Multimodal biometric fusion security inspection and anti-missed detection system

Abstract

This invention discloses a multimodal biometric fusion-based security screening system to prevent missed checks, relating to the fields of artificial intelligence and civil aviation information technology. It includes a risk perception unit, a data acquisition unit, a real-time monitoring unit, a data analysis unit, and a verification unit. This invention proactively identifies and marks high-risk targets in the isolation area through intelligent behavior analysis, enabling precise triage and enhanced verification of passengers entering the isolation area to ensure key targets are under control. Finally, leveraging multimodal biometric fusion technology, continuous identity verification and dynamic tracking are performed at key nodes such as security checkpoints and jet bridges. This closed-loop process not only achieves a shift from passive prevention to proactive intervention but also significantly improves the real-time detection capability and precise interception efficiency of missed checks through a layered prevention and control mechanism, effectively preventing passengers from entering the isolation area or even boarding the plane without security checks, thus comprehensively enhancing the reliability and overall operational efficiency of the security screening system.

Description

Technical Field

[0001] This invention relates to the fields of artificial intelligence and civil aviation information technology, and in particular to a security inspection and anti-missing system based on multimodal biometric fusion. Background Technology

[0002] Security checks are the first physical line of defense in public safety management and a key link in preventing major security incidents. If dangerous goods, people with false identities, or suspicious targets are missed, it could have irreparable consequences for subsequent steps. Therefore, preventing missed checks is directly related to the effectiveness and integrity of the entire security system.

[0003] In the existing civil aviation security system, there is still a risk that some passengers may enter the security checkpoint without completing the full security check process. Especially during peak passenger flow periods, abnormal behaviors such as passengers tailgating or mixing with others are difficult for manual monitoring to detect in a timely manner, which can easily lead to missed checks. On the other hand, even if passengers complete the security check, they may still go to the wrong jet bridge or enter the wrong boarding gate in the jet bridge area. The existing system lacks dynamic collection, analysis and comparison of multimodal features, making it difficult to identify and intervene in a timely manner, which poses a security risk.

[0004] To address the aforementioned technical shortcomings, a solution is proposed. Summary of the Invention

[0005] The purpose of this invention is to solve the technical problem in the prior art where, during peak passenger flow periods, abnormal behaviors such as tailgating and mixing with other passengers are difficult to detect in a timely manner by manual monitoring, resulting in passengers entering the security checkpoint without going through the complete security check process.

[0006] To achieve the above objectives, the present invention adopts the following technical solution: a security inspection and anti-missed detection system based on multimodal biometric fusion, comprising a data acquisition unit, a risk perception unit, a real-time monitoring unit, a data analysis unit, and a verification unit;

[0007] The data acquisition unit is used to collect and verify the identity and photo information of the people passing through the security checkpoint. If the verification is successful, the status of the person passing through the security checkpoint is updated to "security check completed", security check status information is generated and sent to the risk perception unit. At the same time, the identity and photo information of the people who have been security checked are integrated to generate a security check zone face database and sent to the data analysis unit.

[0008] Furthermore, the identity information and photo information of the passersby are collected and verified at the security checkpoint. The specific process is as follows:

[0009] The ID card reader module built into the designated sensing area of ​​the ID card reader reads the encrypted identity information stored in the ID card of the person passing through. The identity information includes name, gender, ethnicity, date of birth, address, ID card number, issuing authority, validity period, and standard ID photo.

[0010] The high-definition camera built into the human verification machine performs liveness detection on passersby, detecting actions such as nodding, shaking, opening the mouth, and blinking; if the liveness detection is successful, a real-time frontal photo of the passerby's face is captured.

[0011] The face comparison algorithm built into the ID card machine extracts feature points from real-time frontal face photos and compares them with standard ID photos to calculate similarity. The feature points include the position, outline, and spacing of facial features.

[0012] Set a similarity threshold Xs. If the calculated similarity score is greater than or equal to the threshold Xs, the verification is considered successful; otherwise, the verification is considered unsuccessful and a re-verification is initiated.

[0013] The similarity threshold Xs is determined based on historical sample data.

