park managers' passive sabotage.

[0054] 2. This embodiment aims to prevent the video surveillance system from being invaded and play pre-recorded videos, so as to avoid losing the function of warning abnormal behavior of abnormal personnel; by detecting the abnormality of video data, when an abnormality is detected, the system will enable a backup server to analyze the real-time collected video data, improve system security, and prevent intrusion attacks.

[0055] By collecting wind speed and direction data and light intensity in real time, the system can sense whether the video data is consistent with the actual environmental conditions; when the offset direction of the flexible ribbon in the video screen is inconsistent with the actual wind direction, or the light intensity does not match the image frame, the system determines that the video may have been tampered with or replaced, and generates activation instructions in time; multiple verification mechanisms effectively prevent hackers from deceiving the monitoring system by playing pre-recorded videos, enhancing the system's ability to resist intrusion and tampering.

[0056] By detecting anomalies in video data, when anomalies are detected, the system automatically enables the backup server for in-depth analysis through the intelligent switching of the data switching switch, ensuring the continuity of monitoring and analysis under abnormal circumstances, ensuring that the monitoring system can continuously and accurately identify and warn abnormal personnel, avoiding monitoring blind spots caused by video replacement, and ensuring the effective implementation of the park's safety precautions. BRIEF DESCRIPTION OF THE DRAWINGS

[0057] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, the drawings required for use in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments recorded in the present invention. For ordinary technicians in this field, other drawings can also be obtained based on these drawings.

[0058] Figure 1 This is a schematic diagram of the intelligent park security risk perception system with prediction capability in Example 1 of the present invention;

[0059] Figure 2 This is a schematic diagram of a smart park security risk perception system with predictive capabilities in Example 2 of the present invention;

[0060] Figure 3 This is a flow chart of the smart park security risk perception method with predictive capability in Example 3 of the present invention. DETAILED DESCRIPTION

[0061] In order to make the purpose, technical solution and advantages of the embodiments of the present invention clearer, the technical solution in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0062] It should be noted that when an element is referred to as being "fixed to" another element, it may be directly on the other element or there may be an intermediate element. When an element is considered to be "connected to" another element, it may be directly connected to the other element or there may be an intermediate element at the same time.

[0063] Example 1

[0064] like Figure 1 As shown, the smart park security risk perception system and method with predictive capabilities in this embodiment include a first acquisition module, a second acquisition module, a first analysis module, a setting module, a second analysis module, a third analysis module, a fourth analysis module, a fifth analysis module and a comprehensive analysis module, and each module is connected by wire and / or wirelessly.

[0065] The first collection module is used to collect facial feature data of people registered for the park access control system and store it in a registration database.

[0066] The second acquisition module is used to collect video data of T monitoring areas in the park in real time.

[0067] The first analysis module extracts the real-time facial feature data appearing in the video data of the t-th monitoring area, t∈T, and compares the real-time facial feature data with the registered facial feature data to obtain a comparison result; the comparison result includes the existence of entry records and the non-existence of entry records, and the real-time facial feature data without entry records is marked as abnormal personnel features.

[0068] Methods for extracting real-time facial feature data include:

[0069] By using personnel recognition software, the area with human faces in the video data frame number image of the t-th monitoring area is intercepted to obtain real-time facial feature data.

[0070] The setting module pre-sets the deliberate avoidance area for each monitoring area. The deliberate avoidance area is the edge area or blocked area of ​​the camera equipment in the monitoring area. It is set by the staff according to the environment of the specific monitoring area. The reason is that the sensitivity in the blocked or edge area may be lower, and abnormal personnel can use this to avoid the automatic alarm mechanism.

[0071] The second analysis module analyzes the video data of T monitoring areas in real time, uses algorithms such as SORT and DeepSORT to track abnormal persons with abnormal personnel characteristics, and records the deliberate avoidance areas where the abnormal personnel characteristics appear in q monitoring areas, marking them as passing avoidance areas, and connects all the recorded passing avoidance areas in sequence to form the movement trajectory of the abnormal personnel, where q is less than or equal to T.

