Human face comparison-based personnel flow track automatic determination system and method

Abstract

The invention relates to the technical field of computer vision, in particular to a human face comparison-based personnel flow trajectory automatic judgment system and method. When a special person passes through a certain camera point location, an embodiment is judged to belong to according to key persons, and activity track information of the person on that day is obtained through a person information retrieval module. And according to the timestamp of the key personnel passing through a certain camera point location and the corresponding definition of special and space-time accompanying personnel, extracting all personnel information and movement track information in the time period under the corresponding camera point location. The feature data of the face organ region is represented by using the region feature data, and is compared with the photo information, so that the accuracy and efficiency of comparison are improved. Compared with the traditional global feature comparison, the regional feature comparison can better process the problems of shielding, angle change and the like.

Description

Technical Field

[0001] The invention relates to the field of computer vision technology, specifically a system and method for automatically determining personnel flow trajectories based on face comparison. Specifically, it involves multi-angle face comparison, facial feature recognition, facial region comparison, and individual trajectory fusion recognition technology. Background Art

[0002] Referring to the existing methods of tracking and determining the flow trajectories of people, the methods for determining the flow trajectories of special groups are relatively simple and discontinuous in time and space. There are situations where relevant personnel trajectory information is left over, the time points are inaccurate, and the credibility is low. At this time, a lot of manpower and material resources are required to investigate, which adds a lot of work to the grassroots staff and increases the technical complexity and resource volume. At the same time, according to the existing face recognition technology and video tracking methods, face recognition and comparison are basically achieved by pre-entering the training model and undergoing multiple training to achieve the purpose of accurate face recognition. This method lacks the timeliness of face comparison. As time goes by, the dynamic changes in personnel information cannot be updated immediately, resulting in low accuracy and reliability of face comparison. The method for automatically determining the flow trajectory of personnel based on face comparison provided by the present invention uses face recognition and face comparison technology, adopts real-time recognition and comparison of face feature vectors, integrates personal daily trajectory data, combines shooting time and monitoring location information, takes the first trajectory point as the starting point and the last trajectory point as the end point, and forms personal activity trajectories in chronological order. No simulation training is required, which provides precise positioning and tracking to help improve the accuracy and efficiency of personnel tracking and reasonably control the trajectories of special groups. Summary of the Invention

[0003] In order to solve the problems described in the above background technology, the present invention provides a method for automatically determining the flow trajectory of people based on face comparison. In the current situation where the flow of special groups of people cannot be effectively monitored, the travel trajectory of special target people and the people accompanying them in time and space can be quickly analyzed.

[0004] In order to achieve the above object, the technical solution adopted by the present invention is as follows: A system for automatically determining the flow trajectory of people based on face comparison, comprising a video image acquisition module, a face identification terminal, a photo information calling module, a photo decoding module, a face capture module, a face recognition module, a database storage module, a face comparison module, and a personnel information retrieval module; The video image acquisition module in the face recognition end detects pedestrian information in real time from the captured video footage, including image and video information. The face recognition module extracts facial image information from the collected image and video information based on facial feature values. The personal information trajectory is stored by saving the successfully recognized facial information and the personal trajectory information retrieved by the database storage module, forming a personal trajectory information result set. The face comparison module compares the image extracted by the face recognition end based on the human feature values ​​with the retrieved photo information. After a successful comparison, the personal ID number is extracted based on the photo information and the identity information of the public security system is retrieved to obtain the personal identity information of the matched person. The personal information trajectory is stored in the database storage module through the slime mold foraging method. The daily movement trajectory of the same person is saved under the personal identity information to track the target information. According to the definitions of close contacts, secondary close contacts, and constant companions, the trajectory information of the target within the previous and next dates is found and data fused. Finally, the trajectory is analyzed by combining the shooting time and monitoring location information to derive daily images and trajectories containing the target and the contact target. The target information and contact target information are further confirmed through feedback search.

