Face recognition system and method based on recursive process

Abstract

A face recognition system and method based on a recursive process are provided. The face recognition system may comprise: a local device for generating, as detection data, a detection result obtained by using an object detection model with respect to an image locally collected using a camera; and a face recognition device for receiving the detection data from the local device and performing face recognition, wherein the face recognition device: stores the detection data in a storage medium; reads a plurality of frames included in a predetermined period from the storage medium; extracts a multi-dimensional face vector configured according to an index of the camera, an index of the period, an index of the frame, and an index of an extracted feature, the face vector being cumulatively stored in the storage medium in units of the period; executes a recursive process for performing identifier matching on the face vector for each of the plurality of frames; and performs face verification or face identification on the basis of the identifier for which matching has been performed through the recursive process.

Description

Facial recognition system and method based on a recursive process

[0001] Cross-citation with related applications

[0002] This application claims the benefit of priority based on Korean Patent Application No. 10-2024-0203036 filed December 31, 2024 and Korean Patent Application No. 10-2025-0153231 filed October 21, 2025, and all contents disclosed in the documents of said Korean patent applications are incorporated herein as part of this specification.

[0003] The disclosure relates to a facial recognition system and method, more specifically, to a facial recognition system and method based on a recursive process.

[0004] Face recognition technology has made significant advancements driven by the development of deep learning. Existing machine learning techniques and image processing technologies alone faced limitations in achieving commercially viable accuracy. This was because the complexity of image data and various environmental factors prevented the provision of sufficient precision in individual identification and facial comparison. Deep learning technology learns from massive volumes of facial images ranging from millions to tens of millions to extract patterns, demonstrating high performance in identifying identical individuals or distinguishing between different people based on these patterns. Such deep learning-based technologies are rapidly being commercialized. The problem, however, is that since current face recognition technology has evolved based on deep learning, it is essentially a form of supervised machine learning. Therefore, perfect identification of the same individual requires prior training on that person, which acts as a fundamental limitation of deep learning-based technology.

[0005] The task to be solved is to provide a facial recognition system and method based on a recursive process.

[0006] A face recognition system according to one embodiment includes: a local device that generates detection results obtained using an object detection model for images collected locally using a camera as detection data; and a face recognition device that receives the detection data from the local device and performs face recognition. The face recognition device stores the detection data in a storage medium, reads a plurality of frames included in a predetermined period in the storage medium, extracts a multidimensional face vector configured according to the index of the camera, the index of the period, the index of the frame, and the index of the extracted feature, and the face vector is accumulated and stored in the storage medium in units of the period, executes a recursive process that performs identifier matching for the face vector for each of the plurality of frames, and performs face verification or face identification based on the identifier matched through the recursive process.

[0007] The above facial recognition device can determine the similarity Sim[D1][D2] between a plurality of features belonging to the first frame and a plurality of features belonging to the second frame between a first frame and a second frame belonging to the same camera and the same period according to the following mathematical formula 1:

[0008] (Mathematical Formula 1)

[0009] Sim[D1][D2] = ∑ | E[A][B][C1][D1] - E[A][B][C2][D2] | b

[0010] Here, A is the index value of the camera (A is a natural number), B is the index value of the period (B is a natural number), C1 is the index value of the first frame (C1 is a natural number), C2 is the index value of the second frame (C2 is a natural number), D1 is the index value of any one of the multiple features belonging to the first frame (D1 is a natural number), D2 is the index value of any one of the multiple features belonging to the second frame (D2 is a natural number), and b represents a constant such that b > 1.

[0011] The above facial recognition device generates a similarity table in which each of a plurality of features belonging to the first frame is used as a row index and each of a plurality of features belonging to the second frame is used as a column index, and in the similarity table, the similarity values ​​between the i-th feature of the first frame (i is a natural number) and the j-th feature of the second frame (j is a natural number) can be arranged at positions corresponding to the intersection cells for the row and the column.

[0012] The above facial recognition device can determine the n-th feature (n is a natural number) among a plurality of features belonging to the second frame that has a minimum similarity to the m-th feature (m is a natural number) belonging to the first frame, and assign the same identifier to the m-th feature of the first frame and the n-th feature of the second frame.

[0013] The above facial recognition device may assign a new identifier to the p-th feature (where p is a natural number) among a plurality of features belonging to the second frame that has a similarity exceeding a predetermined level for all features belonging to the first frame.

[0014] The above facial recognition device, according to the following mathematical formula 2, has a cumulative average feature E A [B][ID] can be determined:

[0015] (Mathematical Formula 2)

[0016] E A [B][ID] = ( Face vector with identifier ID in the first frame + Face vector with identifier ID in the second frame ) / 2

[0017] Here, B represents the index value of the above period (B is a natural number), and ID represents the value of the above identifier (ID is a natural number).

[0018] The above facial recognition device can determine the similarity Sim[ID][D3] between the accumulated average feature and a plurality of features belonging to the third frame according to the following mathematical formula 3:

[0019] (Mathematical Formula 3)

[0020] Sim[ID][D3] = ∑ | E[B][ID] - E[A][B][C3][D3] | b

[0021] Here, A is the index value of the camera (A is a natural number), B is the index value of the period (B is a natural number), C3 is the index value of the third frame (C3 is a natural number), ID is the value of the identifier (ID is a natural number), D3 is the index value of any one of the multiple features belonging to the third frame (D4 is a natural number), and b is a constant such that b > 1.

[0022] The above facial recognition device generates a similarity table in which each of a plurality of features belonging to the cumulative average feature is used as a row index and each of a plurality of features belonging to the third frame is used as a column index, and in the similarity table, the similarity values ​​between the i-th feature of the cumulative average feature (i is a natural number) and the j-th feature of the third frame (j is a natural number) can be arranged at positions corresponding to the intersection cells for the row and the column.

[0023] The above facial recognition device can determine the n-th feature (n is a natural number) that has a minimum similarity to the m-th feature (m is a natural number) belonging to the cumulative average feature among a plurality of features belonging to the third frame, and assign the same identifier to the m-th feature of the cumulative average feature and the n-th feature of the second frame.

[0024] The above facial recognition device can delete an identifier for the p-th feature (p is a natural number) that has a similarity exceeding a predetermined level for all features belonging to the third frame among a plurality of features belonging to the cumulative average feature.

[0025] The above facial recognition device, according to the following mathematical formula 4, cumulative average feature E A [B][ID] can be updated:

[0026] (Mathematical Formula 4)

[0027] E A [B][ID] = ( Face vector with identifier ID in the above cumulative average feature + Face vector with identifier ID in the above third frame ) / 2

[0028] Here, B represents the index value of the above period (B is a natural number), and ID represents the value of the above identifier (ID is a natural number).

[0029] A face recognition method according to one embodiment is a face recognition method performed by a computing device comprising a processor, a storage medium, and a communication interface, wherein the processor receives detection data from a local device through the communication interface, the detection data including detection results obtained using an object detection model for images collected locally using a camera; storing the detection data in the storage medium; reading a plurality of frames included in a predetermined time unit in the storage medium; extracting a multidimensional face vector configured according to the index of the camera, the index of the period, the index of the frame, and the index of the extracted feature, and the face vector is accumulated and stored in the storage medium in the period unit; executing a recursive process that performs identifier matching for the face vector for each of the plurality of frames; and performing face verification or face identification based on the identifier matched through the recursive process.