[0072] Methods for obtaining the movement trajectory of abnormal persons include:

[0073] The actual coordinates of all the passing avoidance areas are mapped to the electronic map of the park, and all the passing avoidance areas are connected in sequence on the electronic map to form the movement trajectory of abnormal personnel.

[0074] The third analysis module intercepts the image frames of abnormal persons appearing in the video data, inputs them into the action recognition model, obtains the avoidance actions of the abnormal persons, the avoidance actions include lowering the head and covering the face, and counts the number of times the abnormal persons lower their heads and cover their faces in a unit time.

[0075] The fourth analysis module collects the light intensity and humidity at the current moment, combines it with the number of times the head is lowered and the face is covered per unit time, and generates an avoidance action coefficient.

[0076] The avoidance action coefficient generation includes:

[0077]

[0078] In the formula, All are preset weight coefficients; XS represents the avoidance action coefficient, the larger the value, the greater the abnormality of the abnormal person; DT represents the number of times the head is lowered per unit time; ZD represents the number of times the face is blocked per unit time; GZ represents the light intensity at the current moment; SD represents the humidity at the current moment; the stronger the light intensity, the more avoidance actions will be performed when blocking the sunlight; similarly, the larger the humidity value, the more avoidance actions will be performed when blocking rain or fog. In order to avoid the influence of light intensity and humidity, the light intensity and humidity are weighted and used as the denominator. The generated avoidance action coefficient can more truly reflect the abnormality of the abnormal person and highlight the real abnormal behavior, thereby reducing the interference of other environmental factors.

[0079] The fifth analysis module is used to obtain the movement characteristic data and height data of abnormal persons based on the video data, and the movement characteristic data includes the average movement speed and the standard deviation of the movement speed.

[0080] The method of extracting the movement characteristic data of abnormal persons includes:

[0081] The position coordinates of the abnormal persons in the continuous images of the video data are mapped to the electronic map of the park to obtain the continuous position coordinate sequence of the abnormal persons in the unit interval time.

[0082] Calculate the moving speed S within the unit time interval:

[0083]

[0084] X i+1 Indicates the horizontal coordinate of the i+1th position, X i Represents the horizontal coordinate of the i-th position, Y i+1 Indicates the ordinate of the i+1th position, Y i Represents the ordinate of the i-th position coordinate.

[0085] The calculated moving speeds are used to establish a moving speed set, and the moving speed average and moving speed standard deviation in the moving speed set are calculated.

[0086] Methods for obtaining abnormal person height data include:

[0087] Using video processing technology, obtain the top and bottom pixel coordinates of the reference object and the abnormal person in the video data image;

[0088] According to the top and bottom pixel coordinates of the reference object, the height of the reference object in the video data image is obtained; similarly, according to the top and bottom pixel coordinates of the abnormal person, the height of the abnormal person in the video data image is obtained; the ratio of the height of the reference object in the video data image to the actual height of the reference object is marked as a scale factor; the height of the abnormal person in the video data image is multiplied by the scale factor to obtain the height data of the abnormal person.

[0089] The comprehensive analysis module is used to input the action trajectory, avoidance action coefficient, movement feature data and height data into the behavior feature analysis model to obtain the behavior analysis results.

[0090] The early warning module issues an early warning if the behavior analysis result is abnormal behavior; and does not issue an early warning if the behavior analysis result is normal behavior.

[0091] The action trajectory refers to the movement path of abnormal personnel in the monitoring area, which consists of a series of space-time coordinate points.

[0092] The action trajectory reflects the direction of movement of abnormal persons, the areas they pass through, and whether they choose routes commonly used by the public; the avoidance action coefficient quantifies the degree to which abnormal persons perform avoidance actions in a unit of time, indicating that the person intentionally avoids the camera, hides his identity, and has potential illegal intentions; moving too fast, higher than the normal moving speed, means that the person is eager to leave the scene or evade tracking; moving too slowly, lower than the normal moving speed, may be observing the environment, looking for targets, or avoiding attention. Speed ​​changes: A stable speed may indicate normal activities, and drastic changes indicate abnormal behavior; a high standard deviation and large speed fluctuations may indicate avoiding surveillance, observing the environment, or looking for opportunities; a low standard deviation and stable speed may indicate performing specific tasks or having clear goals; secondly, the height of abnormal persons is considered to reduce the impact of height on movement speed and improve the prediction accuracy of the feature analysis model.