[0014] The risk perception unit receives security check status information and collects real-time video data of passengers through high-precision wide-angle cameras in the security checkpoint and boarding bridge. Based on the security check status information, it updates the status and locates the corresponding passengers in the real-time video data. It also analyzes the behavior of passengers in the restricted area, calculates the missed detection coefficient, and integrates the data to send comprehensive data to the verification unit.

[0015] The isolation area is the restricted area from the security checkpoint to the boarding gate;

[0016] Further, the missed detection coefficient is calculated to generate annotation information, including the following steps:

[0017] S301. Based on real-time video data, extract multimodal feature vectors, including the following steps:

[0018] Based on the OpenPose+CNN algorithm, a multimodal feature extraction model is constructed to extract multimodal feature vector groups. The input of the multimodal feature extraction model is real-time video data, and the output is behavior trajectory vector and action feature vector.

[0019] S302. Based on the behavior trajectory vector and action feature vector, generate behavior labels, including dwell time, behavior switching frequency, and tailing marker;

[0020] The dwell time refers to the time a user stays in place within the security checkpoint when not queuing.

[0021] The behavior switching frequency refers to the number of times a user switches behavior types per unit of time. The behavior types include standing still, walking, looking at a phone, looking at a sign, looking around, and looking back.

[0022] The trailing marker uses a laser camera to calculate the spatial trajectory of each person in the waiting area. Through time lag synchronization analysis, if the trajectory of a person who passes through later is synchronized with the trajectory of any person who passed through earlier, it is recorded as 1; otherwise, it is recorded as 0.

[0023] S303. Based on behavioral labels, construct abnormal behavior vectors and calculate the false negative coefficient;

[0024] The abnormal behavior vector format is: U={stagnation duration, behavior switching frequency, tailing marker};

[0025] The missed detection coefficient The calculation formula is:

[0026] ;

[0027] ;

[0028] ;

[0029] in, For the duration of the stay, M represents the behavior switching frequency, and M represents the trailing flag. Cross-feature factor; These are the weighting coefficients, obtained through training based on historical experimental data.

[0030] Missed detection coefficient The higher the value, the greater the risk of missed detection;

[0031] The real-time monitoring unit is used to collect multimodal biometric information of passers-by through smart devices deployed in security checkpoints and corridor areas, integrate the information to generate a dynamic dataset, and send it to the data analysis unit.

[0032] Furthermore, the intelligent device includes a telephoto face camera array, an inertial measurement unit, and an edge computing gateway; the dynamic dataset includes facial geometry, texture features, and iris texture features; the edge computing gateway performs real-time preprocessing on the dynamic dataset;

[0033] The data analysis unit is used to receive and process the face database and dynamic dataset of the security check zone, extract the multimodal biometrics of the passers-by from the dynamic dataset and compare them with the corresponding features in the face database of the security check zone, generate the comparison results and send them to the verification unit.

[0034] Furthermore, generating the comparison results includes the following steps:

[0035] Multimodal biometric feature extraction, including facial geometry, texture features, and iris texture features, is performed on a dynamic dataset based on the OpenCV algorithm, generating a comparison vector Bd = (Mb, Wl, Hm).

[0036] Simultaneously, the same multimodal biometric feature extraction is performed on the face database of the security check area to generate a comparison object vector Bx = (MbX, WlX, HmX);

[0037] Where Mb and MbX are facial geometric similarity vectors, Wl and WlX are texture feature similarity vectors, and Hm and HmX are iris feature Hamming distance vectors.

[0038] Based on the alignment vector Bd = (Mb, Wl, Hm) and the alignment object vector Bx = (MbX, WlX, HmX), the difference coefficient Cy is calculated. The specific steps are as follows:

[0039] ;

[0040] in, This is the amplification factor for the facial geometric similarity vector. This is the amplification factor for the texture feature similarity vector. This is the magnification factor of the Hamming distance vector of iris features;

[0041] The magnification factor of the facial geometric similarity vector Amplification factor of texture feature similarity vector and the magnification factor of the Hamming distance vector of iris features The amplification factor of each modal feature is set based on its discrimination ability score in historical security inspection comparison data. The amplification factor of each modal feature is determined according to the normalized ratio of the discrimination ability scores of the modal feature and all modal features. If the acquisition quality of a modal feature is lower than the preset quality threshold, the amplification factor of the modal feature is automatically reduced, and the amplification factor of other modal features is increased proportionally.