[0005] A method for automatically determining personnel flow trajectories based on face comparison includes the following steps: Step 1: Multi-angle face acquisition. The video image acquisition module preprocesses the image information acquired by the real-time detection camera and performs frame segmentation on the dynamic video image sequence to obtain static image information. Background subtraction is then used to detect moving targets, obtaining images containing facial information. Facial feature vectors are extracted from the multi-angle face images. The acquired facial feature vector data and the facial feature vector data obtained by acquiring photo information are recorded in a vector matrix. The images containing facial information are stored in a database for use in step 2.

[0006] Step 2: Facial feature recognition algorithm. Based on the facial information image stored in step 1 and the Euclidean distance of the facial feature vector, the facial feature data scaling weight value is obtained. The facial feature similarity comparison result is obtained based on the facial regression learning function through the influence of the posture of the person when walking, such as looking up, looking down, hair occlusion, and side face. The comparison result is used in step 3. Step 3: Face matching algorithm. Based on the basic information data of the facial regional features and the facial feature similarity comparison results obtained in Step 3, the regional feature data is used to represent the feature data of each organ region corresponding to the face. This is compared with the obtained person's photo information to obtain the final facial similarity result. The similarity result is applied to Step 4.

[0007] Step 4: Personal trajectory data fusion. The comparison results from Step 3 are integrated with daily personal trajectory information through the slime mold foraging method to track the target information. According to the definition of close contacts and temporal and spatial companions, the trajectory information of the target in the days before and after is found and fused. Finally, the trajectory is analyzed by combining the shooting time and monitoring location information, and the image and trajectory containing the target and the contact target are derived. The image and trajectory are sent back to the personal information storage terminal for storage, and the target information and contact target information are further confirmed through feedback search.

[0008] Step 5: Personnel information retrieval. Based on the confirmed target information and contact target information in step 4, the basic information of permanent residents in the public security system is retrieved and associated with the identity information of the person's photo extracted by the face comparison module. The matching basic information of the person is screened and counted to match the person's name, ID card, home address information, and daily activity trajectory; Description of the attached drawings Figure 1 This is a flow chart of the face recognition and comparison method of the present invention; Figure 2 This is a diagram of the architecture of the face recognition and comparison method of the present invention; Figure 3 This is a diagram illustrating the architecture of the facial feature vector recognition and comparison method of the present invention; Figure 4 This is the face recognition comparison result diagram of the present invention. DETAILED DESCRIPTION

[0009] A system for automatically determining the flow trajectory of people based on face comparison, comprising a video image acquisition module, a face identification terminal, a photo information calling module, a photo decoding module, a face capture module, a face recognition module, a database storage module, a face comparison module, and a personnel information retrieval module; The video image acquisition module in the face recognition end detects pedestrian information in real time from the captured video footage, including image and video information. The face recognition module extracts facial image information from the collected image and video information based on facial feature values. The personal information trajectory is stored by saving the successfully recognized facial information and the personal trajectory information retrieved by the database storage module, forming a personal trajectory information result set. The face comparison module compares the image extracted by the face recognition end based on the human feature values ​​with the retrieved photo information. After a successful comparison, the personal ID number is extracted based on the photo information and the identity information of the public security system is retrieved to obtain the personal identity information of the matched person. The personal information trajectory is stored in the database storage module through the slime mold foraging method. The daily movement trajectory of the same person is saved under the personal identity information to track the target information. According to the definitions of close contacts, secondary close contacts, and constant companions, the trajectory information of the target within the previous and next dates is found and data fused. Finally, the trajectory is analyzed by combining the shooting time and monitoring location information to derive daily images and trajectories containing the target and the contact target. The target information and contact target information are further confirmed through feedback search.

[0010] A method for automatically determining personnel flow trajectories based on face comparison includes the following steps: Step 1: Multi-angle face acquisition. The video image acquisition module preprocesses the image information acquired by the real-time detection camera and performs frame segmentation on the dynamic video image sequence to obtain static image information. Background subtraction is then used to detect moving targets, obtaining images containing facial information. Facial feature vectors are extracted from the multi-angle face images. The acquired facial feature vector data and the facial feature vector data obtained by acquiring photo information are recorded in a vector matrix. The images containing facial information are stored in a database for use in step 2.