[0030] The step of executing the above recursive process may include the step of determining the similarity Sim[D1][D2] between a plurality of features belonging to the first frame and a plurality of features belonging to the second frame, between a first frame and a second frame belonging to the same camera and the same period, according to the following mathematical formula 1:

[0031] (Mathematical Formula 1)

[0032] Sim[D1][D2] = ∑ | E[A][B][C1][D1] - E[A][B][C2][D2] | b

[0033] Here, A is the index value of the camera (A is a natural number), B is the index value of the period (B is a natural number), C1 is the index value of the first frame (C1 is a natural number), C2 is the index value of the second frame (C2 is a natural number), D1 is the index value of any one of the multiple features belonging to the first frame (D1 is a natural number), D2 is the index value of any one of the multiple features belonging to the second frame (D2 is a natural number), and b represents a constant such that b > 1.

[0034] The step of executing the above recursive process may further include: determining the n-th feature (n is a natural number) among a plurality of features belonging to the second frame such that the similarity is minimized with respect to the m-th feature (m is a natural number) belonging to the first frame; and assigning the same identifier to the m-th feature of the first frame and the n-th feature of the second frame.

[0035] The step of executing the above recursive process may further include the step of assigning a new identifier to the p-th feature (where p is a natural number) among a plurality of features belonging to the second frame, which has a similarity exceeding a predetermined level for all features belonging to the first frame.

[0036] The step of executing the above recursive process is the cumulative average feature E according to the following mathematical formula 2. A It may include an additional step for determining [B][ID]:

[0037] (Mathematical Formula 2)

[0038] E A [B][ID] = ( Face vector with identifier ID in the first frame + Face vector with identifier ID in the second frame ) / 2

[0039] Here, B represents the index value of the above period (B is a natural number), and ID represents the value of the above identifier (ID is a natural number).

[0040] The step of executing the above recursive process is,

[0041] The method may further include the step of determining the similarity Sim[ID][D3] between the cumulative average feature and a plurality of features belonging to the third frame according to the following mathematical formula 3:

[0042] (Mathematical Formula 3)

[0043] Sim[ID][D3] = ∑ | E[B][ID] - E[A][B][C3][D3] | b

[0044] Here, A is the index value of the camera (A is a natural number), B is the index value of the period (B is a natural number), C3 is the index value of the third frame (C3 is a natural number), ID is the value of the identifier (ID is a natural number), D3 is the index value of any one of the multiple features belonging to the third frame (D4 is a natural number), and b is a constant such that b > 1.

[0045] The step of executing the above recursive process may further include: determining the n-th feature (n is a natural number) among a plurality of features belonging to the third frame that has a minimum similarity to the m-th feature (m is a natural number) belonging to the cumulative average feature; and assigning the same identifier to the m-th feature of the cumulative average feature and the n-th feature of the second frame.

[0046] The step of executing the above recursive process may further include the step of deleting an identifier for the p-th feature (p is a natural number) among a plurality of features belonging to the above cumulative average feature, which has a similarity exceeding a predetermined level for all features belonging to the third frame.

[0047] The step of executing the above recursive process is the cumulative average feature E according to the following mathematical formula 4. A It may include an additional step to update [B][ID]:

[0048] (Mathematical Formula 4)

[0049] E A [B][ID] = ( Face vector with identifier ID in the above cumulative average feature + Face vector with identifier ID in the above third frame ) / 2

[0050] Here, B represents the index value of the above period (B is a natural number), and ID represents the value of the above identifier (ID is a natural number).

[0051] Unlike existing deep learning-based facial recognition algorithms, the present invention does not require prior training with image or video data, so no additional training costs for service enhancement are incurred even as the number of people increases. Accordingly, all additional costs, such as cloud costs for video collection and storage, video data labeling and anonymization processing costs, cloud costs for deep learning training, MLops and DEVops-related costs, and labor and incidental expenses for skilled personnel, can be eliminated or minimized. Furthermore, the present invention enables real-time database creation even for people seen for the first time and adopts an algorithm that performs training and testing processes simultaneously to enable real-time facial recognition and management using only feature data extracted from a single image. Moreover, since the collection of video or images is not required, personally identifiable information may not be collected.

[0052] FIG. 1 is a drawing for explaining a facial recognition system according to one embodiment.

[0053] FIG. 2 is a drawing for explaining a facial recognition system and method according to one embodiment.

[0054] FIG. 3 is a drawing for explaining an example of implementation of a facial recognition system and method according to one embodiment.

[0055] FIG. 4 is a drawing for explaining a facial recognition method according to one embodiment.

[0056] FIG. 5 is a drawing for explaining a facial recognition method according to one embodiment.

[0057] FIGS. 6 to 10 are drawings for explaining an example of implementation regarding a facial recognition method and device according to one embodiment.

[0058] FIG. 11 is a block diagram illustrating a computing device according to one embodiment.

[0059] Embodiments of the present invention are described below with reference to the attached drawings so that those skilled in the art can easily implement them. However, the present invention may be embodied in various different forms and is not limited to the embodiments described herein. Furthermore, in order to clearly explain the present invention in the drawings, parts unrelated to the explanation have been omitted, and similar parts throughout the specification are denoted by similar reference numerals.

[0060] Throughout the specification and claims, when a part is described as "comprising" a certain component, this means that, unless specifically stated otherwise, it does not exclude other components but may include additional components. Terms including ordinal numbers, such as first, second, etc., may be used to describe various components, but said components are not limited by said terms. Such terms are used solely for the purpose of distinguishing one component from another.

[0061] Terms such as "...part," "...unit," and "module" as used in the specification may refer to a unit capable of processing at least one function or operation described in this specification, and may be implemented as hardware or circuit, software, or a combination of hardware or circuit and software. Additionally, at least some components or functions of the recursive process-based facial recognition system and method according to the embodiments described below may be implemented as a program or software, and the program or software may be stored on a computer-readable medium.

[0062] FIG. 1 is a drawing for explaining a facial recognition system according to one embodiment.

[0063] Referring to FIG. 1, a facial recognition system (1) according to one embodiment may include a facial recognition device (10) and local devices (20, 21). The facial recognition device (10) and local devices (20, 21) may exchange data with each other through a network (40).

[0064] The facial recognition device (10) and the local devices (20, 21) may be computing devices. Specifically, the facial recognition device (10) and the local devices (20, 21) may each include a processor, a storage medium, and a communication interface. Here, the term "storage medium" is used as a concept that comprehensively includes memory devices and storage devices (or storage). The facial recognition device (10) and the local devices (20, 21) may each execute program code or instructions loaded into one or more memory devices through one or more processors. For example, the facial recognition device (10) and the local devices (20, 21) may each be implemented as a computing device (50) as described below in relation to FIG. 11. In this case, one or more processors may correspond to the processor (510) of the computing device (50), and one or more memory devices may correspond to the memory (520) of the computing device (50). Program code or instructions can be executed by one or more processors to perform the functions of a facial recognition device (10) and local devices (20, 21) according to embodiments described in this specification.

[0065] In some embodiments, the local device (20, 21) may be placed locally where the images to be analyzed are collected, and the facial recognition device (10) may be placed at a remote location. For example, the local device (20, 21) may be installed and operated at a store (e.g., a cafe) where images are to be collected and analyzed, and the facial recognition device (10) may be installed and operated at another location outside the store.