[0042] Set difference thresholds cy1 and cy2, where cy1 < cy2;

[0043] If Cy > cy2, then the comparison result is consistent with the testimonial evidence;

[0044] If cy1 < Cy ≤ cy2, then the comparison result is an anomaly in the human evidence comparison.

[0045] If Cy≤cy1, then the comparison result is inconsistent with the evidence.

[0046] The verification unit is used to receive comprehensive data and comparison results, confirm whether the person passing through is a missed user based on the comprehensive data and comparison results, and perform a second verification on the person passing through if it is confirmed that there is a risk of missed detection.

[0047] Furthermore, based on the comprehensive data and comparison results, it is confirmed whether the person passing through is a user who has missed the check. If a risk of missed check is confirmed, a second verification is performed on the person passing through, including the following steps:

[0048] S601. Within the isolation zone, based on the false negative coefficient in the comprehensive data. Set a threshold Lj for the missed detection coefficient. When the missed detection coefficient of the passing personnel is... If the value is ≥Lj, the person passing through is identified as a high-risk individual for missed detection, and a second verification and correct route guidance are provided to that person.

[0049] when If the value is less than Lj, then the person passing through is considered a normal user;

[0050] S602. Inside the security checkpoint, if the identity verification results of a person passing through do not match, it is determined that the person has missed the security check and a re-security check process is triggered.

[0051] If the comparison result of the person passing through is abnormal, it is determined that the person passing through is at risk of being missed during security check, and a second verification is performed on them; if the second verification result is that the person passing through matches the person and the document, the person passing through is identified as a normal security check user; otherwise, the person passing through is identified as having missed security check and a re-security check process is triggered.

[0052] When the verification result of the person passing through the passage matches the identity document, the person passing through the passage is directly identified as a normal security check user;

[0053] Inside the covered walkway, if the identity verification results of a person do not match, it is determined that the person has entered the wrong covered walkway, and the correct route is provided.

[0054] If the verification result of the person passing through is abnormal, it is determined that the person passing through is at risk of taking the wrong bridge and a second verification is performed; if the second verification result is that the person passing through matches the identity document, the person passing through is identified as a normal user; otherwise, it is determined that the person passing through has taken the wrong bridge and the correct path guidance is implemented.

[0055] When the verification result of the person passing through matches the identity document, the person passing through is directly identified as a normal user.

[0056] In summary, due to the adoption of the above technical solution, the beneficial effects of the present invention are:

[0057] This multimodal biometric fusion-based security screening system first proactively identifies and marks high-risk targets in the isolation area through intelligent behavior analysis, then strengthens the verification of passengers entering the isolation area to ensure that major risk targets are under control. Finally, using multimodal biometric fusion technology, it continuously verifies identity and dynamically tracks key nodes such as security checkpoints and jet bridges. This closed-loop process not only realizes the transformation from passive prevention to proactive intervention, but also significantly improves the real-time detection capability and precise interception efficiency of missed detection behavior through a layered prevention and control mechanism. At the same time, it effectively avoids flight delays caused by passengers going to the wrong boarding gate, and comprehensively enhances the reliability, intelligence level and overall operational efficiency of the security screening system. Attached Figure Description

[0058] Figure 1 This is a schematic diagram of a security inspection and anti-missed detection system that integrates multimodal biometric features. Detailed Implementation

[0059] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0060] Example: Figure 1 As shown, the multimodal biometric fusion security inspection and anti-missed detection system includes a data acquisition unit, a risk perception unit, a real-time monitoring unit, a data analysis unit, and a verification unit.

[0061] The data acquisition unit is used to collect and verify the identity and photo information of the people passing through the security checkpoint. If the verification is successful, the status of the person passing through the security checkpoint is updated to "security check completed", security check status information is generated and sent to the risk perception unit. At the same time, the identity and photo information of the people who have been security checked are integrated to generate a security check zone face database and sent to the data analysis unit.

[0062] The security checkpoint collects and verifies the identity and photo information of passers-by. The specific process is as follows:

[0063] The ID card reader module built into the designated sensing area of ​​the ID card reader reads the encrypted identity information stored in the ID card of the person passing through. The identity information includes name, gender, ethnicity, date of birth, address, ID card number, issuing authority, validity period, and standard ID photo.