[0011] Step 2: Facial feature recognition algorithm. Based on the facial information image stored in step 1 and the Euclidean distance of the facial feature vector, the facial feature data scaling weight value is obtained. The facial feature similarity comparison result is obtained based on the facial regression learning function through the influence of the posture of the person when walking, such as looking up, looking down, hair occlusion, and side face. The comparison result is used in step 3. Step 3: Face matching algorithm. Based on the basic information data of the facial regional features and the facial feature similarity comparison results obtained in Step 3, the regional feature data is used to represent the feature data of each organ region corresponding to the face. This is compared with the obtained person's photo information to obtain the final facial similarity result. The similarity result is applied to Step 4.

[0012] Step 4: Personal trajectory data fusion. The comparison results from Step 3 are integrated with daily personal trajectory information through the slime mold foraging method to track the target information. According to the definition of close contacts and temporal and spatial companions, the trajectory information of the target in the days before and after is found and fused. Finally, the trajectory is analyzed by combining the shooting time and monitoring location information, and the image and trajectory containing the target and the contact target are derived. The image and trajectory are sent back to the personal information storage terminal for storage, and the target information and contact target information are further confirmed through feedback search.

[0013] Step 5: Personnel information retrieval. Based on the confirmed target information and contact target information in step 4, the basic information of permanent residents in the public security system is retrieved and associated with the identity information of the person's photo extracted by the face comparison module. The matching basic information of the person is screened and counted to match the person's name, ID card, home address information, and daily activity trajectory; Furthermore, in the above step 1, multi-angle face acquisition includes the following: The face collection end device and the face recognition end device are connected via a dedicated line link; 1) The face collection terminal device includes: The video image acquisition module is used to detect pedestrian information in the video screen captured by the camera in real time, perform frame cutting on the dynamic video image sequence to obtain static image information, and then use the background subtraction method to detect moving targets to obtain images containing facial information. The image information is pre-processed, and the shooting time and monitoring location information are recorded and saved to the database.

[0014] The database storage module is used to store the facial images identified by the face recognition module and save the facial information and daily trajectory information for the fourteen days before and after as agreed.

[0015] 2) The face recognition terminal device includes: The face recognition module is used to identify the image information and video information collected by the video acquisition end module, and extract the facial feature vectors in the multi-angle face images according to the facial feature values ​​of the human body based on the pre-processed image information.

[0016] 3) The facial feature vector data collected and the facial feature vector data obtained by obtaining photo information are recorded using a vector matrix of w=n*m, as shown below: , , ; in, ={x1, x2, x3...xn} represents the first row of the vector matrix, xi is the facial feature point vector of the i-th collected image, ={y1, y2, y3...yn} represents the first row of the obtained photo vector matrix, and yj is the facial feature point vector of the jth photo information.

[0017] Furthermore, in the above step 2, the facial feature recognition algorithm includes the following: The photo information calling module is used to call the photo information of people in the public security system, and decode the binary code of the photo called out from the public security system according to the prescribed algorithm, identify the photo information of the ID card and compare it with the face recognized in the face recognition module.

[0018] The distance formula of the facial feature vector of the facial feature data collected by the face recognition module is: (1) 、 are the horizontal and vertical coordinates of the i-th facial feature pixel on the face image; The distance formula of the facial feature vector of the face feature data obtained by obtaining the photo information is: (2) Taking into account the size of the faces captured due to the distance between the camera and the pedestrians during video capture, the scaling weighted value of the facial feature data captured by the face acquisition module is calculated using formulas (1) and (2): (3) Because the facial images collected by the face acquisition module are affected by the posture of the person when walking, such as looking up, looking down, hair occlusion, side face, etc., the face posture coefficient is introduced. Its calculation formula is as follows: (4) in is a scaling factor used to avoid errors caused by inconsistencies between the scale of the face image captured by the face acquisition module and the standard face posture image obtained by acquiring photo information.