[0066] In some embodiments, the local device (20, 21) may be implemented as an edge artificial intelligence device, and the facial recognition device (10) may be implemented as a central server. The local device (20, 21), which is an edge artificial intelligence device, is a device capable of processing data in real time at the site where the data is generated, and can collect data from peripheral devices such as cameras, sensors, and IoT (Internet of Things) devices, and perform preliminary analysis or inference thereon. The facial recognition device (10) is implemented as a central server and can integrate and analyze data transmitted from the local device.

[0067] The local device (20, 21) can collect video locally. For example, the local device (20, 21) can collect video captured inside the store using a device such as a camera installed in the store. The local device (20, 21) can obtain detection results for the collected video using an object detection model. The object detection model may be an artificial intelligence algorithm that detects the location of a specific object in an image or video and classifies the object. The object detection model is based on a Convolutional Neural Network (CNN) and can perform tasks such as predicting the area where an object exists in an input image and determining the type of the object. The object detection model may include, for example, YOLO (You Only Look Once).

[0068] The local device (20, 21) can generate detection results obtained using an object detection model on collected images as detection data. Here, detection data refers only to the results of recognizing an object among the tasks performed by the object detection model, and may refer to detection results for which no final inference has been performed.

[0069] The facial recognition device (10) can receive detection data from local devices (20, 21). Additionally, the facial recognition device (10) can perform facial recognition using an inference module. That is, the facial recognition device (10) can provide the detection data received from the local devices (20, 21) to an inference module that performs facial analysis of a detected person. The inference results inferred by the inference module can be aggregated by a predetermined identifier (e.g., ID) to generate inference information by identifier. Accordingly, inference information for an anonymous person assigned an ID can be aggregated, and inference information for another anonymous person assigned a different ID can be aggregated.

[0070] The facial recognition device (10) can store detection data received from local devices (20, 21) in a storage medium. In some embodiments, the facial recognition device (10) can store detection data in a database.

[0071] The face recognition device (10) can read multiple frames, for example, N frames (N is a natural number), included in a predetermined time unit, for example, a 5-minute unit, from a storage medium. The face recognition device (10) can execute a recursive process that performs identifier matching for face vectors for each of the multiple frames. That is, the face recognition device can perform a recursive process for each of the N frames. The face recognition device (10) can perform face verification or face identification based on the identifiers matched through the recursive process.

[0072] Facial verification is a process of confirming whether a user's facial data matches specific pre-registered facial data, and it is typically used in applications such as identity authentication or access control. In this process, a similarity score is calculated between the input facial data and the target facial data, and the identity can be determined based on a pre-set threshold. On the other hand, facial identification is a process of identifying who a face belongs to by comparing input facial data with multiple facial records stored in a database, and it can be utilized in applications such as person identification or criminal investigation. In this process, a match is performed between the input facial data and all facial data included in the database, and the facial data with the highest matching score represents the identity of the input face.

[0073] Executing a recursive process may be as follows. A face recognition device (10) calculates the similarity between one or more preceding frame face vectors derived from a preceding frame among a plurality of frames and one or more succeeding frame face vectors derived from a succeeding frame, and can match an identifier between one or more preceding frame face vectors and one or more succeeding frame face vectors according to the similarity.

[0074] Matching may be performed as follows. A face recognition device (10) may match an identifier between a first preceding frame face vector and a first succeeding frame face vector when the similarity value between a first preceding frame face vector among one or more preceding frame face vectors and a first succeeding frame face vector among one or more succeeding frame face vectors is less than or equal to a preset first reference value.

[0075] Meanwhile, the facial recognition device (10) may assign a new identifier to the second subsequent frame facial vector when all similarity values ​​calculated between the second subsequent frame facial vector among the one or more subsequent frame facial vectors for each of the one or more preceding frame facial vectors exceed a preset second reference value.

[0076] Executing a recursive process may further include updating the value of a face vector among one or more trailing frame face vectors where an identifier is matched.

[0077] In some embodiments, the local device (20, 21) can generate detection data by using an object detection model on images collected locally using a camera. The face recognition device (10) can store the detection data in a storage medium, e.g., a database, and extract a multidimensional face vector by reading multiple frames included in a predetermined period from the storage medium. Here, the face vector may be configured according to the index of the camera, the index of the period, the index of the frame, and the index of the extracted feature. The face recognition device (10) can accumulate and store the face vector in the storage medium in units of a predetermined period. Below, an embodiment regarding a recursive process executed by the face recognition device (10) will be described in detail.

[0078] The facial recognition device (10) can determine the similarity between features between multiple frames belonging to the same camera and the same period. Specifically, the facial recognition device (10) can determine the similarity between multiple features belonging to the first frame and multiple features belonging to the second frame between a first frame and a second frame belonging to the same camera and the same period. Here, the first frame may be a preceding frame for the second frame, and the second frame may be a succeeding frame for the first frame.

[0079] In some embodiments, the facial recognition device (10) can determine the similarity Sim[D1][D2] according to the following mathematical formula 1.

[0080] (Mathematical Formula 1)

[0081] Sim[D1][D2] = ∑ | E[A][B][C1][D1] - E[A][B][C2][D2] | b

[0082] Here, A is the index value of the camera (A is a natural number), B is the index value of the period (B is a natural number), C1 is the index value of the first frame (C1 is a natural number), C2 is the index value of the second frame (C2 is a natural number), D1 is the index value of any one of the multiple features belonging to the first frame (D1 is a natural number), D2 is the index value of any one of the multiple features belonging to the second frame (D2 is a natural number), and b can represent a constant such that b > 1.

[0083] Subsequently, the facial recognition device (10) can generate a similarity table. Here, the similarity table may represent a data structure in which each of the multiple features belonging to the first frame is used as a row index, and each of the multiple features belonging to the second frame is used as a column index. In the similarity table, at the position corresponding to the intersection cell for the row and column, the similarity values ​​between the i-th feature of the first frame (i is a natural number) and the j-th feature of the second frame (j is a natural number) may be arranged.

[0084] The facial recognition device (10) can assign an identifier according to a feature based on a similarity table. Specifically, the facial recognition device (10) can determine the nth feature (n is a natural number) that has the minimum similarity to the mth feature (m is a natural number) belonging to the first frame among a plurality of features belonging to the second frame, and assign the same identifier to the mth feature of the first frame and the nth feature of the second frame. Meanwhile, the facial recognition device (10) can assign a new identifier to the pth feature (p is a natural number) that has a similarity exceeding a predetermined level for all features belonging to the first frame among a plurality of features belonging to the second frame.

[0085] Next, the facial recognition device (10) can determine a cumulative average feature. Here, the cumulative average feature can be determined based on the average of the facial vector of the first frame and the facial vector of the second frame corresponding to a specific identifier.

[0086] In some embodiments, the facial recognition device (10) has a cumulative average feature E according to the following mathematical formula 2. A [B][ID] can be determined.

[0087] (Mathematical Formula 2)

[0088] E A [B][ID] = ( Face vector with identifier ID in the 1st frame + Face vector with identifier ID in the 2nd frame ) / 2

[0089] Here, B can represent the index value of the period (B is a natural number), and ID can represent the value of the identifier (ID is a natural number).

[0090] Next, the facial recognition device (10) can determine the similarity between a cumulative average feature and a plurality of features belonging to a third frame. Here, the third frame may be a subsequent frame to the second frame.

[0091] In some embodiments, the facial recognition device (10) can determine the similarity Sim[ID][D3] between a cumulative average feature and a plurality of features belonging to a third frame according to the following mathematical formula 3.