[0064] The high-definition camera built into the human verification machine performs liveness detection on passersby, detecting actions such as nodding, shaking, opening the mouth, and blinking; if the liveness detection is successful, a real-time frontal photo of the passerby's face is captured.

[0065] The face comparison algorithm built into the ID card machine extracts feature points from real-time frontal face photos and compares them with standard ID photos to calculate similarity. The feature points include the position, outline, and spacing of facial features.

[0066] Set a similarity threshold Xs. If the calculated similarity score is greater than or equal to the threshold Xs, the verification is considered successful; otherwise, the verification is considered unsuccessful and a re-verification is initiated.

[0067] The similarity threshold Xs is determined based on historical sample data.

[0068] The risk perception unit receives security check status information and collects real-time video data of passengers through high-precision wide-angle cameras in the security checkpoint and boarding bridge. Based on the security check status information, it updates the status and locates the corresponding passengers in the real-time video data. It also analyzes the behavior of passengers in the restricted area, calculates the missed detection coefficient, and integrates the data to send comprehensive data to the verification unit.

[0069] The restricted area is the area from security checkpoint to boarding gate;

[0070] The calculation of the missed detection coefficient and the generation of annotation information include the following steps:

[0071] S301. Based on real-time video data, extract multimodal feature vectors, including the following steps:

[0072] Based on the OpenPose+CNN algorithm, a multimodal feature extraction model is constructed to extract multimodal feature vector groups. The input of the multimodal feature extraction model is real-time video data, and the output is behavior trajectory vector and action feature vector.

[0073] S302. Generate behavior labels based on behavior trajectory vectors and action feature vectors, including dwell time, behavior switching frequency, and tailing markers;

[0074] Dwell time indicates the time a user stays in place within the security checkpoint when not queuing.

[0075] Behavior switching frequency refers to the number of times a user switches behavior types within a unit of time. Behavior types include stationary, walking, looking at a phone, looking at a sign, looking around, and looking back.

[0076] The trailing marker uses a laser camera to calculate the spatial trajectory of each person in the waiting area. Through time lag synchronization analysis, if the trajectory of a person who passes through later is synchronized with the trajectory of any person who passed through earlier, it is recorded as 1; otherwise, it is recorded as 0.

[0077] S303. Based on behavioral labels, construct abnormal behavior vectors and calculate the false negative coefficient;

[0078] The abnormal behavior vector format is: U={dwelling duration, behavior switching frequency, trailing marker};

[0079] Missed detection coefficient The calculation formula is:

[0080] ;

[0081] ;

[0082] ;

[0083] in, For the duration of the stay, M represents the behavior switching frequency, and M represents the trailing flag. Cross-feature factor; These are the weighting coefficients, obtained through training based on historical experimental data.

[0084] Missed detection coefficient The higher the value, the greater the risk of missed detection;

[0085] The real-time monitoring unit is used to collect multimodal biometric information of passers-by through smart devices deployed in security checkpoints and corridor areas, integrate the information to generate a dynamic dataset, and send it to the data analysis unit.

[0086] The intelligent device includes a telephoto face camera array, an inertial measurement unit, and an edge computing gateway; the dynamic dataset includes facial geometry, texture features, and iris texture features; the edge computing gateway performs real-time preprocessing on the dynamic dataset;

[0087] The data analysis unit is used to receive and process the face database and dynamic dataset of the security check zone, extract the multimodal biometrics of the passers-by from the dynamic dataset and compare them with the corresponding features in the face database of the security check zone, generate the comparison results and send them to the verification unit.

[0088] Generating the alignment results includes the following steps:

[0089] Multimodal biometric feature extraction, including facial geometry, texture features, and iris texture features, is performed on a dynamic dataset based on the OpenCV algorithm, generating a comparison vector Bd = (Mb, Wl, Hm).

[0090] Simultaneously, the same multimodal biometric feature extraction is performed on the face database of the security check area to generate a comparison object vector Bx = (MbX, WlX, HmX);

[0091] Where Mb and MbX are facial geometric similarity vectors, Wl and WlX are texture feature similarity vectors, and Hm and HmX are iris feature Hamming distance vectors.