[0093] The training methods of the feature analysis model include:

[0094] K groups of action trajectories, avoidance action coefficients, abnormal personnel heights, and mobile feature data are collected in advance and combined into K groups of feature vectors, where K is an integer greater than 1. Supervised learning methods, such as support vector machine (SVM), random forest, deep neural network, etc., are used to train the behavior feature analysis model. The normal behaviors and abnormal behaviors pre-labeled in the feature vectors constitute a data set of known behaviors. The data set is divided into a training set and a test set for feature analysis model training and verification, and a feature analysis model that meets the preset accuracy is output.

[0095] By inputting multiple behavioral characteristics into the behavioral characteristic analysis model, the behavior of abnormal personnel can be comprehensively and accurately evaluated; each characteristic reflects the behavioral patterns and potential risks of personnel from different angles, and its connection with the behavioral analysis results is close and complex; through in-depth analysis of these connections, the model can more effectively identify abnormal behavior, enhance the park's security risk perception capabilities, and effectively reduce false alarms. High-accuracy early warnings not only improve the efficiency of security prevention, but also reduce the impact of passive sabotage on park managers.

[0096] Example 2

[0097] See also Figure 2 As shown, this embodiment provides a smart park security risk perception system with prediction capability, and also includes:

[0098] The third acquisition module is used to collect wind speed and direction data in each monitoring area, specifically through an anemometer or ultrasonic anemometer installed in each monitoring area.

[0099] The monitoring anomaly analysis module analyzes whether the video data of each monitoring area is abnormal based on wind speed and direction data or light intensity. If it is abnormal, an enable instruction is generated. If it is normal, no enable instruction is generated, and the enable instruction is sent to the backup server and the data switching switch. The backup server can complete real-time analysis of the video data. The backup server is turned on according to the enable instruction, the input end of the data switching switch is connected to the data output line of the video acquisition device, and the output end of the data switching switch is connected to the analysis server and the backup server respectively. When the data switching switch receives the enable instruction, the video data is transmitted to the backup server for analysis according to the enable instruction.

[0100] Methods for analyzing whether the video data of each monitoring area is abnormal based on wind speed and direction data or light intensity include:

[0101] When the wind force level in the rth monitoring area is greater than the preset wind force level, the offset direction of the flexible ribbon in the video data of the rth monitoring area is analyzed to see whether it is consistent with the wind direction. If not, an activation instruction is generated; if consistent, no activation instruction is generated.

[0102] If the wind force level is less than or equal to the preset wind force level, it means there is no wind or the wind force is small, which will cause the flexible ribbon to deviate by a limited range and easily cause the erroneous generation of the activation instruction. Therefore, when the wind force level is less than or equal to the preset wind force level, the real-time collected light intensity and the real-time image frame in the video data of the rth monitoring area are input into the monitoring anomaly analysis model to determine whether the image frame is abnormal. If the image frame is abnormal, an activation instruction is generated; if the image frame is normal, no activation instruction is generated.

[0103] The method for analyzing the deviation direction of the flexible ribbon in the video data of the r-th monitoring area includes:

[0104] In the display screen of the video data, a plane coordinate system is preset, and the fixed end of the flexible ribbon is used as the origin of the plane coordinate system. The coordinates of each pixel point of the area occupied by the flexible ribbon in the display screen at the current moment are obtained, and the quadrant of the plane coordinate system where the flexible ribbon is located is determined according to the coordinates of each pixel point. According to the preset indication direction of the quadrant, the offset direction of the flexible ribbon is determined.

[0105] First quadrant: Both x and y are positive and located to the upper right of the origin.

[0106] Second quadrant: x is negative, y is positive, and is located to the upper left of the origin.

[0107] The third quadrant: Both x and y are negative values, located to the lower left of the origin.