[0092] (Mathematical Formula 3)

[0093] Sim[ID][D3] = ∑ | E[B][ID] - E[A][B][C3][D3] | b

[0094] Here, A is the index value of the camera (A is a natural number), B is the index value of the period (B is a natural number), C3 is the index value of the third frame (C3 is a natural number), ID is the value of the identifier (ID is a natural number), D3 is the index value of any one of the multiple features belonging to the third frame (D4 is a natural number), and b is a constant such that b > 1.

[0095] Subsequently, the facial recognition device (10) can generate a similarity table. Here, the similarity table may represent a data structure in which each of the multiple features belonging to the cumulative average feature is used as a row index, and each of the multiple features belonging to the third frame is used as a column index. In the similarity table, the similarity values ​​between the i-th feature of the cumulative average feature (i is a natural number) and the j-th feature of the third frame (j is a natural number) may be arranged at positions corresponding to the intersection cells for the row and column.

[0096] The facial recognition device (10) can assign an identifier according to a feature based on a similarity table. Specifically, the facial recognition device (10) can determine the nth feature (n is a natural number) that has the minimum similarity to the mth feature (m is a natural number) belonging to the cumulative average feature among a plurality of features belonging to the third frame, and can assign the same identifier to the mth feature of the cumulative average feature and the nth feature of the second frame. Meanwhile, the facial recognition device (10) can delete the identifier for the pth feature (p is a natural number) that has a similarity exceeding a predetermined level for all features belonging to the third frame among a plurality of features belonging to the cumulative average feature.

[0097] Next, the facial recognition device (10) can update the cumulative average feature. Here, the cumulative average feature can be updated based on the average of the facial vector of the cumulative average feature corresponding to a specific identifier and the facial vector of the third frame.

[0098] In some embodiments, the facial recognition device (10) has a cumulative average feature E according to the following mathematical formula 4. A [B][ID] can be updated.

[0099] (Mathematical Formula 4)

[0100] E A [B][ID] = ( Face vector with identifier ID in the cumulative average feature + Face vector with identifier ID in the third frame ) / 2

[0101] Here, B can represent the index value of the period (B is a natural number), and ID can represent the value of the identifier (ID is a natural number).

[0102] According to the present embodiment, by recursively calculating the similarity of face vectors between temporally consecutive image frames, updating or assigning a new identifier using the similarity result, and accumulating and updating the average value of feature vectors corresponding to the same identifier, stable and reliable recognition results can be obtained even under the temporal continuity of images.

[0103] Specifically, for multiple frames belonging to the same period, by determining the similarity between features of a preceding frame and features of a succeeding frame, it is possible to identify the same person's vector based on the relative distance between features, even if the facial vectors obtained from individual frames differ slightly due to changes in lighting, facial orientation, or facial expressions. In particular, by arranging the similarity calculated in this way into a table-like data structure, the interrelationships between features across frames can be intuitively identified, and feature vectors of the same person can be efficiently matched by selecting the cell with the lowest similarity. This allows for the maintenance of a consistent identifier even if facial vectors corresponding to the same person exist across multiple frames, and consequently provides recognition results that are robust against inter-frame fluctuations or noise. Meanwhile, if a specific feature within the similarity table has a high similarity distance exceeding a predetermined threshold for all existing features, that feature is identified as a new person and assigned a new identifier; thereby, the recognition target group can be automatically expanded without separate initialization or manual processing even when a new person appears. This is very useful in environments where multiple people enter and exit in real time, and can significantly improve the scalability and autonomy of the system.

[0104] Furthermore, consistency in recognition results over time can be maintained by continuously calculating the average of face vectors corresponding to the same identifier to generate a cumulative average feature, which is then periodically updated. In other words, by repeatedly calculating the average of feature vectors across frames corresponding to the same identifier, the impact of temporary image distortion or detection errors in some frames on the overall recognition result can be mitigated. This cumulative averaging process stabilizes the distribution of face vectors along the time axis, thereby improving the overall robustness of the system. Moreover, by hierarchizing the similarity calculation and identifier matching processes within the same cycle into an inter-frame comparison stage and a subsequent frame comparison stage using the cumulative average feature, computational efficiency can also be expected. Consequently, real-time recognition is possible while maintaining fast processing speeds, even when there are a large number of frames or features.

[0105] In other words, the facial recognition device accumulates and stores facial vectors on a periodic basis, and by referencing the accumulated average features of the previous period during subsequent frame recognition, it can maintain recognition continuity between sessions. Since the average vector accumulated from the previous period is used as the initial reference value for the next period even after one period ends, the same person can be stably tracked without interruption in recognition even at the boundaries between periods. This enables the provision of consistent ID tracking results across the entire range, regardless of discontinuities in viewpoints or divisions of periodic storage intervals, even in environments where video streams are continuously input from multiple cameras. Furthermore, since the recursive process updates only the feature vectors corresponding to the matched identifiers, it eliminates the need to reprocess all data in every period, thereby reducing the processing load on the database. Consequently, real-time facial recognition operating with low latency is possible even in environments where large volumes of video data are continuously input, and efficient use of data storage space and computational resources is enabled.

[0106] FIG. 2 is a drawing for explaining a facial recognition system and method according to one embodiment.

[0107] Referring to FIG. 2, in relation to a facial recognition system and method according to one embodiment, in step (S201), a local device (20) selects a frame from a local image, and in step (S202), for the selected frame, a detection result based on an object detection model can be generated as detection data. Subsequently, in step (S203), the local device (20) can transmit the detection data to a facial recognition device (10).

[0108] In step (S204), the facial recognition device (10) may execute a recursive process for identifier matching. Subsequently, in step (S205), the facial recognition device (10) may perform facial recognition based on the identifier for which matching has been performed.

[0109] Meanwhile, in step (S206), the local device (20) can generate detection data based on an object detection model for another frame. Here, the other frame may be a frame after the frame selected in step (S202). Additionally, in step (S207), the local device (20) can generate detection data based on an object detection model for yet another frame. Here, the other frame may be a frame after the frame selected in step (S206). Next, in step (S208), the local device (20) can perform compression and encryption on the detection data generated in steps (S206) through (S207). Subsequently, in step (S209), the local device (20) can transmit the detection data, after compression and encryption are completed, to the face recognition device (10). That is, the local device (20) can repeat the task of generating detection data frame by frame from the local image, and when a certain amount of repetition is completed, it can collect the generated detection data and provide it to the face recognition device (10) at once. In step (S210), the face recognition device (10) can execute a recursive process for identifier matching on the collected and received data. Subsequently, in step (S211), the face recognition device (10) can perform face recognition based on the identifier for which matching has been performed.

[0110] FIG. 3 is a drawing for explaining an example of implementation of a facial recognition system and method according to one embodiment.

[0111] Referring to FIG. 3, the detection data may include a set of multiple values ​​corresponding to each of a predetermined set of items.

[0112] The facial recognition detection data (31) may be implemented to include values ​​regarding items such as screen coordinates, height (height), width, center, screen depth, number of bounding boxes, and recognition score, but this is an example of implementation to aid understanding and does not limit the scope of the present invention.

[0113] Next, the human body recognition detection data (32) may be implemented to include values ​​regarding items such as screen coordinates, height (face size), width, center, screen depth, number of bounding boxes, recognition score, and whether it is front or back, but this is an example of implementation to aid understanding and does not limit the scope of the present invention.

[0114] FIG. 4 is a drawing for explaining a facial recognition method according to one embodiment.

