Disclosure of Invention
The method is a key field in the aspect of temporarily managing and processing behaviors of key personnel in key areas. However, the conventional method is often limited to a single feature representation and a limited prediction model, so that difficulty exists in accurately classifying or clustering the complex personnel information, and the multi-dimensional prediction system has large calculation amount in a large amount of data analysis processing process, high requirement on hardware and high analysis cost. Therefore, the invention aims to improve the analysis and early warning capacity of personnel in the key area by adopting a multidimensional prediction model and an abnormal behavior recognition model based on a laminated neural network, and realize the rapid analysis of the behaviors of the personnel in the key area.
In a first aspect, the present invention provides a method for analyzing personnel behaviors in a key area, including:
s100: acquiring a multidimensional risk identification model, a human body characteristic identification model and a target track tracking model;
acquiring the multi-dimensional risk identification model based on a stacked neural network, wherein the output of the multi-dimensional risk identification model is 1 or 0, wherein 1 represents that personnel have risks, and 0 represents that personnel do not have risks;
the human body characteristic recognition model comprises a human face characteristic recognition module and a morphological characteristic recognition module;
The target track tracking model comprises a track tracking module and a target matching module, the motion track of the personnel is estimated through the track tracking module, and the motion track of the personnel is updated through the target matching module, so that the continuous tracking of the motion track of the personnel is realized;
s200: collecting identity authentication information of a person entering a key area at an entrance of the key area, and dividing a management and control level of the person according to the identity authentication information, wherein the management and control level is a primary management and control person or a secondary management and control person;
the identity authentication information comprises: physiological characteristic parameters, face images and body state images;
s300: according to the control level of the personnel entering the key area, acquiring multi-dimensional characteristic information of the personnel entering the key area at the entrance of the key area, wherein the multi-dimensional characteristic information is one type of multi-dimensional characteristic information or two types of multi-dimensional characteristic information;
acquiring a type of multidimensional characteristic information of the primary management and control personnel; the multi-dimensional characteristic information comprises: face feature information, posture feature information, occupation feature information and economic feature information;
acquiring second-class multidimensional characteristic information of the secondary control personnel; the second-class multidimensional feature information includes: face feature information, occupation feature information and economic feature information;
Inputting the face image and the body state image of the primary management and control personnel into a human body feature recognition model to obtain face feature information and body state feature information of the primary management and control personnel;
inputting the face image of the secondary control personnel into a human body characteristic recognition model to obtain face characteristic information of the secondary control personnel;
the job characteristic information includes a job of the person at the current point in time and a job within one year before the current point in time;
the economic characteristic information comprises income and expense of personnel at the current time point and income and expense of the personnel in a year before the current time point;
s400: tracking and analyzing the human body characteristic matching state and the motion trail of the personnel in the heavy point area through the human body characteristic recognition model and the target trail tracking model to acquire the motion trail of the personnel;
s500: inputting the motion trail, professional characteristic information and economic characteristic information of the personnel into a multidimensional risk identification model, and outputting a judgment result of whether the personnel has risks;
s600: and reporting the identity authentication information corresponding to the personnel with risk to a superior system by counting the personnel with risk.
Further, the step of obtaining the multi-dimensional risk identification model includes:
step 1: acquiring a training sample set, wherein each training sample comprises a motion trail of a person, professional characteristic information, economic characteristic information and multidimensional risk parameters;
step 2: training 3 neural networks, wherein the 3 neural networks are a first layer neural network, a second layer neural network and a third layer neural network respectively, the inputs of the first layer neural network, the second layer neural network and the third layer neural network are movement tracks, occupational characteristic information and economic characteristic information respectively, and multidimensional risk parameters are output respectively;
step 3: combining the trained 3 neural networks in a serial manner to form a laminated neural network;
step 4: training a laminated neural network, wherein the input of the laminated neural network is movement track, occupational characteristic information and economic characteristic information, and a multidimensional risk identification model is obtained.
Further, the 3 neural networks are connected in series in the order of the third layer neural network, the first layer neural network and the second layer neural network.
Further, the face feature recognition module is a face recognition algorithm based on a deep convolutional neural network, and the morphological feature recognition module is a feature extraction algorithm based on a generated countermeasure.