[0019] Considering that the face acquisition module will collect multiple different facial information of the same person within the video frame time, in order to ensure the balance of the sample, the facial feature data obtained by obtaining photo information will be given a higher weight. Moreover, a person's photo information is stored in the ID card database. Therefore, the weighted function expression is obtained as follows: (5) Where N is the face acquisition module that will collect multiple different faces of the same person in multiple video frames. Based on the face regression learning function, the formula for facial feature similarity comparison is obtained: (6) in is the facial feature vector obtained by obtaining photo information, The facial feature vector data of the face is collected through the face recognition module.

[0020] Furthermore, in step 3 above, the face comparison algorithm includes the following: According to the basic information data of the facial region features, the data of the regional feature a (1≤a≤7) is used to represent the feature data of each organ region corresponding to the face, the data of the regional feature H1 represents the feature data of the forehead, the data of the regional feature H2 represents the feature data of the left eye, the data of the regional feature H3 represents the feature data of the right eye, the data of the regional feature H4 represents the feature data of the nose, the data of the regional feature H5 represents the feature data of the mouth, the data of the regional feature H6 represents the feature data of the left cheek, and the data of the regional feature H7 represents the feature data of the right cheek; Furthermore, when the average value of the facial features obtained by the face acquisition module is greater than the preset threshold value of 0.7, it can be considered to be the same person. The average value vector is calculated as follows: (7) The calculation formula of the average value vector obtained by obtaining photo information is as follows: (8) According to the face comparison algorithm, the final face similarity formula is as follows: (9) When the difference between the average vector calculated by the face acquisition module and the average vector obtained by obtaining photo information is greater than 0.7, that is, when the similarity is greater than 70%, it can be considered that they are the same person.

[0021] Furthermore, in step 4 above, the fusion of personal trajectory data includes the following: The slime mold is initialized based on the first personal photo with a successful face match. The facial feature vector of the first personal photo is obtained, and the shooting time and monitoring location are recorded. The slime mold's expansion and contraction process extracts the facial feature vector of the person successfully matched that day and compares it with the first facial feature vector. Referring to the three-person face matching algorithm in step 2, the slime mold continues to contract until all facial feature vectors obtained that day are matched. The slime mold contraction process ends, and the trajectory matching and recognition process is complete. The successfully recognized face information and trajectory information are returned to the personal information trajectory storage terminal.

[0022] Assume that the facial feature vectors of M images extracted based on the slime mold model are (10) in is the facial feature point vector of the i-th collected image, is the face pose coefficient, is the deviation vector of the feature pixel of the i-th face when the slime mold forages.

[0023] According to the learning characteristics of slime mold, the face regression learning function is used to obtain the final objective function of face image feature points and trajectory information. (11) in Collect the facial feature point vector for the first slime mold foraging image. is the trajectory offset between the i-th facial feature point vector and the first facial feature point vector.

[0024] The personal information trajectory storage terminal uses the slime mold foraging method to save the daily movement trajectory of the same person under the personal identity information, which is used to track the target information. According to the definition of close contacts and temporal and spatial companions, the trajectory information of the target in the previous and subsequent dates is found for data fusion. Finally, the trajectory is analyzed by combining the shooting time and monitoring location information, thereby deriving daily images and trajectories containing the target and the contact target, and further confirming the target information and the contact target information through feedback search.

[0025] The following is a clear and complete description of the technical solutions in the embodiments of this application, in conjunction with the accompanying drawings. A method for automatically determining close proximity and spatiotemporal accompaniment of people based on face comparison includes a video image acquisition module, a database storage module, a face recognition module, a photo information retrieval module, and a person information retrieval module.

[0026] In this embodiment, the video image acquisition module is used to detect pedestrian information in real time in the video images captured by the camera, capture facial image information and video information, and return pedestrian detection information in real time. It records the capture location and time information, combines historical trajectory data to mark high-traffic areas (high-nutrient areas), and dynamically improves the sampling rate of the area through the slime mold weight feedback mechanism.