[0115] Referring to FIG. 4, a face recognition method according to one embodiment may include the steps of: receiving detection data from a local device including detection results obtained using an object detection model for images collected locally (S401); storing the detection data in a storage medium (S402); reading a plurality of frames included in a predetermined time unit in the storage medium (S403); executing a recursive process for performing identifier matching for face vectors for each of the plurality of frames (S404); and performing face verification or face identification based on the identifier matched through the recursive process (S405).

[0116] For more detailed information regarding the facial recognition method according to the present embodiment, reference may be made to other embodiments described in this specification; therefore, redundant descriptions are omitted here.

[0117] FIG. 5 is a drawing for explaining a facial recognition method according to one embodiment.

[0118] Referring to FIG. 5, a face recognition method according to one embodiment may include the steps of: reading a plurality of frames included in a predetermined time unit in a storage medium (S501); calculating a similarity between one or more preceding frame face vectors derived from a preceding frame among the plurality of frames and one or more subsequent frame face vectors derived from a subsequent frame (S502); matching an identifier between one or more preceding frame face vectors and one or more subsequent frame face vectors according to the similarity (S503); updating the value of a face vector among one or more subsequent frame face vectors for which an identifier has been matched (S504); and increasing the indices of the preceding frame and the subsequent frame and performing a similarity calculation (S505).

[0119] For more detailed information regarding the facial recognition method according to the present embodiment, reference may be made to other embodiments described in this specification; therefore, redundant descriptions are omitted here.

[0120] FIGS. 6 to 10 are drawings for explaining an example of implementation regarding a facial recognition method and device according to one embodiment.

[0121] Referring to FIG. 6, the facial recognition device (10) can determine the similarity between features between multiple frames belonging to the same camera and the same period. Specifically, the facial recognition device (10) can determine the similarity between multiple features belonging to the first frame (F[1][1][1]) and multiple features belonging to the second frame (F[1][1][2]) between a first frame (F[1][1][1]) and a second frame (F[1][1][2]) belonging to the same camera and the same period. Here, the first frame (F[1][1][1]) may be a preceding frame for the second frame (F[1][1][2]), and the second frame (F[1][1][2]) may be a succeeding frame for the first frame (F[1][1][1]).

[0122] In some embodiments, the facial recognition device (10) can determine the similarity Sim[D1][D2] according to the following mathematical formula 1.

[0123] (Mathematical Formula 1)

[0124] Sim[D1][D2] = ∑ | E[A][B][C1][D1] - E[A][B][C2][D2] | b

[0125] Here, A is the index value of the camera (A is a natural number), B is the index value of the period (B is a natural number), C1 is the index value of the first frame (F[1][1][1]) (C1 is a natural number), C2 is the index value of the second frame (F[1][1][2]) (C2 is a natural number), D1 is the index value of any one of the multiple features belonging to the first frame (F[1][1][1]) (D1 is a natural number), D2 is the index value of any one of the multiple features belonging to the second frame (F[1][1][2]) (D2 is a natural number), and b can represent a constant such that b > 1.

[0126] Specifically, the facial recognition device (10) can determine the similarity between feature 1 (E[1][1][1][1][1]) of the first frame (F[1][1][1]) and feature 1 (E[1][1][2][1]) of the second frame (F[1][1][2]). For example, feature 1 (E[1][1][1][1]) of the first frame (F[1][1][1]) is [-1.209242, 0.048394, ...] T And, feature 1 (E[1][1][2][1]) of the second frame (F[1][1][2]) is [2.245223, -2.246733, ...] T In this case, the similarity can be determined as follows.

[0127] Sim[1][1] = | -1.209242 - 2.245223 | b + | 0.048394 - -2.246733 | b + ...

[0128] Similarly, the facial recognition device (10) can determine the similarity between feature 1 (E[1][1][1][1][1]) of the first frame (F[1][1][1]) and feature 2 (E[1][1][2][2]) of the second frame (F[1][1][2]), the similarity between feature 1 (E[1][1][1][1]) of the first frame (F[1][1][1]) and feature 3 (E[1][1][2][3]) of the second frame (F[1][1][2]), and the similarity between feature 1 (E[1][1][1][1]) of the first frame (F[1][1][1]) and feature 4 (E[1][1][2][4]) of the second frame (F[1][1][2]).

[0129] Subsequently, the facial recognition device (10) has a similarity between feature 2 (E[1][1][1][2]) of the first frame (F[1][1][1]) and feature 1 (E[1][1][2][1]) of the second frame (F[1][1][2]), a similarity between feature 2 (E[1][1][1][2]) of the first frame (F[1][1][1]) and feature 2 (E[1][1][2][2]) of the second frame (F[1][1][2]), a similarity between feature 2 (E[1][1][1][2]) of the first frame (F[1][1][1]) and feature 3 (E[1][1][2][3]) of the second frame (F[1][1][2]), and the first The similarity between feature 2 (E[1][1][1][2]) of frame (F[1][1][1]) and feature 4 (E[1][1][2][4]) of the second frame (F[1][1][2]) can be determined.

[0130] Subsequently, the facial recognition device (10) has a similarity between feature 3 (E[1][1][1][3]) of the first frame (F[1][1][1]) and feature 1 (E[1][1][2][1]) of the second frame (F[1][1][2]), a similarity between feature 3 (E[1][1][1][3]) of the first frame (F[1][1][1]) and feature 2 (E[1][1][2][2]) of the second frame (F[1][1][2]), a similarity between feature 3 (E[1][1][1][3]) of the first frame (F[1][1][1]) and feature 3 (E[1][1][2][3]) of the second frame (F[1][1][2]), and the first The similarity between feature 3 (E[1][1][1][3]) of frame (F[1][1][1]) and feature 4 (E[1][1][2][4]) of frame 2 (F[1][1][2]) can be determined.

[0131] Subsequently, referring to FIGS. 7 and FIGS. 8, the facial recognition device (10) can generate a similarity table. Here, the similarity table may be implemented as a data structure in which each of the multiple features belonging to the first frame (F[1][1][1]) is used as a row index, and each of the multiple features belonging to the second frame (F[1][1][2]) is used as a column index. In the similarity table, at the position corresponding to the intersection cell for the row and column, the similarity values ​​between the i-th feature (i is a natural number) of the first frame (F[1][1][1]) and the j-th feature (j is a natural number) of the second frame (F[1][1][2]) may be arranged.

[0132] The facial recognition device (10) can assign an identifier according to a feature based on a similarity table. Here, the facial recognition device (10) determines the n-th feature (n is a natural number) that has the minimum similarity to the m-th feature (m is a natural number) belonging to the first frame (F[1][1][1]) among a plurality of features belonging to the second frame (F[1][1][2]), and can assign the same identifier to the m-th feature of the first frame (F[1][1][1]) and the n-th feature of the second frame (F[1][1][2]).

[0133] Specifically, among the multiple features belonging to the second frame (F[1][1][2]), the minimum similarity value for Feature 1 of the first frame (F[1][1][1]) is 3000, so that the same identifier "ID1" can be assigned to Feature 1 of the first frame (F[1][1][1]) and Feature 1 of the second frame (F[1][1][2]). Meanwhile, among the multiple features belonging to the second frame (F[1][1][2]), the minimum similarity value for Feature 2 of the first frame (F[1][1][1]) is 650, so that the same identifier "ID2" can be assigned to Feature 2 of the first frame (F[1][1][1]) and Feature 2 of the second frame (F[1][1][2]). Meanwhile, among the multiple features belonging to the second frame (F[1][1][2]), the minimum similarity value for feature 3 of the first frame (F[1][1][1]) is 3800, so that the same identifier "ID3" can be assigned to feature 3 of the first frame (F[1][1][1]) and feature 4 of the second frame (F[1][1][2]).