Further, the track tracking module adopts a multi-target path prediction algorithm based on a generated countermeasure model, and the target matching module adopts a nearest neighbor tracking algorithm.
Further, the S200 includes:
s201: collecting identity authentication information of a person entering a key area at an entrance of the key area, wherein the identity authentication information comprises: physiological characteristic parameters, face images and body state images;
s202: setting an upper limit value and a reference value of the physiological characteristic parameters of a normal human body;
s203: according to the physiological characteristic parameters, the upper limit value of the physiological characteristic parameters and the reference value, calculating personnel grading parameters, wherein the calculation formula of the personnel grading parameters is as follows:wherein:αclassifying parameters for personnel;Sphysiological characteristic parameters;Sc a reference value for physiological characteristic parameters;Sm an upper limit value of the physiological characteristic parameter;
s204: dividing the personnel into first-level management and control personnel and second-level management and control personnel according to the personnel grading parameters of the personnel;
when the personnel grading parameter of the personnel is more than 1, the personnel is a first-level management and control personnel; when the personnel grading parameter of the personnel is equal to 1, the personnel is a secondary management personnel.
Further, the physiological characteristic parameter is blood oxygen saturation or heart rate.
Further, the S400 includes:
s401: collecting an image to be detected in real time in a key area through a camera;
s402: inputting the image to be detected into a human body characteristic recognition model to obtain human face characteristic information and morphological characteristic information of a person in the image to be detected;
s403: matching the face characteristic information or the body state characteristic information acquired in the step S402 with the face characteristic information or the body state characteristic information in the multidimensional characteristic information, and identifying the person in the image to be detected as a primary management and control person or a secondary management and control person;
when the person in the image to be detected is a primary management and control person, acquiring face characteristic information and body characteristic information matched with the face characteristic information or body characteristic information acquired in the step S402 from the multi-dimensional characteristic information, and entering into the step S404;
when the person in the image to be detected is a diode control person, entering S405;
s404: human body characteristic matching state analysis of primary management and control personnel;
when the face characteristic information and the body characteristic information acquired from the multidimensional characteristic information and the S402 by the primary management and control personnel are consistent, entering S405;
when the face characteristic information and the body characteristic information acquired from the multidimensional characteristic information and the S402 are inconsistent, judging that the primary management and control personnel have risks, and entering S600;
S405: acquiring a motion trail of a person in an image to be detected;
and inputting the face characteristic information and the morphological characteristic information of the person in the image to be detected into a target track tracking model, and carrying out tracking analysis on the motion track of the person in the image to be detected to obtain the motion track of the person.
Further, in S404:
when the similarity of the face characteristic information and the body characteristic information obtained from the multidimensional characteristic information and the S402 by the primary management and control personnel is more than or equal to 90%, judging that the face characteristic information and the body characteristic information are consistent;
and when the similarity of the face characteristic information or the body characteristic information obtained from the multidimensional characteristic information and the S402 by the primary management and control personnel is less than 90%, judging that the face characteristic information or the body characteristic information is inconsistent.
In a second aspect, the present invention provides a system for analyzing personnel behavior in a key area, including: the system comprises a model training module, a personnel grading module, a multidimensional feature information acquisition module, a motion trail tracking module, a risk judging module and a risk reporting module;
the model training module is used for acquiring a multidimensional risk identification model, a human body characteristic identification model and a target track tracking model;
acquiring the multi-dimensional risk identification model based on a stacked neural network, wherein the output of the multi-dimensional risk identification model is 1 or 0, wherein 1 represents that personnel have risks, and 0 represents that personnel do not have risks;
The human body characteristic recognition model comprises a human face characteristic recognition module and a morphological characteristic recognition module;
the target track tracking model comprises a track tracking module and a target matching module, the motion track of the personnel is estimated through the track tracking module, and the motion track of the personnel is updated through the target matching module, so that the continuous tracking of the motion track of the personnel is realized;
the personnel grading module is used for collecting identity authentication information of personnel entering the key area at the entrance of the key area, and dividing the management and control grade of the personnel according to the identity authentication information, wherein the management and control grade is a primary management and control personnel or a secondary management and control personnel;