[0108] The fourth quadrant: x is positive, y is negative, and is located to the lower right of the origin.

[0109] The method for determining the quadrant of the plane coordinate system where the flexible ribbon is located according to the coordinates of each pixel point includes:

[0110] The quadrant where the largest number of pixels are located is taken as the quadrant where the flexible ribbon is located.

[0111] The training methods for monitoring anomaly analysis models include:

[0112] Collect E groups of training data in advance, each group of training data includes light intensity and image frames corresponding to the light intensity, and E is an integer greater than 1; use the training data as the input of a monitoring anomaly analysis model, the monitoring anomaly analysis model uses the light intensity corresponding to each group of image frames as output, and uses the actual light intensity corresponding to each group of image frames as a prediction target; minimize the sum of prediction errors of all light intensities as a training target; train the monitoring anomaly analysis model until the sum of prediction errors converges and stops training, the monitoring anomaly analysis model uses a deep learning model (such as a convolutional neural network) to predict light intensity from image frames.

[0113] This embodiment aims to prevent the video surveillance system from being invaded and play pre-recorded videos, so as to avoid losing the function of warning abnormal behavior of abnormal personnel; by detecting the abnormality of video data, when an abnormality is detected, the system will enable a backup server to analyze the real-time collected video data, improve system security, and prevent intrusion attacks.

[0114] By collecting wind speed and direction data and light intensity in real time, the system can sense whether the video data is consistent with the actual environmental conditions; when the offset direction of the flexible ribbon in the video screen is inconsistent with the actual wind direction, or the light intensity does not match the image frame, the system determines that the video may have been tampered with or replaced, and generates activation instructions in time; multiple verification mechanisms effectively prevent hackers from deceiving the monitoring system by playing pre-recorded videos, enhancing the system's ability to resist intrusion and tampering.

[0115] By detecting anomalies in video data, when anomalies are detected, the system automatically enables the backup server for in-depth analysis through the intelligent switching of the data switching switch, ensuring the continuity of monitoring and analysis under abnormal circumstances, ensuring that the monitoring system can continuously and accurately identify and warn abnormal personnel, avoiding monitoring blind spots caused by video replacement, and ensuring the effective implementation of the park's safety precautions.

[0116] Example 3

[0117] like Figure 3 As shown, the intelligent park security risk perception method with prediction capability in this embodiment includes:

[0118] Collect facial feature data of people registered at the park access control;

[0119] Collect video data of T monitoring areas in the park in real time;

[0120] Extract the real-time facial feature data appearing in the video data of the t-th monitoring area, t∈T, and compare the real-time facial feature data with the registered facial feature data to obtain a comparison result; the comparison result includes the presence of an entry record and the absence of an entry record, and mark the real-time facial feature data of the person without an entry record as an abnormal person feature;

[0121] Pre-set deliberate avoidance areas for each monitoring area;

[0122] Use the tracking algorithm to track the abnormal person with abnormal personnel characteristics, and record the deliberate avoidance area where the abnormal personnel characteristics appear in q monitoring areas, mark it as the passing avoidance area, and connect all the recorded passing avoidance areas in sequence to form the action trajectory of the abnormal person, q is less than or equal to T;

[0123] Intercept the image frames where abnormal persons appear in the video data, input them into the action recognition model, obtain the avoidance actions of the abnormal persons, and count the number of avoidance actions per unit time;

[0124] Collect the light intensity and humidity at the current moment, combine them with the number of avoidance actions, and generate the avoidance action coefficient;

[0125] Obtain movement feature data and height data of abnormal persons based on video data;

[0126] Input the action trajectory, avoidance action coefficient, movement feature data and height data into the behavior feature analysis model to obtain the behavior analysis results;

[0127] If the behavior analysis result is abnormal behavior, an early warning will be issued.