[0134] Meanwhile, the facial recognition device (10) may assign a new identifier to the p-th feature (p is a natural number) that has a similarity exceeding a predetermined level for all features belonging to the first frame (F[1][1][1]) among a plurality of features belonging to the second frame (F[1][1][2]). Specifically, a new identifier may be assigned to feature 3 that has a similarity exceeding a predetermined level for all features belonging to the first frame (F[1][1][1]) among a plurality of features belonging to the second frame (F[1][1][2]).

[0135] Next, the facial recognition device (10) can determine a cumulative average feature. Here, the cumulative average feature can be determined based on the average of the facial vector of the first frame (F[1][1][1]) and the facial vector of the second frame (F[1][1][2]) corresponding to a specific identifier.

[0136] In some embodiments, the facial recognition device (10) has a cumulative average feature E according to the following mathematical formula 2. A [B][ID] can be determined.

[0137] (Mathematical Formula 2)

[0138] E A [B][ID] = ( Face vector with identifier ID in the first frame (F[1][1][1]) + Face vector with identifier ID in the second frame (F[1][1][2]) / 2

[0139] Here, B can represent the index value of the period (B is a natural number), and ID can represent the value of the identifier (ID is a natural number).

[0140] Specifically, the facial recognition device (10) has a cumulative average feature (E) for identifier 1 of cycle 1. A [1][1]), cumulative average feature E for identifier 2 of cycle 1 A [1][2], cumulative average feature E for identifier 3 of cycle 1 A [1][3], cumulative average feature E for identifier 4 of cycle 1 A [1][4] can be determined.

[0141] Next, the facial recognition device (10) can determine the similarity between a cumulative average feature and a plurality of features belonging to a third frame (F[1][1][3]) as shown in FIG. 6. Here, the third frame (F[1][1][3]) may be a subsequent frame to the second frame (F[1][1][2]).

[0142] In some embodiments, the facial recognition device (10) can determine the similarity Sim[ID][D3] between a cumulative average feature and a plurality of features belonging to a third frame (F[1][1][3]) according to the following mathematical formula 3.

[0143] (Mathematical Formula 3)

[0144] Sim[ID][D3] = ∑ | E[B][ID] - E[A][B][C3][D3] | b

[0145] Here, A is the index value of the camera (A is a natural number), B is the index value of the period (B is a natural number), C3 is the index value of the third frame (F[1][1][3]) (C3 is a natural number), ID is the value of the identifier (ID is a natural number), D3 is the index value of any one of the multiple features belonging to the third frame (F[1][1][3]) (D4 is a natural number), and b is a constant such that b > 1.

[0146] Specifically, the facial recognition device (10) has a cumulative average feature (E A The similarity between [1][1]) and the third frame (F[1][1][3]) and feature 1 (E[1][1][3][1]) can be determined. For example, the cumulative average feature (E A [1][1]) is [-1.209242, 0.048394, ...] T And, feature 1 (E[1][1][3][1]) of the 3rd frame (F[1][1][3]) is [2.245223, -2.246733, ...] T In this case, the similarity can be determined as follows.

[0147] Sim[1][1] = | -1.209242 - 2.245223 | b + | 0.048394 - -2.246733 | b + ...

[0148] Similarly, the facial recognition device (10) has a cumulative average feature (E A [1][1]) and the similarity of feature 2 (E[1][1][3][2]) of the third frame (F[1][1][3]), and the cumulative average feature (E A The similarity between [1][1]) and feature 3 (E[1][1][3][3]) of the third frame (F[1][1][3]) can be determined.

[0149] Afterwards, the facial recognition device (10) has a cumulative average feature (E ASimilarity between [1][2]) and the third frame (F[1][1][3]) and feature 1 (E[1][1][3][1]), cumulative average feature (E A [1][2]) and the similarity of feature 2 (E[1][1][3][2]) of the third frame (F[1][1][3]), and the cumulative average feature (E A [1][2]) and the similarity between feature 3 (E[1][1][3][3]) of the third frame (F[1][1][3]) can be determined.

[0150] Afterwards, the facial recognition device (10) has a cumulative average feature (E A Similarity between [1][3]) and Feature 1 (E[1][1][3][1]) of the third frame (F[1][1][3]), cumulative average feature (E A [1][3]) and the similarity of feature 2 (E[1][1][3][2]) of the third frame (F[1][1][3]), and the cumulative average feature (E A The similarity between [1][3]) and feature 3 (E[1][1][3][3]) of the third frame (F[1][1][3]) can be determined.

[0151] Afterwards, the facial recognition device (10) has a cumulative average feature (E A Similarity between [1][4]) and Feature 1 (E[1][1][3][1]) of the third frame (F[1][1][3]), cumulative average feature (E A [1][4]) and the similarity of feature 2 (E[1][1][3][2]) of the third frame (F[1][1][3]), and the cumulative average feature (E A [1][4]) and the similarity between feature 3 (E[1][1][3][3]) of the third frame (F[1][1][3]) can be determined.

[0152] Subsequently, referring to FIGS. 9 and FIGS. 10, the facial recognition device (10) can generate a similarity table. Here, the similarity table may be implemented as a data structure in which each of the multiple features belonging to the cumulative average feature is used as a row index, and each of the multiple features belonging to the third frame (F[1][1][3]) is used as a column index. In the similarity table, the similarity values ​​between the i-th feature of the cumulative average feature (i is a natural number) and the j-th feature of the third frame (F[1][1][3]) (j is a natural number) may be arranged at positions corresponding to the intersection cells for the row and column.

[0153] The facial recognition device (10) can assign an identifier according to a feature based on a similarity table. Specifically, the facial recognition device (10) determines the nth feature (n is a natural number) that has the minimum similarity to the mth feature (m is a natural number) belonging to the cumulative average feature among a plurality of features belonging to the third frame (F[1][1][3]), and can assign the same identifier to the mth feature of the cumulative average feature and the nth feature of the second frame (F[1][1][2]).

[0154] Specifically, among the multiple features belonging to the third frame (F[1][1][3]), the minimum similarity value for the feature corresponding to ID1 of the cumulative average feature is 598, so the same identifier "ID1" can be assigned to the feature corresponding to ID1 of the cumulative average feature and feature 2 of the third frame (F[1][1][3]). Meanwhile, among the multiple features belonging to the third frame (F[1][1][3]), the minimum similarity value for the feature corresponding to ID2 of the cumulative average feature is 1359, so the same identifier "ID2" can be assigned to the feature corresponding to ID2 of the cumulative average feature and feature 1 of the third frame (F[1][1][3]). Meanwhile, among the multiple features belonging to the third frame (F[1][1][3]), the minimum similarity value for the feature corresponding to ID4 of the cumulative average feature is 1728, so the same identifier "ID4" can be assigned to the feature corresponding to ID4 of the cumulative average feature and feature 3 of the third frame (F[1][1][3]).

[0155] Meanwhile, the facial recognition device (10) can delete the identifier for the p-th feature (p is a natural number) that has a similarity exceeding a predetermined level for all features belonging to the third frame (F[1][1][3]) among a plurality of features belonging to the cumulative average feature. Specifically, it can delete the identifier "ID3" for the feature corresponding to ID3 of the cumulative average feature that has a similarity exceeding a predetermined level for all features belonging to the third frame (F[1][1][3]) among a plurality of features belonging to the cumulative average feature.