the identity authentication information comprises: physiological characteristic parameters, face images and body state images;
the multi-dimensional characteristic information acquisition module acquires multi-dimensional characteristic information of personnel entering the key area at the entrance of the key area according to the control level of the personnel entering the key area, wherein the multi-dimensional characteristic information is one type of multi-dimensional characteristic information or two types of multi-dimensional characteristic information;
acquiring a type of multidimensional characteristic information of the primary management and control personnel; the multi-dimensional characteristic information comprises: face feature information, posture feature information, occupation feature information and economic feature information;
Acquiring second-class multidimensional characteristic information of the secondary control personnel; the second-class multidimensional feature information includes: face feature information, occupation feature information and economic feature information;
inputting the face image and the body state image of the primary management and control personnel into a human body feature recognition model to obtain face feature information and body state feature information of the primary management and control personnel;
inputting the face image of the secondary control personnel into a human body characteristic recognition model to obtain face characteristic information of the secondary control personnel;
the job characteristic information includes a job of the person at the current point in time and a job within one year before the current point in time;
the economic characteristic information comprises income and expense of personnel at the current time point and income and expense of the personnel in a year before the current time point;
the motion trail tracking module is used for carrying out tracking analysis on the human body characteristic matching state and the motion trail of the personnel in the heavy point area through the human body characteristic recognition model and the target trail tracking model to acquire the motion trail of the personnel;
the risk judging module inputs the motion trail, professional characteristic information and economic characteristic information of the personnel into a multidimensional risk identification model and outputs a judging result of whether the personnel has risks or not;
And the risk reporting module is used for reporting the identity authentication information corresponding to the personnel with risk to a superior system when the statistical judgment result is the personnel with risk.
The invention has the beneficial effects that:
(1) According to the method for analyzing the behaviors of personnel in the key areas, the behaviors of the personnel entering the key areas are analyzed from multiple dimensions of social tracks, professional states and economic states, a multi-dimensional representation framework is utilized, extraction models are respectively used for constructing theme dimensions such as social, employment and economy, fusion judgment of multi-dimensional risks is achieved by designing personnel risk identification models based on stacked neural networks, accurate classification or clustering of complex personnel information is achieved, a localization mechanism is trained by researching models, the models can be updated and iterated continuously, and accuracy is improved.
(2) In order to improve the recognition precision of risk personnel, when acquiring face images of personnel, in order to prevent the behaviors of using a deception system such as photos, videos and the like, physiological characteristic parameters are introduced into the system, physiological characteristic parameters of the personnel are acquired while human body characteristics (faces, body states) are acquired, normal and abnormal early-stage pre-judgment is carried out on the behaviors of the personnel based on the physiological characteristic parameters, when the physiological characteristic parameters of an operating personnel are abnormal, the abnormal behaviors of the operating personnel can be initially judged to be abnormal, the abnormal personnel are divided into primary management and control personnel, in the follow-up track tracking process, the human body characteristics of the primary management and control personnel are continuously compared with the human body characteristics acquired when entering an important area, when the human body characteristics of the personnel are found to be inconsistent with the human body characteristics acquired when entering the important area in the follow-up real-time detection process, the human body characteristics of the personnel can be recognized to be the human face image behaviors, the personnel can be directly judged to be camouflaged as the personnel with risks, the judgment efficiency of the risk personnel is further improved, and the safety of the important area is ensured.
(3) Based on a risk identification model of a stacked network, firstly, training a neural network model of each single dimension, combining a plurality of neural networks in a serial connection mode after the single-layer network is trained, and when the plurality of neural networks are combined, as the measurement mode of each dimension to the risk is different, and the contribution degree of each dimension to final risk parameters is also different, the application provides a multi-dimensional fusion mechanism.
Description of the embodiments
The technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present application; it is apparent that the described embodiments are only some embodiments of the present application, not all embodiments, and that all other embodiments obtained by persons of ordinary skill in the art without making creative efforts based on the embodiments in the present application are within the protection scope of the present application.