[0027] Furthermore, a face capture module is used to extract clear face images from the video information recorded by the video acquisition module, assign initial weights to the detected faces (based on image clarity, frontal orientation, etc.) in combination with slime mold bionic logic, dynamically update the similarity threshold between the faces and the key personnel database, sort by weight, and preferentially transmit high-weighted faces to the recognition module; Furthermore, the face recognition module is used to identify the image information and video information collected by the video acquisition end module, extract the human face image information according to the human face feature value, calculate the variance of each latitude, retain the first N dimensions with the largest variance (high nutrient channel), sparsely store the remaining dimensions, and save the face feature vector data, and record the shooting time and monitoring location information, and output the compressed feature vector to the comparison module to reduce storage pressure; Furthermore, the photo information calling module is used to call the photo information of people in the public security system, decode the binary code of the photo called out from the public security system according to the prescribed algorithm, identify the photo information of the ID card and compare it with the face recognized by the face recognition module, establish the retrieval network topology of the key personnel database (with the person ID and the last appearance position as the node), and then perform slime mold path planning. For frequently queried personnel (high-nutrient nodes), the retrieval path is shortened (such as caching to memory) and the index priority is reduced for personnel who have not been queried for a long time. Furthermore, the face comparison module is used to compare the image extracted by the face recognition end according to the human feature value with the face photo information within the current comparison time range according to the timestamp. Based on the slime mold expansion exploration and dynamic threshold oscillation method, if the comparison similarity is greater than the set threshold, the comparison is successful and the identity information of the current person in the photo information is extracted at the same time; Furthermore, the personnel information retrieval module will retrieve the basic information of permanent residents and discrete trajectory information from the public security system, and associate it with the identity information of the personnel photos extracted by the face comparison module. Using the trajectory segment fusion method, the matching basic information of the personnel will be screened and counted to match the personnel's name, ID card, home address information and daily activity trajectory. Furthermore, the personnel information retrieved in the personnel information retrieval module is associated with the nucleic acid information and signaling data to obtain the complete three core elements of personnel data, including name, ID number, and telephone number, and generate a personnel identity information report by location.

[0028] Example 1: Identification of key personnel In this embodiment, the video capture end camera captures the image of the human face, captures the face of pedestrians within the camera coverage area, and captures the facial image information and video information, including the following steps: S-01. Face recognition: Use the video capture end camera to capture facial images. Capture facial images within the camera's coverage area, capturing both image and video information. Photos that do not capture the full facial features are discarded. S-02. Feature extraction: The facial recognition module locates the nose, eyes, and mouth of the captured facial image to determine the facial area. The corresponding facial feature values ​​are obtained using traditional facial recognition methods. The data is stored and the shooting time and monitoring location information are recorded to determine the comparison time range. S-03 photo information call, through the photo information call module, call the public security system personnel photo information, the photo information obtained in binary format base64 decoded, identify the face photo information, and obtain facial feature values ​​according to the face information; S-04. Face comparison: compare the facial image feature values ​​with the feature values ​​obtained from the face information according to the facial features one by one, calculate the similarity between the face identified by the face information and the faces in the face feature database, that is, find the face in the photo information that is most similar to the face to be identified based on the calculated similarity.

[0029] Based on the set comparison time, the corresponding data in the facial feature database is compared with the facial feature values ​​obtained from the face information, and the similarity between the face in the face information and each face in the facial feature database is calculated. Based on the timestamp of the facial feature value storage, facial photos within the required comparison time range are selected and compared with the face recognized by the face recognition module. If the same photo information contains multiple photos, the above steps are repeated until all photos are compared. In this embodiment, the corresponding facial feature data is extracted from the facial template of the face to be identified based on the set comparison time range, and this data is compared one by one with the facial feature data obtained from the face information to calculate the similarity between the face to be identified and the face obtained from the face information.

[0030] According to the face regression learning function, the formula for facial feature similarity comparison is: in is the facial feature vector obtained by obtaining photo information, The facial feature vector data of the face is collected through the face recognition module.