[0128] The above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit the same. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that the technical solutions described in the aforementioned embodiments may still be modified, or some of the technical features may be replaced by equivalents. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A smart park security risk perception method with predictive capabilities, characterized in that: include: Collect facial feature data of people registered at the park access control; Collect video data of T monitoring areas in the park in real time; Extract the real-time facial feature data appearing in the video data of the t-th monitoring area, t∈T, and compare the real-time facial feature data with the registered facial feature data to obtain a comparison result; the comparison result includes the presence of an entry record and the absence of an entry record, and mark the real-time facial feature data of the person without an entry record as an abnormal person feature; Pre-set deliberate avoidance areas for each monitoring area; Use the tracking algorithm to track the abnormal person with abnormal personnel characteristics, and record the deliberate avoidance areas where the abnormal personnel characteristics appear in q monitoring areas, marking them as passing avoidance areas, and connect all the recorded passing avoidance areas in sequence to form the movement trajectory of the abnormal person, q is less than or equal to T; Intercept the image frames where abnormal persons appear in the video data, input them into the action recognition model, obtain the avoidance actions of the abnormal persons, and count the number of avoidance actions per unit time; Collect the light intensity and humidity at the current moment, combine them with the number of avoidance actions, and generate the avoidance action coefficient; Obtain movement feature data and height data of abnormal persons based on video data; Input the action trajectory, avoidance action coefficient, movement feature data and height data into the behavior feature analysis model to obtain the behavior analysis results; If the behavior analysis result is abnormal behavior, an early warning will be issued.

2. The method for sensing security risks in an intelligent park with predictive capability according to claim 1 is characterized in that: The avoidance actions include lowering the head and covering the face, and the number of times of lowering the head and covering the face in a unit time is counted; The avoidance action coefficient generation includes: The avoidance action value is obtained by weighted summing up the number of times of lowering the head and covering the face in unit time, the environmental impact value is obtained by weighted summing up the light intensity and humidity at the current moment, and the avoidance action value is divided by the environmental impact value to obtain the avoidance action coefficient.

3. The method for sensing security risks in an intelligent park with predictive capability according to claim 1 is characterized in that: The method of extracting the movement characteristic data of abnormal persons includes: The moving characteristic data include the moving speed mean and moving speed standard deviation; Mapping the position coordinates of the abnormal person in the continuous images in the video data to the electronic map of the park to obtain a continuous position coordinate sequence of the abnormal person in the unit interval time; Calculate the moving speed S within the unit time interval: X i+1 Indicates the horizontal coordinate of the i+1th position, X i Indicates the horizontal coordinate of the i-th position, Y i+1 Indicates the ordinate of the i+1th position, Y i Represents the ordinate of the i-th position coordinate; The calculated moving speeds are used to establish a moving speed set, and the moving speed average and moving speed standard deviation in the moving speed set are calculated.

4. The method for sensing security risks in an intelligent park with predictive capability according to claim 1 is characterized in that: Methods for obtaining abnormal person height data include: Using video processing technology, obtain the top and bottom pixel coordinates of the reference object and the abnormal person in the video data image; Obtaining the height of the reference object in the video data image according to the top and bottom pixel coordinates of the reference object; Similarly, the height of the abnormal person in the video data image is obtained according to the top and bottom pixel coordinates of the abnormal person; The ratio of the height of the reference object in the video data image to the actual height of the reference object is marked as a scale factor; the height of the abnormal person in the video data image is multiplied by the scale factor to obtain the height data of the abnormal person.

5. The method for sensing security risks in an intelligent park with predictive capability according to claim 1 is characterized in that: The training methods of the feature analysis model include: K groups of action trajectories, avoidance action coefficients, abnormal personnel heights, and mobile feature data are collected in advance and combined into K groups of feature vectors, where K is an integer greater than 1. A supervised learning method is used to train a behavioral feature analysis model using a support vector machine, random forest, or deep neural network. The normal behaviors and abnormal behaviors pre-labeled in the feature vectors form a data set of known behaviors. The data set is divided into a training set and a test set for feature analysis model training and verification, and a feature analysis model that meets the preset accuracy is output.