The embodiment of the application provides a method and a system for analyzing personnel behaviors in a key area, and particularly relates to a method for analyzing personnel behaviors in a key area, which comprises the following steps of:
s100: acquiring a multidimensional risk identification model, a human body characteristic identification model and a target track tracking model;
acquiring a multidimensional risk identification model based on a laminated neural network, wherein the output of the multidimensional risk identification model is 1 or 0, wherein 1 represents that personnel have risks, and 0 represents that personnel do not have risks;
The human body characteristic recognition model comprises a human face characteristic recognition module and a body characteristic recognition module;
the target track tracking model comprises a track tracking module and a target matching module, wherein the track tracking module is used for estimating the motion track of the personnel, and the target matching module is used for updating the motion track of the personnel so as to realize the continuous tracking of the motion track of the personnel;
s200: collecting identity authentication information of a person entering the key area at an entrance of the key area, and dividing the management and control level of the person according to the identity authentication information, wherein the management and control level is a primary management and control person or a secondary management and control person;
the identity authentication information includes: physiological characteristic parameters, face images and body state images;
s300: according to the management and control level of the personnel entering the key area, acquiring multi-dimensional characteristic information of the personnel entering the key area at the entrance of the key area, wherein the multi-dimensional characteristic information is one type of multi-dimensional characteristic information or two types of multi-dimensional characteristic information;
acquiring a type of multidimensional characteristic information of a first-level management and control person; one type of multidimensional feature information includes: face feature information, posture feature information, occupation feature information and economic feature information;
acquiring second-class multidimensional characteristic information of a secondary management and control person; the second-class multidimensional feature information includes: face feature information, occupation feature information and economic feature information;
Inputting the face image and the body state image of the primary management and control personnel into a human body feature recognition model to obtain face feature information and body state feature information of the primary management and control personnel;
inputting the face image of the secondary control personnel into a human body characteristic recognition model to obtain the face characteristic information of the secondary control personnel;
the occupation characteristic information includes occupation of the person at the current time point and occupation within one year before the current time point;
the economic characteristic information comprises income and expense of personnel at the current time point and income and expense of the personnel in one year before the current time point;
s400: the human body characteristic matching state and the motion trail of the personnel in the heavy point area are tracked and analyzed through the human body characteristic recognition model and the target trail tracking model, and the motion trail of the personnel is obtained;
s500: inputting the motion trail, occupational feature information and economic feature information of the personnel into a multidimensional risk identification model, and outputting a judgment result of whether the personnel has risks or not;
s600: and reporting the identity authentication information corresponding to the personnel with risk to a superior system by counting the personnel with risk.
According to the embodiment, the behavior analysis is carried out on the personnel entering the key areas from a plurality of dimensions of social tracks, professional states and economic states, the multidimensional representation framework is utilized, the extraction models are respectively used for constructing theme dimensions of social, employment, economy and the like, and fusion judgment of multidimensional risks is achieved by designing a personnel risk identification model based on a laminated neural network, so that accurate classification or clustering of complex personnel information is achieved, a localization mechanism is trained by researching the model, the model can be updated and iterated continuously, and accuracy is improved.
In order to improve the recognition precision of risk personnel, when acquiring face images of personnel, in order to prevent the behaviors of using a deception system such as photos, videos and the like, physiological characteristic parameters are introduced into the system, physiological characteristic parameters of the personnel are acquired while human body characteristics (faces, body states) are acquired, normal and abnormal early-stage pre-judgment is carried out on the behaviors of the personnel based on the physiological characteristic parameters, when the physiological characteristic parameters of an operating personnel are abnormal, the abnormal behaviors of the operating personnel can be initially judged to be abnormal, the abnormal personnel are divided into primary management and control personnel, in the follow-up track tracking process, the human body characteristics of the primary management and control personnel are continuously compared with the human body characteristics acquired when entering an important area, when the human body characteristics of the personnel are found to be inconsistent with the human body characteristics acquired when entering the important area in the follow-up real-time detection process, the human body characteristics of the personnel can be recognized to be the human face image behaviors, the personnel can be directly judged to be camouflaged as the personnel with risks, the judgment efficiency of the risk personnel is further improved, and the safety of the important area is ensured.
The multidimensional risk recognition model, the human body characteristic recognition model and the target track tracking model are obtained through the step S100 and are used for subsequent personnel track recognition and personnel risk judgment.
In order to accurately and real-timely identify high-risk personnel, the multi-dimensional risk identification model is a basis for realizing multi-dimensional risk identification, and the embodiment provides the multi-dimensional risk identification model, so that the high-risk personnel can be accurately identified, the workload of manually checking data is reduced, and meanwhile, the multi-dimensional risk identification model has the characteristics of small training sample size and high real-time performance.