[0092] Based on the alignment vector Bd = (Mb, Wl, Hm) and the alignment object vector Bx = (MbX, WlX, HmX), the difference coefficient Cy is calculated. The specific steps are as follows:

[0093] ;

[0094] in, This is the amplification factor for the facial geometric similarity vector. This is the amplification factor for the texture feature similarity vector. This is the magnification factor of the Hamming distance vector of iris features;

[0095] Amplification factor of facial geometric similarity vector Amplification factor of texture feature similarity vector and the magnification factor of the Hamming distance vector of iris features The amplification factor of each modal feature is set based on its discrimination ability score in historical security inspection comparison data. The amplification factor of each modal feature is determined according to the normalized ratio of the discrimination ability scores of the modal feature and all modal features. If the acquisition quality of a modal feature is lower than the preset quality threshold, the amplification factor of the modal feature is automatically reduced, and the amplification factor of other modal features is increased proportionally.

[0096] Set difference thresholds cy1 and cy2, where cy1 < cy2;

[0097] If Cy > cy2, then the comparison result is consistent with the testimonial evidence;

[0098] If cy1 < Cy ≤ cy2, then the comparison result is an anomaly in the human evidence comparison.

[0099] If Cy≤cy1, then the comparison result is inconsistent with the evidence.

[0100] The verification unit is used to receive comprehensive data and comparison results, confirm whether the person passing through is a missed user based on the comprehensive data and comparison results, and perform a second verification on the person passing through if it is confirmed that there is a risk of missed detection.

[0101] Based on comprehensive data and comparison results, it is determined whether the person passing through is a user who missed the check. If a risk of missed check is confirmed, a second verification is performed on the person passing through, including the following steps:

[0102] S601. Within the isolation zone, based on the false negative coefficient in the comprehensive data. Set a threshold Lj for the missed detection coefficient. When the missed detection coefficient of the passing personnel is... If the value is ≥Lj, the person passing through is identified as a high-risk individual for missed detection, and a second verification and correct route guidance are provided to that person.

[0103] when If the value is less than Lj, then the person passing through is considered a normal user;

[0104] S602. Inside the security checkpoint, if the identity verification results of a person passing through do not match, it is determined that the person has missed the security check and a re-security check process is triggered.

[0105] If the comparison result of the person passing through is abnormal, it is determined that the person passing through is at risk of being missed during security check, and a second verification is performed on them; if the second verification result is that the person passing through matches the person and the document, the person passing through is identified as a normal security check user; otherwise, the person passing through is identified as having missed security check and a re-security check process is triggered.

[0106] When the verification result of the person passing through the passage matches the identity document, the person passing through the passage is directly identified as a normal security check user;

[0107] Inside the covered walkway, if the identity verification results of a person do not match, it is determined that the person has entered the wrong covered walkway, and the correct route is provided.

[0108] If the verification result of the person passing through is abnormal, it is determined that the person passing through is at risk of taking the wrong bridge and a second verification is performed; if the second verification result is that the person passing through matches the identity document, the person passing through is identified as a normal user; otherwise, it is determined that the person passing through has taken the wrong bridge and the correct path guidance is implemented.

[0109] When the verification result of the person passing through matches the identity document, the person passing through is directly identified as a normal user.

[0110] By using intelligent behavior analysis, high-risk targets are proactively identified and marked in the isolation area, enabling precise triage and enhanced verification of passengers entering the isolation area to ensure that key targets are under control. Finally, multimodal biometric fusion technology is used to continuously verify identity and dynamically track key nodes such as security checkpoints and jet bridges. This closed-loop process not only realizes the transformation from passive prevention to proactive intervention, but also significantly improves the real-time detection capability and precise interception efficiency of missed detection behavior through a layered prevention and control mechanism. At the same time, it effectively avoids flight delays caused by passengers going to the wrong boarding gate, and comprehensively enhances the reliability, intelligence level and overall operational efficiency of the security system.

[0111] The size of the interval and threshold is set to facilitate comparison. The size of the threshold depends on the amount of sample data and the number of bases set by those skilled in the art for each set of sample data; as long as it does not affect the ratio between the parameter and the quantized value.

[0112] The above formulas are all dimensionless calculations. The formulas are derived from software simulations based on a large amount of collected data to obtain the most recent real-world results. The preset parameters in the formulas are set by those skilled in the art according to the actual situation.