[0156] Next, the facial recognition device (10) can update the cumulative average feature. Here, the cumulative average feature can be updated based on the average of the facial vector of the cumulative average feature corresponding to a specific identifier and the facial vector of the third frame (F[1][1][3]).

[0157] In some embodiments, the facial recognition device (10) has a cumulative average feature E according to the following mathematical formula 4. A [B][ID] can be updated.

[0158] (Mathematical Formula 4)

[0159] E A [B][ID] = ( Face vector with identifier ID in the cumulative average feature + Face vector with identifier ID in the third frame (F[1][1][3]) ) / 2

[0160] Here, B can represent the index value of the period (B is a natural number), and ID can represent the value of the identifier (ID is a natural number).

[0161] Specifically, the facial recognition device (10) has a cumulative average feature (E) for identifier 1 of cycle 1. A [1][1]) reflects the cumulative average feature (E[1][1][3][2]) of the third frame (F[1][1][3]). A [1][1]) can be updated. In addition, the facial recognition device (10) can update the cumulative average feature E for identifier 2 of cycle 1. A [1][2] is a cumulative average feature (E[1][1][3][1]) that reflects feature 1 (E[1][1][3][1]) of the third frame (F[1][1][3]). A [1][2]) can be updated. The facial recognition device (10) has a cumulative average feature E for identifier 4 of cycle 1. A [1][4] is a cumulative average feature (E[1][1][3][3]) that reflects feature 4 (E[1][1][3][3]) of the third frame (F[1][1][3]). A [1][4]) can be updated. Meanwhile, the cumulative average feature E for identifier 3 of cycle 1 A [1][3] may not change.

[0162] By repeating this recursive process in other cycles, facial recognition based on the recursive process can be implemented.

[0163] FIG. 11 is a block diagram illustrating a computing device according to one embodiment.

[0164] Referring to FIG. 11, a facial recognition system and method according to embodiments may be implemented using a computing device (50). This computing device (50) may be implemented as various types of electronic devices, servers, or similar devices, and its functions may be implemented through a combination of software and hardware.

[0165] The computing device (50) may include at least one of a processor (501) communicating via a bus (509), a memory (502), a storage device (503), a display device (504), a network interface device (505) providing access to a network (40) for communication with other entities, and an input / output interface device (506) providing a user input interface or a user output interface. Of course, the computer device (50) may additionally include any electronic device necessary to implement the technical concept described in this specification, although not shown in FIG. 11.

[0166] The processor (501) can be implemented as various types of computing devices, such as an MCU (Micro Controller Unit), AP (Application Processor), CPU (Central Processing Unit), GPU (Graphic Processing Unit), NPU (Neural Processing Unit), QPU (Quantum Processing Unit), etc. The processor (501) is a semiconductor device that executes instructions stored in memory (502) or storage device (503) and can perform a core role in the system. Program code and data stored in memory (502) or storage device (503) instruct the processor (501) to perform specific tasks, thereby enabling the operation of the entire system. The processor (501) can be configured to implement the functions or methods described above in relation to FIGS. 1 to 10.

[0167] The memory (502) and storage device (503) may include various forms of volatile or non-volatile storage media for storing and accessing data of the system. For example, the memory (502) may include read-only memory (ROM) or random access memory (RAM). In some embodiments, the memory (502) may be embedded inside the processor (501), in which case the data transfer speed between the memory (502) and the processor (501) may be very fast. In some other embodiments, the memory (502) may be located outside the processor (501), in which case the memory (502) may be connected to the processor (501) through various data buses or interfaces. Such connection may be made through various known means, for example, a PCIe (Peripheral Component Interconnect Express) interface for high-speed data transfer or a memory controller. Meanwhile, examples of storage devices (503) include HDD (Hard Disk Drive) or SSD (Solid State Drive), and the scope of the present invention is not limited to the elements listed above for the purpose of explanation.

[0168] In some embodiments, at least some configurations or functions of the facial recognition system and method according to the embodiments may be implemented as a program or software executed on a computing device (50), and the program or software may be stored on a computer-readable recording medium or storage medium. Specifically, a computer-readable recording medium or storage medium according to one embodiment may have a program recorded thereon for executing steps included in the implementation of the facial recognition system and method according to the embodiments on a computer including a processor (501) that executes a program or instructions stored in a memory (502) or a storage device (503).

[0169] In some embodiments, at least some configurations or functions of the facial recognition system and method according to the embodiments may be implemented using hardware or circuits of the computing device (50), or may be implemented using separate hardware or circuits that can be electrically connected to the computing device (50).

[0170] According to the embodiments described so far, unlike existing deep learning-based facial recognition algorithms, the present invention does not require prior training of image or video data, so no additional training costs for service enhancement are incurred even if the number of people increases. Accordingly, all additional costs, such as cloud costs for image collection and storage, image data labeling and anonymization processing costs, cloud costs for deep learning training, MLops and DEVops-related costs, and labor and incidental costs for skilled personnel, can be eliminated or minimized. Furthermore, the present invention enables real-time database creation even for people seen for the first time and adopts an algorithm that performs training and testing processes simultaneously to enable real-time facial recognition and management using only feature data extracted from a single image. Moreover, since the collection of video or images is not required, personal identification information may not be collected.

[0171] Although embodiments of the present invention have been described in detail above, the scope of the present invention is not limited thereto, and various modifications and improvements by those skilled in the art to which the present invention belongs, using the basic concept of the present invention as defined in the following claims, also fall within the scope of the present invention.

Claims

1. A local device that generates detection results obtained using an object detection model on images collected locally using a camera as detection data; and It includes a facial recognition device that receives detection data from the local device and performs facial recognition, The above facial recognition device is, The above detection data is stored in a storage medium, and A plurality of frames included in a predetermined period in the above storage medium are read, and A multidimensional face vector configured according to the index of the camera, the index of the period, the index of the frame, and the index of the extracted feature is extracted, and the face vector is accumulated and stored in the storage medium in units of the period. A recursive process is executed to perform identifier matching for the face vector for each of the plurality of frames, and Performing face verification or face identification based on the identifier for which matching was performed through the recursive process above. Facial recognition system.

2. In Paragraph 1, The above facial recognition device is, A facial recognition system that determines a similarity Sim[D1][D2] between a plurality of features belonging to the first frame and a plurality of features belonging to the second frame, between a first frame and a second frame belonging to the same camera and the same period according to the following mathematical formula 1: (Mathematical Formula 1) Sim[D1][D2] = ∑ | E[A][B][C1][D1] - E[A][B][C2][D2] | b Here, A is the index value of the camera (A is a natural number), B is the index value of the period (B is a natural number), C1 is the index value of the first frame (C1 is a natural number), C2 is the index value of the second frame (C2 is a natural number), D1 is the index value of any one of the multiple features belonging to the first frame (D1 is a natural number), D2 is the index value of any one of the multiple features belonging to the second frame (D2 is a natural number), and b represents a constant such that b > 1.

3. In Paragraph 2, The above facial recognition device is, A similarity table is created in which each of the plurality of features belonging to the first frame is used as a row index, and each of the plurality of features belonging to the second frame is used as a column index. A facial recognition system in which, in the above similarity table, the similarity values ​​between the i-th feature (i is a natural number) of the first frame and the j-th feature (j is a natural number) of the second frame are arranged at positions corresponding to the intersection cells for the row and the column.