In this embodiment, the multi-dimensional risk recognition model adopts a stacked neural network model, as shown in fig. 2, and the step of obtaining the multi-dimensional risk recognition model includes:
step 1: acquiring a training sample set, wherein each training sample comprises a motion trail of a person, professional characteristic information, economic characteristic information and multidimensional risk parameters;
step 2: training 3 neural networks, wherein the 3 neural networks are a first layer of neural network, a second layer of neural network and a third layer of neural network respectively, the inputs of the first layer of neural network, the second layer of neural network and the third layer of neural network are motion trail, occupational characteristic information and economic characteristic information respectively, and multidimensional risk parameters are output respectively;
step 3: combining the trained 3 neural networks in a serial manner to form a laminated neural network;
step 4: training the laminated neural network, wherein the input of the laminated neural network is motion trail, occupational characteristic information and economic characteristic information, and obtaining a multidimensional risk identification model.
Specifically, in step 2, a BP neural network is adopted to build three network models with three layers, wherein the 3 neural networks are respectively a first layer neural network, a second layer neural network and a third layer neural network, and each network model comprises: the method comprises the steps of firstly, respectively inputting track offset risk parameters, occupational risk parameters and economic risk parameters of a preprocessed training sample set into an input layer, outputting a value at an output end through the action of an action function of the calculation layer, wherein the output value is 1 or 0, 1 represents that the person is at risk, 0 represents that the person is not at risk, calculating an error according to an error calculation function, and if the requirement of the expected error is not met, feeding the information just output back to the input layer and the calculation layer, acting on the input layer and the calculation layer, and adjusting the weight and related functions to achieve the effect of optimizing the output value.
And 3, combining a plurality of neural networks in a serial connection mode, and training the combined multi-layer neural network through the step 4 to realize the fusion of the multi-dimensional risk judgment factors.
When the multi-layer neural networks are combined, as the measurement modes of each dimension on the risk are different, and the contribution degree of each dimension on the final risk parameter is also different, the application provides a multi-dimensional fusion mechanism, and according to the influence of different dimensions on the risk parameter, the positions of the neural networks in the layer classifier are arranged, and through analysis and verification of a large amount of data, the influence of different dimensions on the risk parameter is in the following order from the judgment of three dimensions of social track, occupation state and economic state on the risk factors of personnel: economic status > social track > professional status.
For the above reasons, in this embodiment, the 3 neural networks are connected in series in the order of the third-layer neural network, the first-layer neural network, and the second-layer neural network.
Based on a risk identification model of a stacked network, firstly, training a neural network model of each single dimension, combining a plurality of neural networks in a serial connection mode after the single-layer network is trained, and when the plurality of neural networks are combined, as the measurement mode of each dimension to the risk is different, and the contribution degree of each dimension to final risk parameters is also different, the application provides a multi-dimensional fusion mechanism.
The human body characteristic recognition module is used for recognizing human body characteristics of people in the acquired images, in order to ensure recognition accuracy, the human face characteristic recognition module is a human face recognition algorithm based on a deep convolutional neural network, and the morphological characteristic recognition module is a characteristic extraction algorithm based on generation countermeasure.
The face recognition algorithm based on the neural convolutional neural network is mature, the extraction accuracy is high, and detailed description of the embodiment is omitted here.
The morphological feature recognition module is a feature extraction algorithm based on the generation of the countermeasure, and compared with a general deep learning feature extraction method, the morphological feature recognition module generates new data by generating the countermeasure, so that a network of feature extraction reduces intra-class feature variation among the same images to the greatest extent and distinguishes inter-class features among different images. In this embodiment, the encoder is used as a backbone network for recognition learning, and the physical characteristics of the person are learned by using images generated under different conditions.
In order to realize the actual situation of motion to be considered when the motion trail of a plurality of people is predicted, the motion trail of target people is also affected by the activities of surrounding people.
The multi-target path prediction algorithm based on the generated countermeasure model is based on the encoder/decoder structure of the generated countermeasure, and a pooling module is provided for simulating the interaction between pedestrians, and the target personnel prediction track is finally obtained through MLP and maximum pooling processing.