[0113] The above description is only a preferred embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any equivalent substitutions or modifications made by those skilled in the art within the scope of the technology disclosed in the present invention, based on the technical solution and inventive concept of the present invention, should be covered within the scope of protection of the present invention.

Claims

1. A security inspection and missed detection prevention system based on multimodal biometric fusion, characterized in that, It includes a data acquisition unit, a risk perception unit, a real-time monitoring unit, a data analysis unit, and a verification unit; The data acquisition unit is used to collect and verify the identity and photo information of the people passing through the security checkpoint. If the verification is successful, the status of the person passing through the security checkpoint is updated to "security check completed", security check status information is generated and sent to the risk perception unit. At the same time, the identity and photo information of the people who have been security checked are integrated to generate a security check zone face database and sent to the data analysis unit. The risk perception unit receives security check status information and collects real-time video data of passengers through high-precision wide-angle cameras in the security checkpoint and boarding bridge. Based on the security check status information, it updates the status and locates the corresponding passengers in the real-time video data. It also analyzes the behavior of passengers in the restricted area, calculates the missed detection coefficient, and integrates the data to send comprehensive data to the verification unit. The isolation area is the restricted area from the security checkpoint to the boarding gate; The real-time monitoring unit is used to collect multimodal biometric information of passers-by through smart devices deployed in security checkpoints and corridor areas, integrate the information to generate a dynamic dataset, and send it to the data analysis unit. The data analysis unit is used to receive and process the face database and dynamic dataset of the security check zone, extract the multimodal biometrics of the passers-by from the dynamic dataset and compare them with the corresponding features in the face database of the security check zone, generate the comparison results and send them to the verification unit. The verification unit is used to receive comprehensive data and comparison results, confirm whether the person passing through is a missed user based on the comprehensive data and comparison results, and perform a second verification on the person passing through if a missed detection risk is confirmed.

2. The multimodal biometric fusion security inspection and anti-missed detection system according to claim 1, characterized in that, The security checkpoint collects and verifies the identity and photo information of passers-by. The specific process is as follows: The ID card reader module built into the designated sensing area of ​​the ID card reader reads the encrypted identity information stored in the ID card of the person passing through. The identity information includes name, gender, ethnicity, date of birth, address, ID card number, issuing authority, validity period, and standard ID photo. The high-definition camera built into the human verification machine performs liveness detection on passersby, detecting actions such as nodding, shaking, opening the mouth, and blinking; if the liveness detection is successful, a real-time frontal photo of the passerby's face is captured. The face comparison algorithm built into the ID card machine extracts feature points from real-time frontal face photos and compares them with standard ID photos to calculate similarity. The feature points include the position, outline, and spacing of facial features. Set a similarity threshold Xs. If the calculated similarity score is greater than or equal to the threshold Xs, the verification is considered successful; otherwise, the verification is considered unsuccessful and a re-verification is initiated. The similarity threshold Xs is determined based on historical sample data.

3. The multimodal biometric fusion security inspection and anti-missed detection system according to claim 1, characterized in that, The calculation of the missed detection coefficient and the generation of annotation information include the following steps: S301. Based on real-time video data, extract multimodal feature vectors, including the following steps: Based on the OpenPose+CNN algorithm, a multimodal feature extraction model is constructed to extract multimodal feature vector groups. The input of the multimodal feature extraction model is real-time video data, and the output is behavior trajectory vector and action feature vector. S302. Based on the behavior trajectory vector and action feature vector, generate behavior labels, including dwell time, behavior switching frequency, and tailing marker; The dwell time refers to the time a user stays in place within the security checkpoint when not queuing. The behavior switching frequency refers to the number of times a user switches behavior types per unit of time. The behavior types include standing still, walking, looking at a phone, looking at a sign, looking around, and looking back. The trailing marker uses a laser camera to calculate the spatial trajectory of each person in the waiting area. Through time lag synchronization analysis, if the trajectory of a person who passes through later is synchronized with the trajectory of any person who passed through earlier, it is recorded as 1; otherwise, it is recorded as 0. S303. Based on behavioral labels, construct abnormal behavior vectors and calculate the missed detection coefficient; The abnormal behavior vector format is: U={stagnation duration, behavior switching frequency, tailing marker}; The missed detection coefficient The calculation formula is: ； ； ； in, For the duration of the stay, M represents the behavior switching frequency, and M represents the trailing flag. Cross-feature factor; These are the weighting coefficients, obtained through training based on historical experimental data. Missed detection coefficient The higher the value, the greater the risk of missed detection.