4. In Paragraph 2, The above facial recognition device is, Among a plurality of features belonging to the second frame, the n-th feature (n is a natural number) that has a minimum similarity to the m-th feature (m is a natural number) belonging to the first frame is determined, and A facial recognition system that assigns the same identifier to the m-th feature of the first frame and the n-th feature of the second frame.

5. In Paragraph 4, The above facial recognition device is, A facial recognition system that assigns a new identifier to the p-th feature (p is a natural number) among a plurality of features belonging to the second frame, which has a similarity exceeding a predetermined level for all features belonging to the first frame.

6. In Paragraph 5, The above facial recognition device is, Cumulative average feature E according to the following mathematical formula 2 A Facial recognition system determining [B][ID]: (Mathematical Formula 2) E A [B][ID] = ( Face vector with identifier ID in the first frame + Face vector with identifier ID in the second frame ) / 2 Here, B represents the index value of the above period (B is a natural number), and ID represents the value of the above identifier (ID is a natural number).

7. In Paragraph 6, The above facial recognition device is, A facial recognition system that determines the similarity Sim[ID][D3] between the above cumulative average feature and a plurality of features belonging to a third frame according to the following mathematical formula 3: (Mathematical Formula 3) Sim[ID][D3] = ∑ | E[B][ID] - E[A][B][C3][D3] | b Here, A is the index value of the camera (A is a natural number), B is the index value of the period (B is a natural number), C3 is the index value of the third frame (C3 is a natural number), ID is the value of the identifier (ID is a natural number), D3 is the index value of any one of the multiple features belonging to the third frame (D4 is a natural number), and b is a constant such that b > 1.

8. In Paragraph 7, The above facial recognition device is, A similarity table is created in which each of the multiple features belonging to the above cumulative average feature is used as a row index, and each of the multiple features belonging to the above third frame is used as a column index. A facial recognition system in which, in the above similarity table, the similarity values ​​between the i-th feature (i is a natural number) of the cumulative average feature and the j-th feature (j is a natural number) of the third frame are arranged at positions corresponding to the intersection cells for the above row and the above column.

9. In Paragraph 7, The above facial recognition device is, Among the plurality of features belonging to the third frame, the n-th feature (n is a natural number) that has a minimum similarity to the m-th feature (m is a natural number) belonging to the cumulative average feature is determined, and A facial recognition system that assigns the same identifier to the m-th feature of the cumulative average feature and the n-th feature of the second frame.

10. In Paragraph 9, The above facial recognition device is, A facial recognition system that deletes an identifier for the p-th feature (p is a natural number) having a similarity exceeding a predetermined level for all features belonging to the third frame among a plurality of features belonging to the above cumulative average feature.

11. In Paragraph 10, The above facial recognition device is, Cumulative average feature E according to the following mathematical formula 4 A Facial recognition system updating [B][ID]: (Mathematical Formula 4) E A [B][ID] = ( Face vector with identifier ID in the above cumulative average feature + Face vector with identifier ID in the above third frame ) / 2 Here, B represents the index value of the above period (B is a natural number), and ID represents the value of the above identifier (ID is a natural number).

12. A facial recognition method performed by a computing device comprising a processor, a storage medium, and a communication interface, wherein The step of the processor receiving detection data from a local device, including detection results obtained using an object detection model on images collected locally using a camera, through the communication interface; A step of storing the above detection data in the storage medium; A step of reading a plurality of frames included in a predetermined period in the above storage medium; A step of extracting a multidimensional face vector configured according to the index of the camera, the index of the period, the index of the frame, and the index of the extracted feature, and accumulating and storing the face vector in the storage medium in units of the period; A step of executing a recursive process for performing identifier matching for the face vector for each of the plurality of frames; and A step comprising performing face verification or face identification based on the identifier for which matching was performed through the recursive process above. Facial recognition method.

13. In Paragraph 12, The step of executing the above recursive process is, A face recognition method comprising the step of determining a similarity Sim[D1][D2] between a plurality of features belonging to the first frame and a plurality of features belonging to the second frame, between a first frame and a second frame belonging to the same camera and the same period according to the following mathematical formula 1: (Mathematical Formula 1) Sim[D1][D2] = ∑ | E[A][B][C1][D1] - E[A][B][C2][D2] | b Here, A is the index value of the camera (A is a natural number), B is the index value of the period (B is a natural number), C1 is the index value of the first frame (C1 is a natural number), C2 is the index value of the second frame (C2 is a natural number), D1 is the index value of any one of the multiple features belonging to the first frame (D1 is a natural number), D2 is the index value of any one of the multiple features belonging to the second frame (D2 is a natural number), and b represents a constant such that b > 1.

14. In Paragraph 13, The step of executing the above recursive process is, A step of determining the n-th feature (n is a natural number) among a plurality of features belonging to the second frame, such that the similarity is minimized with respect to the m-th feature (m is a natural number) belonging to the first frame; and A facial recognition method further comprising the step of assigning the same identifier to the m-th feature of the first frame and the n-th feature of the second frame.

15. In Paragraph 14, The step of executing the above recursive process is, A facial recognition method further comprising the step of assigning a new identifier to a p-th feature (p is a natural number) among a plurality of features belonging to the second frame, which has a similarity exceeding a predetermined level for all features belonging to the first frame.

16. In Paragraph 15, The step of executing the above recursive process is, Cumulative average feature E according to the following mathematical formula 2 A A facial recognition method including an additional step of determining [B][ID]: (Mathematical Formula 2) E A [B][ID] = ( Face vector with identifier ID in the first frame + Face vector with identifier ID in the second frame ) / 2 Here, B represents the index value of the above period (B is a natural number), and ID represents the value of the above identifier (ID is a natural number).

17. In Paragraph 16, The step of executing the above recursive process is, A face recognition method further comprising the step of determining the similarity Sim[ID][D3] between the cumulative average feature and a plurality of features belonging to a third frame according to the following mathematical formula 3: (Mathematical Formula 3) Sim[ID][D3] = ∑ | E[B][ID] - E[A][B][C3][D3] | b Here, A is the index value of the camera (A is a natural number), B is the index value of the period (B is a natural number), C3 is the index value of the third frame (C3 is a natural number), ID is the value of the identifier (ID is a natural number), D3 is the index value of any one of the multiple features belonging to the third frame (D4 is a natural number), and b is a constant such that b > 1.

18. In Paragraph 17, The step of executing the above recursive process is, A step of determining the n-th feature (n is a natural number) that has a minimum similarity to the m-th feature (m is a natural number) belonging to the cumulative average feature among a plurality of features belonging to the third frame; and A facial recognition method further comprising the step of assigning the same identifier to the m-th feature of the cumulative average feature and the n-th feature of the second frame.

19. In Paragraph 18, The step of executing the above recursive process is, A facial recognition method further comprising the step of deleting an identifier for the p-th feature (p is a natural number) having a similarity exceeding a predetermined level for all features belonging to the third frame among a plurality of features belonging to the above-mentioned cumulative average feature.

20. In Paragraph 19, The step of executing the above recursive process is, Cumulative average feature E according to the following mathematical formula 4 A A facial recognition method that includes an additional step of updating [B][ID]: (Mathematical Formula 4) E A [B][ID] = ( Face vector with identifier ID in the above cumulative average feature + Face vector with identifier ID in the above third frame ) / 2 Here, B represents the index value of the above period (B is a natural number), and ID represents the value of the above identifier (ID is a natural number).

Images