[0031] According to the basic information data of the facial regional features, the data of regional feature a (1≤a≤7) is used to represent the feature data of each organ region corresponding to the face. The data of regional feature H1 represents the feature data of the forehead, the data of regional feature H2 represents the feature data of the left eye, the data of regional feature H3 represents the feature data of the right eye, the data of regional feature H4 represents the feature data of the nose, the data of regional feature H5 represents the feature data of the mouth, the data of regional feature H6 represents the feature data of the left cheek, and the data of regional feature H7 represents the feature data of the right cheek. The facial characteristics obtained by the face acquisition module are calculated as follows: The calculation formula of the average value vector obtained by obtaining photo information is as follows: According to the face comparison algorithm, the final face similarity formula is as follows: That is, when the difference between the average value vector calculated by the face acquisition module and the average value vector obtained by obtaining photo information is greater than 0.7, that is, when the similarity is greater than 70%, they can be considered to be the same person.

[0032] Mark the matched photo information.

[0033] S-05. Acquisition of identity information: After the facial image feature value is successfully compared with the feature value photo information obtained according to the facial features, the person's ID number information is extracted according to the sequence of the photos.

[0034] S-06. Report generation: The ID number obtained through identity information is associated with signaling data to obtain basic information and trajectory of personnel flow investigation.

[0035] Example 2: Special Personnel Trajectory Process Judgment In this embodiment, when a special person passes a camera point, their identification as a key person is determined, and their activity trajectory for that day is retrieved through the personnel information retrieval module. Based on the timestamp of the key person's passing through a camera point, and in accordance with the corresponding definitions of special and spatiotemporal accompanying persons, all personnel information and activity trajectory information for that period of time at the corresponding camera point is extracted.

[0036] (1) Face matching algorithm: Advantages: By using regional feature data to represent the characteristics of facial organ areas and comparing them with photo information, the accuracy and efficiency of the comparison are improved. Compared with traditional global feature comparison, regional feature comparison can better handle problems such as occlusion and angle changes.

[0037] (2) Personal information track storage terminal: Advantages: Using the slime mold foraging method to save and analyze personal trajectory information, this method mimics the behavior of slime molds in finding food and can efficiently process and analyze large amounts of data. Compared with traditional data processing methods, it is more intelligent and adaptive.

[0038] (3) Facial feature recognition algorithm: Advantages: By considering factors such as walking posture and occlusion, the face pose coefficient is introduced to improve recognition accuracy. This method is more robust than simple feature matching and can adapt to more practical application scenarios.

[0039] (4) Multi-angle face acquisition: Advantages: Through the video image acquisition module and background subtraction method, multi-angle facial images can be extracted from dynamic videos, increasing the comprehensiveness and accuracy of recognition. Compared with single-angle facial recognition, multi-angle recognition can provide richer information.

[0040] (5) Personnel information retrieval: Advantages: By linking with the basic permanent resident information of the public security system, it can quickly retrieve and match personnel information, improving retrieval efficiency and accuracy. This method is more efficient and accurate than traditional manual retrieval.

[0041] (6) Face collection equipment and face recognition equipment: Advantages: Dedicated link connection ensures data transmission security and real-time performance. At the same time, the separation of the acquisition and identification ends improves the flexibility and scalability of the system.

[0042] (7) Database storage module: Advantages: It can store large amounts of facial images and trajectory information, supporting long-term data preservation and analysis. Compared with traditional storage methods, this modular design is easier to manage and maintain.

Claims

1. A system for automatically determining personnel flow trajectories based on face comparison, characterized by It includes video image acquisition module, face identification terminal, photo information calling module, photo decoding module, face capture module, face recognition module, database storage module, face comparison module, and personnel information retrieval module; The video image acquisition module in the face recognition terminal detects pedestrian information in real time in the captured video screen. The pedestrian information includes image information and video information; The face recognition module extracts facial image information based on the facial feature values ​​of the collected image information and video information. The personal information trajectory is saved by saving the successfully recognized facial information and the personal trajectory information called by the database storage module to form a personal trajectory information result set; the face comparison module compares the image extracted by the face recognition end based on the human feature values ​​with the called photo information. After a successful comparison, the personal ID number is extracted according to the photo information, and the identity information of the public security system is called to obtain the personal identity information of the compared person; the personal information trajectory is saved to the database storage module through the slime mold foraging method, and the movement trajectory of the same person every day is saved under the personal identity information to track the target information. According to the definition of close contacts, secondary close contacts, and constant accompanying persons, the trajectory information of the target in the previous and next dates is found for data fusion. Finally, the trajectory is analyzed by combining the shooting time and monitoring location information, thereby deriving daily images and trajectories containing the target and the contact target, and the target information and contact target information are further confirmed through feedback search.