4. The multimodal biometric fusion security inspection and anti-missed detection system according to claim 1, characterized in that, The intelligent device includes a telephoto face camera array, an inertial measurement unit, and an edge computing gateway; the dynamic dataset includes facial geometry, texture features, and iris texture features. The edge computing gateway performs real-time preprocessing of dynamic datasets.

5. The multimodal biometric fusion security inspection and anti-missed detection system according to claim 3, characterized in that, Generating the alignment results includes the following steps: Multimodal biometric feature extraction, including facial geometry, texture features, and iris texture features, is performed on a dynamic dataset based on the OpenCV algorithm, generating a comparison vector Bd = (Mb, Wl, Hm). Simultaneously, the same multimodal biometric feature extraction is performed on the face database of the security check area to generate a comparison object vector Bx = (MbX, WlX, HmX); Where Mb and MbX are facial geometric similarity vectors, Wl and WlX are texture feature similarity vectors, and Hm and HmX are iris feature Hamming distance vectors. Based on the alignment vector Bd = (Mb, Wl, Hm) and the alignment object vector Bx = (MbX, WlX, HmX), the difference coefficient Cy is calculated. The specific steps are as follows: ; in, This is the amplification factor for the facial geometric similarity vector. This is the amplification factor for the texture feature similarity vector. This is the magnification factor for the Hamming distance vector of iris features; The magnification factor of the facial geometric similarity vector Amplification factor of texture feature similarity vector and the magnification factor of the Hamming distance vector of iris features The amplification factor of each modal feature is set based on its discrimination ability score in historical security inspection comparison data. The amplification factor of each modal feature is determined according to the normalized ratio of the discrimination ability scores of the modal feature and all modal features. If the acquisition quality of a modal feature is lower than the preset quality threshold, the amplification factor of the modal feature is automatically reduced, and the amplification factor of other modal features is increased proportionally. Set difference thresholds cy1 and cy2, where cy1 < cy2; If Cy > cy2, then the comparison result is consistent with the testimonial evidence; If cy1 < Cy ≤ cy2, then the comparison result is an anomaly in the human evidence comparison. If Cy≤cy1, then the comparison result is inconsistent with the evidence.

6. The multimodal biometric fusion security inspection and anti-missed detection system according to claim 5, characterized in that, Based on comprehensive data and comparison results, it is determined whether the person passing through is a user who missed the check. If a risk of missed check is confirmed, a second verification is performed on the person passing through, including the following steps: S601. Within the isolation zone, based on the false negative coefficient in the comprehensive data. Set a threshold Lj for the missed detection coefficient. When the missed detection coefficient of the passing personnel is... If the value is ≥Lj, the person passing through is identified as a high-risk individual for missed detection, and a second verification and correct route guidance are provided to the person passing through. when If the value is less than Lj, then the person passing through is considered a normal user; S602. Inside the security checkpoint, if the identity verification results of a person passing through do not match, it is determined that the person has missed the security check and a re-security check process is triggered. If the comparison result of the person passing through is abnormal, it is determined that the person passing through is at risk of being missed during security check, and a second verification is performed on them; if the second verification result is that the person passing through matches the person and the document, the person passing through is identified as a normal security check user; otherwise, the person passing through is identified as having missed security check and a re-security check process is triggered. When the verification result of the person passing through the passage matches the identity document, the person passing through the passage is directly identified as a normal security check user; Inside the covered walkway, if the identity verification results of a person do not match, it is determined that the person has entered the wrong covered walkway, and the correct route is provided. If the verification result of the person passing through is abnormal, it is determined that the person passing through is at risk of taking the wrong bridge and a second verification is performed; if the second verification result is that the person passing through matches the identity document, the person passing through is identified as a normal user; otherwise, it is determined that the person passing through has taken the wrong bridge and the correct path guidance is implemented. When the verification result of the person passing through matches the identity document, the person passing through is directly identified as a normal user.

Images