The track tracking module consists of a generator, a pooling module and a discriminator, wherein the generator is based on an LSTM framework for encoding and decoding, the pooling module is used for connecting the hidden states of encoding and decoding, and finally the discriminator is used for judging whether the track is true or not.
And the target matching module adopts a nearest neighbor tracking algorithm to track the motion trail of the personnel in the image.
Step 200, collecting identity authentication information of personnel entering a key area, when collecting the identity authentication information of the personnel, in order to prevent behaviors of using a cheating system such as a photo, a video and the like, the system introduces physiological characteristic parameters, collects physiological characteristic parameters of the personnel while collecting human body characteristics (human face and body state), and performs early pre-judgment on normal and abnormal behaviors of the personnel based on the physiological characteristic parameters, when the physiological characteristic parameters of an operating personnel are abnormal, the operating personnel can be primarily judged to have abnormal behaviors, and the abnormal personnel are classified as first-level management and control personnel.
As shown in fig. 3, S200 includes:
s201: collecting identity authentication information of a person entering a key area at an entrance of the key area, wherein the identity authentication information comprises: physiological characteristic parameters, face images and body state images;
S202: setting an upper limit value and a reference value of physiological characteristic parameters of a normal human body;
s203: according to the physiological characteristic parameters, the upper limit value of the physiological characteristic parameters and the reference value, personnel grading parameters are calculated, and the calculation formula of the personnel grading parameters is as follows:wherein:αclassifying parameters for personnel;Sphysiological characteristic parameters;Sc a reference value for physiological characteristic parameters;Sm an upper limit value of the physiological characteristic parameter;
s204: dividing personnel into primary management and control personnel and secondary management and control personnel according to personnel grading parameters of the personnel;
when the personnel grading parameters of the personnel are more than 1, the personnel are first-level management and control personnel;
when the personnel grading parameter of the personnel is equal to 1, the personnel is a secondary control personnel.
Because the corresponding physiological characteristic parameters also fluctuate when the human body is stressed, the physiological characteristic parameters in the embodiment are blood oxygen saturation or heart rate.
Taking blood oxygen saturation as an example, paO2>The blood oxygen saturation curve of normal people is in a flat section with the blood oxygen saturation of more than 8kPa (60 mmHg) and more than 90 percentAs above, the blood oxygen saturation value is considered to be normal when it is within the range of 90.+ -.10, so that the reference value of the physiological characteristic parameter in this embodimentSc Set to 90, upper limit value of physiological characteristic parameterSm The calculation formula of the personnel grading parameter is 100:when the blood oxygen saturation is between the reference value Sc and the upper limit value Sm, the blood oxygen saturation is considered to be normal, and when the acquired blood oxygen saturation is 95, a living body correction parameter alpha=1 is calculated; when the acquired blood oxygen saturation is 75, the in-vivo correction parameter α=e is calculated.
The personnel grading parameter introduces an allowable physiological characteristic parameter variation range, the personnel grading parameter calculated when the measured physiological characteristic parameter is within the allowable variation range is 1, and the personnel grading parameter calculated when the measured physiological characteristic parameter is not within the allowable variation range is far greater than 1.
Meanwhile, in order to further improve the precision of personnel grading parameters, the physiological characteristic parameters are used for simultaneously acquiring the blood oxygen saturation and the heart rate, the personnel grading parameters are jointly determined by the blood oxygen saturation and the heart rate, the personnel grading parameters are the average value of the personnel grading parameters calculated according to the blood oxygen saturation and the personnel grading parameters calculated according to the heart rate, and specifically, the calculation formula of the personnel grading parameters is as follows:wherein:ɑ1 for the personnel grading parameter calculated from the blood oxygen saturation,Sc1 is a reference value for the blood oxygen saturation,Sm1 Is the upper limit value of the blood oxygen saturation;S1 is the detected blood oxygen saturation;ɑ2 classifying parameters for a person calculated from a heart rate;Sc2 is a reference value for the heart rate,Sm2 is the upper limit value of heart rate;S2 is the heart rate detected.
For heart rate, the heart rate of a normal person is 60-100 times per minute, and the heart rate is tenseThe heart rate is generally about 80-100 times/min, even up to 120 times/min, based on the reference value of the heart rate in this embodimentSc2 Set to 60, upper limit value of physiological characteristic parameterSm2 100.