6. The method for sensing security risks in an intelligent park with predictive capabilities according to claim 1, characterized in that: Also includes: Collect wind speed and direction data for each monitoring area; According to the wind speed and direction data or the light intensity, the video data of each monitoring area is analyzed to see if it is abnormal. If it is abnormal, an enable instruction is generated. If it is normal, no enable instruction is generated, and the enable instruction is sent to the backup server and the data switching switch. The backup server can complete the real-time analysis of the video data. The backup server is turned on according to the enable instruction, and the data switching switch transmits the video data to the backup server for analysis according to the enable instruction.

7. The method for sensing security risks in an intelligent park with predictive capability according to claim 6 is characterized in that: Methods for analyzing whether the video data of each monitoring area is abnormal based on wind speed and direction data or light intensity include: When the wind force level of the r-th monitoring area is greater than the preset wind force level, the flexible ribbon offset direction in the video data of the r-th monitoring area is analyzed to see whether it is consistent with the wind direction. If not, an activation instruction is generated; if consistent, no activation instruction is generated; If the wind force level is less than or equal to the preset wind force level, the real-time collected light intensity and the real-time image frame in the video data of the rth monitoring area are input into the monitoring anomaly analysis model to determine whether the image frame is abnormal. If the image frame is abnormal, an enable instruction is generated; if the image frame is normal, no enable instruction is generated.

8. The method for sensing security risks in an intelligent park with predictive capability according to claim 7 is characterized in that: The method for analyzing the deviation direction of the flexible ribbon in the video data of the r-th monitoring area includes: In the display screen of the video data, a plane coordinate system is preset, and the fixed end of the flexible ribbon is used as the origin of the plane coordinate system. The coordinates of each pixel point of the area occupied by the flexible ribbon in the display screen at the current moment are obtained, and the quadrant of the plane coordinate system where the flexible ribbon is located is determined according to the coordinates of each pixel point. According to the preset indication direction of the quadrant, the offset direction of the flexible ribbon is determined.

9. The method for sensing security risks in an intelligent park with predictive capability according to claim 7, characterized in that: The training method of the monitoring anomaly analysis model includes: E groups of training data are collected in advance, each group of training data includes light intensity and image frames corresponding to the light intensity, and E is an integer greater than 1; the training data is used as the input of a monitoring anomaly analysis model, and the monitoring anomaly analysis model uses the light intensity corresponding to each group of image frames as output, and the actual light intensity corresponding to each group of image frames as a prediction target; minimizing the sum of prediction errors of all light intensities is used as a training target; the monitoring anomaly analysis model is trained until the sum of prediction errors converges and the training is stopped, and the monitoring anomaly analysis model adopts a convolutional neural network.

10. A smart park security risk perception system with predictive capability, used to implement the smart park security risk perception method with predictive capability described in any one of 1-9, characterized in that: include: The first collection module is used to collect facial feature data of people registered for the park access control system and store it in the registration database; The second acquisition module is used to collect video data of T monitoring areas in the park in real time; The first analysis module extracts the real-time facial feature data appearing in the video data of the t-th monitoring area, t∈T, and compares the real-time facial feature data with the registered facial feature data to obtain a comparison result; the comparison result includes the presence of an entry record and the absence of an entry record, and the real-time facial feature data of the person without an entry record is marked as an abnormal person feature; Setting module, pre-set the deliberate avoidance area for each monitoring area; The second analysis module uses a tracking algorithm to track abnormal persons with abnormal personnel characteristics, and records the deliberate avoidance areas where the abnormal personnel characteristics appear in q monitoring areas, marking them as passing avoidance areas, and connects all the recorded passing avoidance areas in sequence to form the movement trajectory of the abnormal person, where q is less than or equal to T; The third analysis module intercepts the image frames where abnormal persons appear in the video data, inputs them into the action recognition model, obtains the avoidance actions of the abnormal persons, and counts the number of avoidance actions per unit time; The fourth analysis module collects the light intensity and humidity at the current moment, combines them with the number of avoidance actions, and generates an avoidance action coefficient; A fifth analysis module, used to obtain movement feature data and height data of abnormal persons based on video data; A comprehensive analysis module is used to input the action trajectory, avoidance action coefficient, movement feature data and height data into the behavior feature analysis model to obtain the behavior analysis results; The early warning module issues an early warning if the behavior analysis result is abnormal behavior.

Images