2. A method for automatically determining personnel flow trajectories based on face comparison, characterized in that The steps include: Step 1: Multi-angle face acquisition Through the video image acquisition module, the image information obtained by the real-time detection camera is preprocessed, and the dynamic video image sequence is frame-cut to obtain static picture information. Then, the background subtraction method is used to detect moving targets to obtain pictures containing facial information, and the facial feature vectors in the multi-angle face images are extracted. The facial feature vector data collected and the facial feature vector data obtained by obtaining photo information are recorded in a vector matrix, and the pictures containing facial information are saved in the database for use in step 2; Step 2: Facial feature recognition algorithm Based on the facial information image stored in step 1 and the Euclidean distance of the facial feature vector, the facial feature data scaling weight value is obtained. The facial feature similarity comparison result is obtained based on the facial regression learning function by taking into account the influence of the posture of the person when walking, such as looking up, looking down, hair occlusion, and profile. The comparison result is then used in step 3; Step 3: Face matching algorithm Based on the basic information data of the facial regional features and the facial feature similarity comparison results obtained in step 3, the feature data of each organ region corresponding to the face is represented by the regional feature data, and compared with the obtained person's photo information to obtain the final facial similarity result, and the similarity result is applied to step 4; Step 4: Personal trajectory data fusion The comparison results from step 3 are integrated with daily personal trajectory information through the slime mold foraging method to track the target information. According to the definition of close contacts and temporal and spatial companions, the trajectory information of the target in the days before and after is found for data fusion. Finally, the trajectory is analyzed by combining the shooting time and monitoring location information to derive images and trajectories containing the target and the contact target. These images and trajectories are sent back to the personal information storage terminal for storage, and the target information and contact target information are further confirmed through feedback search. Step 5: Personnel Information Retrieval According to the confirmation of target information and contact target information in step 4, the basic information of permanent residents from the public security system's identity information is retrieved and associated with the identity information of the person's photo extracted by the face comparison module. The matched basic information of the person is screened and counted to match the person's name, ID card, home address information, and daily activity trajectory.

3. The method for automatically determining the flow trajectory of people based on face comparison according to claim 2 is characterized in that In step 1, multi-angle face acquisition includes the following: The face collection end device and the face recognition end device are connected via a dedicated line link; 1) The face collection terminal device includes: The video image acquisition module is used to detect pedestrian information in real time in the video images captured by the camera, cut the dynamic video image sequence into frames to obtain static image information, and then use background subtraction to detect moving targets to obtain images containing facial information. The image information is pre-processed, and the shooting time and monitoring location information are recorded and saved to the database. The database storage module is used to store facial images recognized by the face recognition module and save facial information and daily trajectory information for the previous and next fourteen days as agreed. 2) The face recognition terminal device includes: The face recognition module is used to identify the image information and video information collected by the video acquisition end module, and extract the facial feature vectors from the multi-angle face images according to the facial feature values ​​of the human body based on the pre-processed image information; 3) The facial feature vector data collected and the facial feature vector data obtained by obtaining photo information are recorded using a vector matrix of w=n*m, as shown below: = , = , W = n × m; in, ={x1, x2, x3...xn} represents the first row of the vector matrix, xi is the facial feature point vector of the i-th collected image, ={y1, y2, y3...yn} represents the first row of the obtained photo vector matrix, and yj is the facial feature point vector of the jth photo information.