Step S300, acquiring multi-dimensional characteristic information of a person entering a key area at an entrance of the key area, wherein the multi-dimensional characteristic information is one type of multi-dimensional characteristic information or two types of multi-dimensional characteristic information; acquiring a type of multidimensional characteristic information of a first-level management and control person; one type of multidimensional feature information includes: face feature information, posture feature information, occupation feature information and economic feature information; acquiring second-class multidimensional characteristic information of a secondary management and control person; the second-class multidimensional feature information includes: face feature information, occupation feature information, and economic feature information.
Inputting the face image and the body state image of the primary management and control personnel into a human body feature recognition model to obtain face feature information and body state feature information of the primary management and control personnel;
And inputting the face image of the secondary control personnel into a human body characteristic recognition model to obtain the face characteristic information of the secondary control personnel.
In order to realize accurate identification of face feature information and physical feature information, a camera shooting component is a high-definition camera, and in some specific embodiments, the camera shooting component is a high-resolution monitoring gun camera or ball camera supporting more than 200 ten thousand pixels on the market, so that real-time and clear video stream pictures of import and export personnel in key areas can be provided.
In the follow-up track tracking process, the human body characteristics of the primary management and control personnel are continuously compared with the human body characteristics collected when entering the key area, when the human body characteristics of the personnel are found to be inconsistent with the human body characteristics collected when entering the key area in the follow-up real-time detection process, the human face image camouflage behavior of the personnel can be considered, the personnel can be directly judged as the personnel with risks, the personnel with risks are reported, the judging efficiency of the risks is further improved, and the safety of the key area is ensured.
Specifically, as shown in fig. 4, S400 includes:
s401: collecting an image to be detected in real time in a key area through a camera;
s402: inputting the image to be detected into a human body characteristic recognition model to obtain human face characteristic information and morphological characteristic information of a person in the image to be detected;
S403: matching the face characteristic information or the body state characteristic information obtained in the step S402 with the face characteristic information or the body state characteristic information in the multidimensional characteristic information, and identifying the person in the image to be detected as a primary management and control person or a secondary management and control person;
when the person in the image to be detected is a primary management and control person, face feature information and body state feature information matched with the face feature information or body state feature information acquired in the step S402 are acquired from the multi-dimensional feature information, and the step S404 is entered;
when the person in the image to be detected is a diode control person, entering S405;
s404: human body characteristic matching state analysis of primary management and control personnel;
when the face characteristic information and the body characteristic information acquired from the step S402 are consistent, the step S405 is entered;
when the face characteristic information and the body characteristic information acquired from the multi-dimensional characteristic information and the S402 are inconsistent, judging that the primary management and control personnel have risks, and entering S600;
s405: acquiring a motion trail of a person in an image to be detected;
and inputting the face characteristic information and the body characteristic information of the person in the image to be detected into a target track tracking model, and carrying out tracking analysis on the motion track of the person in the image to be detected to obtain the motion track of the person.
The consistency judgment of the personnel posture features is judged according to the similarity of the face features and the posture features corresponding to the same personnel acquired in S402 and S300, in this embodiment, in S404:
when the similarity of the face characteristic information and the posture characteristic information obtained from the multidimensional characteristic information and the S402 by the first-level management and control personnel is more than or equal to 90%, judging that the face characteristic information and the posture characteristic information are consistent;
and when the similarity of the face characteristic information or the body characteristic information obtained from the multidimensional characteristic information and the S402 by the first-level management and control personnel is less than 90%, judging that the face characteristic information or the body characteristic information is inconsistent.
S500: inputting the motion trail, professional characteristic information and economic characteristic information of the personnel into a multidimensional risk identification model, outputting a judging result of whether the personnel has risks, outputting 1 when the personnel has risks and outputting 0 when the personnel has no risks.
S600: and reporting the identity authentication information corresponding to the person with risk to a superior system by counting the person with risk in real time as a judgment result.
Reporting the identity authentication information corresponding to the person at risk to a superior system, and further calling the information of the related person in the superior system to verify the information of the person.
In a second aspect, the present embodiment provides a system for analyzing personnel behavior in a key area, as shown in fig. 5, including: the system comprises a model training module, a personnel grading module, a multidimensional feature information acquisition module, a motion trail tracking module, a risk judging module and a risk reporting module;
the model training module is used for acquiring a multi-dimensional risk recognition model, a human body characteristic recognition model and a target track tracking model;
acquiring a multidimensional risk identification model based on a laminated neural network, wherein the output of the multidimensional risk identification model is 1 or 0, wherein 1 represents that personnel have risks, and 0 represents that personnel do not have risks;
the human body characteristic recognition model comprises a human face characteristic recognition module and a body characteristic recognition module;
the target track tracking model comprises a track tracking module and a target matching module, wherein the track tracking module is used for estimating the motion track of the personnel, and the target matching module is used for updating the motion track of the personnel so as to realize the continuous tracking of the motion track of the personnel;
the personnel grading module is used for collecting identity authentication information of personnel entering the key area at the entrance of the key area, and dividing the management and control grade of the personnel according to the identity authentication information, wherein the management and control grade is a primary management and control personnel or a secondary management and control personnel;
The identity authentication information includes: physiological characteristic parameters, face images and body state images;
the multi-dimensional characteristic information acquisition module is used for acquiring multi-dimensional characteristic information of personnel entering the key area at the entrance of the key area according to the control level of the personnel entering the key area, wherein the multi-dimensional characteristic information is one type of multi-dimensional characteristic information or two types of multi-dimensional characteristic information;
acquiring a type of multidimensional characteristic information of a first-level management and control person; one type of multidimensional feature information includes: face feature information, posture feature information, occupation feature information and economic feature information;
acquiring second-class multidimensional characteristic information of a secondary management and control person; the second-class multidimensional feature information includes: face feature information, occupation feature information and economic feature information;
inputting the face image and the body state image of the primary management and control personnel into a human body feature recognition model to obtain face feature information and body state feature information of the primary management and control personnel;
inputting the face image of the secondary control personnel into a human body characteristic recognition model to obtain the face characteristic information of the secondary control personnel;
the occupation characteristic information includes occupation of the person at the current time point and occupation within one year before the current time point;
the economic characteristic information comprises income and expense of personnel at the current time point and income and expense of the personnel in one year before the current time point;
The motion trail tracking module is used for carrying out tracking analysis on the human body characteristic matching state and the motion trail of the personnel in the heavy point area through the human body characteristic recognition model and the target trail tracking model to acquire the motion trail of the personnel;
the risk judging module inputs the motion trail, occupational characteristic information and economic characteristic information of the personnel into a multidimensional risk identification model and outputs a judging result of whether the personnel has risks or not;
and the risk reporting module is used for reporting the identity authentication information corresponding to the personnel with risk to the upper system when the statistical judgment result is the personnel with risk.
The key area personnel behavior analysis system in the implementation realizes automatic identification, is low in cost, simple, convenient and obvious in effect, realizes the matching, comparison and identification of the whole area of the human face in the key area, solves the problem of face camouflage, solves the problems of complicated construction process and large calculation amount through the laminated network risk realization model, can realize the rapid early warning of the key area personnel behavior risk with high precision and automation, and has the advantages of reducing the workload of workers and improving the working efficiency.
It should be noted that, in this embodiment, each module (or unit) is in a logic sense, and in a specific implementation, a plurality of modules (or units) may be combined into one module (or unit), and one module (or unit) may be split into a plurality of modules (or units).
The key area personnel behavior analysis system in the implementation realizes automatic identification, is low in cost, simple, convenient and obvious in effect, can accurately and rapidly analyze and report the risk of key area personnel in application scenes of dense crowds, and has the advantages of reducing workload of workers and improving working efficiency.
It will be appreciated by those skilled in the art that all or part of the flow of the method of the above embodiment may be implemented by a program to instruct related hardware, and the program may be stored in a computer readable storage medium, and the program may include the flow of the embodiment of the above methods when executed. The storage medium may be a magnetic disk, an optical disk, a Read-only memory (ROM), a random access memory (Random Access Memory, RAM), or the like.
The foregoing is merely a preferred embodiment of the invention, and although the invention has been described in detail with reference to the foregoing embodiments, it will be apparent to those skilled in the art that modifications may be made to the embodiments described in the foregoing embodiments, or equivalents may be substituted for some of the features thereof. Modifications, equivalents, and alternatives falling within the spirit and principles of the invention are intended to be included within the scope of the invention.