4. The method for automatically determining the flow trajectory of people based on face comparison according to claim 2 is characterized in that In step 2, the facial feature recognition algorithm includes the following: The photo information calling module is used to call the photo information of the person in the public security system, decode the binary code of the photo called from the public security system according to the prescribed algorithm, identify the photo information of the ID card and compare it with the face recognized by the face recognition module; The distance formula of the facial feature vector of the facial feature data collected by the face recognition module is: (1) 、 are the horizontal and vertical coordinates of the i-th facial feature pixel on the face image; The distance formula of the facial feature vector of the face feature data obtained by obtaining the photo information is: (2) Taking into account the size of the faces captured due to the distance between the camera and the pedestrians during video capture, the scaling weighted value of the facial feature data captured by the face acquisition module is calculated using formulas (1) and (2): (3) Because the facial images collected by the face acquisition module are affected by the posture of the person when walking, such as looking up, looking down, hair occlusion, side face, etc., the face posture coefficient is introduced. Its calculation formula is as follows: (4) in is the scaling factor, which is used to avoid errors caused by the inconsistency between the scale of the face image collected by the face acquisition module and the standard face posture image obtained by acquiring photo information; Considering that the face acquisition module will collect multiple different facial information of the same person within the video frame time, in order to ensure the balance of the sample, the facial feature data obtained by obtaining photo information will be given a higher weight. Moreover, a person's photo information is stored in the ID card database. Therefore, the weighted function expression is obtained as follows: (5) Where N is the face acquisition module that will collect multiple different faces of the same person in multiple video frames. Based on the face regression learning function, the formula for facial feature similarity comparison is obtained: (6) in is the facial feature vector obtained by obtaining photo information, The facial feature vector data of the face is collected through the face recognition module.

5. The method for automatically determining the flow trajectory of people based on face comparison according to claim 2 is characterized in that In step 3, the face comparison algorithm includes the following: According to the basic information data of the facial region features, the data of the regional feature a (1≤a≤7) is used to represent the feature data of each organ region corresponding to the face, the data of the regional feature H1 represents the feature data of the forehead, the data of the regional feature H2 represents the feature data of the left eye, the data of the regional feature H3 represents the feature data of the right eye, the data of the regional feature H4 represents the feature data of the nose, the data of the regional feature H5 represents the feature data of the mouth, the data of the regional feature H6 represents the feature data of the left cheek, and the data of the regional feature H7 represents the feature data of the right cheek. When the average value of the facial features obtained by the face acquisition module is greater than the preset threshold of 0.7, it can be considered to be the same person. The average value vector is calculated as follows: (7) The calculation formula of the average value vector obtained by obtaining photo information is as follows: (8) According to the face comparison algorithm, the final face similarity formula is as follows: (9) When the difference between the average vector calculated by the face acquisition module and the average vector obtained by obtaining photo information is greater than 0.7, that is, when the similarity is greater than 70%, it can be considered that they are the same person.

6. The method for automatically determining the flow trajectory of people based on face comparison according to claim 2 is characterized in that In step 4, the personal trajectory data fusion includes the following: Using the first personal photo with a successful face match as a benchmark, the slime mold is initialized to obtain the facial feature vector of the first personal photo, and the shooting time and monitoring location information are recorded. The facial feature vector of the person successfully matched that day is extracted through the slime mold's expansion and contraction process, and compared with the first facial feature vector. Referring to the three-person face matching algorithm in step 2, the slime mold continues to contract until all facial feature vectors obtained that day are matched. The slime mold contraction process ends, and the trajectory matching and recognition process is completed. The successfully recognized face information and trajectory information are returned to the personal information trajectory storage terminal. Assume that the facial feature vectors of M images extracted based on the slime mold model are (10) in is the facial feature point vector of the i-th collected image, is the face pose coefficient, is the deviation vector of the feature pixel point of the i-th face when the slime mold forages; According to the learning characteristics of slime mold, the face regression learning function is used to obtain the final objective function of face image feature points and trajectory information. (11) in Collect the facial feature point vector for the first slime mold foraging image. is the trajectory offset between the i-th facial feature point vector and the first facial feature point vector; The personal information trajectory storage terminal uses the slime mold foraging method to save the daily movement trajectory of the same person under the personal identity information, which is used to track the target information. According to the definition of close contacts and temporal and spatial companions, the trajectory information of the target in the previous and subsequent dates is found for data fusion. Finally, the trajectory is analyzed by combining the shooting time and monitoring location information, thereby deriving daily images and trajectories containing the target and the contact target, and further confirming the target information and the contact target information through feedback search.

Images