Stranger face recognition system, method, legged humanoid robot and medium

Abstract

The application discloses a stranger face recognition system and method, a foot type humanoid robot and a medium, which are applied to the technical field of image processing, adopt a multi-task face recognition model, perform face boundary frame prediction and face key point extraction on an environment image to obtain a face key point set in each face boundary frame, align the face key point set in each face boundary frame to a standard template face image to obtain a non-distortion face image, extract a face feature vector of a face to be recognized from each non-distortion face image, calculate the similarity between the face feature vector of each face to be recognized and the face feature vector of each temporary reference face in a lightweight face database, and determine whether the face to be recognized is a stranger face based on the similarity. In this way, stranger face recognition can be realized on the foot type humanoid robot, network delay caused by cloud comparison is avoided, real-time interaction response in a public environment is realized, and human-computer interaction efficiency is improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of image processing, and in particular to a stranger face recognition system and method, a foot-type humanoid robot and a medium. BACKGROUND

[0002] With the rapid development of foot-type humanoid robots and embodied intelligence, foot-type humanoid robots in public environments are increasingly widely used in human-computer interaction, service guidance and safety monitoring and other tasks. As one of the core means of human-computer interaction, face recognition technology can achieve identity recognition, behavior recording and personalized services based on identity by detecting and comparing user face information.

[0003] In crowded and dynamically high environment public places such as shopping malls, stations and airports, foot-type humanoid robots need to achieve real-time face recognition and interaction in complex lighting, frequent personnel flow and multiple target simultaneous appearance and other complex scenes, which puts higher requirements on the robustness, power consumption and response delay of the face recognition system.

[0004] The existing face recognition system mostly adopts a cloud comparison method, and the foot-type humanoid robot uploads the collected face image to the cloud for comparison and verification. The cloud comparison process introduces network delay, which is difficult to meet the requirements of real-time interaction of the robot, thereby affecting the interaction efficiency of the robot in public places. SUMMARY

[0005] The present application provides a stranger face recognition system, method, foot-type humanoid robot and medium to solve the problem of low efficiency of stranger face recognition of foot-type humanoid robots in public environments. The technical solutions provided by the present application are as follows:

[0006] On the one hand, the present application provides a stranger face recognition system deployed in a foot-type humanoid robot, comprising a video stream acquisition module, an image quality optimization module, a face feature extraction module and a face identity determination module.

[0007] The video stream acquisition module is integrated on the body of the foot-type humanoid robot and is used to acquire environmental video streams in public environments.

[0008] The image quality optimization module is connected with the video stream acquisition module and is used to use a lightweight quality repair model to repair the quality of environmental images in the environmental video streams to obtain target environmental images.

[0009] The face feature extraction module is connected with the image quality optimization module, and is configured to adopt a multi-task face recognition model to perform face bounding box prediction and face key point extraction on the target environment image to obtain a face bounding box set and a face key point set in each face bounding box in the face bounding box set; based on a standard template key point set, the face key point set in each face bounding box is aligned to a standard template face image to obtain a non-distortion face image corresponding to each face bounding box; a multi-task feature extraction model is adopted to extract a global face visual feature vector and a multi-dimensional face attribute feature vector of the face to be recognized from the non-distortion face image corresponding to each face bounding box as a face feature vector;

[0010] The face identity determination module is connected with the face feature extraction module, and is configured to calculate the similarity between the face feature vector of the face to be recognized in each face bounding box and the face feature vector of each temporary reference face dynamically stored in the local lightweight face database, and determine whether the face to be recognized in each face bounding box is a stranger face based on the similarity; wherein the face feature vector of each temporary reference face dynamically stored in the local lightweight face database is the face feature vector of each stranger face recognized in a recent set time range.

[0011] Optionally, the image quality optimization module is configured to, for each frame of environment image in the environment video stream, calculate an image quality score of the environment image by using an image quality evaluation function; when the image quality score is less than a first quality threshold, discarding the environment image; when the image quality score is greater than or equal to the first quality threshold and less than a second quality threshold, repairing the motion blur, image noise and illumination anomaly of the environment image by using a lightweight quality repair model to obtain a target environment image; when the image quality score is greater than or equal to the second quality threshold, determining the environment image as the target environment image.

[0012] Optionally, the multi-task face recognition model comprises a first feature map extraction branch and a three-level detection head connected in parallel with the first feature map extraction branch; wherein the three-level detection head is a face two-classification detection branch, a face bounding box regression branch and a face key point regression branch.

[0013] The face feature extraction module is configured to, after extracting first shared semantic feature maps from the target environment image by a first feature map extraction branch in the multi-task face recognition model, perform face binary classification detection, face bounding box detection, and face key point detection in parallel based on the first shared semantic feature maps by a face binary classification detection branch, a face bounding box regression branch, and a face key point regression branch in the multi-task face recognition model to obtain face binary classification detection results, face bounding box detection results, and face key point detection results; and perform post-processing on the face binary classification detection results, the face bounding box detection results, and the face key point detection results to obtain a face bounding box set corresponding to the target environment image and a face key point set in each face bounding box in the face bounding box set.

[0014] Optionally, the face feature extraction module is configured to, for each face bounding box, calculate an affine transformation matrix between the face key point set in the face bounding box and a standard template key point set, and align the face key point set in the face bounding box to a calibrated template face image based on the affine transformation matrix to obtain a non-distortion face image corresponding to the face bounding box.

[0015] Optionally, the multi-task feature extraction model includes a second feature map extraction branch and a two-level detection head connected in parallel to the second feature map extraction branch; the two-level detection head includes a global visual feature extraction branch and a multi-attribute feature extraction branch.

[0016] The face feature extraction module is configured to, after extracting second shared semantic feature maps from the non-distortion face image corresponding to each face bounding box by a second feature map extraction branch in the multi-task feature extraction model, perform global visual feature extraction and multi-attribute feature extraction in parallel based on the second shared semantic feature maps corresponding to each face bounding box by a global visual feature extraction branch and a multi-attribute feature extraction branch in the multi-task feature extraction model to obtain a global face visual feature vector and a multi-dimensional face attribute feature vector corresponding to each face bounding box; and perform post-processing on the global face visual feature vector and the multi-dimensional face attribute feature vector corresponding to each face bounding box to obtain a face feature vector corresponding to each face bounding box.

[0017] Optionally, the face identity determination module is configured to calculate the Euclidean distance between the face feature vector of each face in the face bounding box and the face feature vector of each temporary reference face dynamically stored in the local lightweight face database by using a similarity calculation method based on the Euclidean distance; for each face in the face bounding box, if the Euclidean distance between the face feature vector of the face to be identified and the face feature vector of any temporary reference face dynamically stored in the local lightweight face database is less than the distance threshold, it is determined that the face to be identified and the temporary reference face are the same face, and if the Euclidean distance between the face feature vector of the face to be identified and the face feature vector of all temporary reference faces dynamically stored in the local lightweight face database is greater than or equal to the distance threshold, it is determined that the face to be identified is a stranger face.

[0018] Optionally, the stranger face recognition system provided by the present application further comprises:

[0019] The data management module is connected with the face identity determination module and is configured to, for each face in the face bounding box, if the face to be identified and any temporary reference face in the local lightweight face database are the same face, update the entry time of any temporary reference face to the current time, if the face to be identified is a stranger face, assign a unique temporary identity to the stranger face, store the stranger face as a temporary reference face together with the face feature vector and the unique temporary identity to the local lightweight face database, and take the current time as the entry time; and automatically delete the unique temporary identity and the face feature vector of the temporary reference face in the local lightweight face database whose entry duration exceeds the storage duration threshold.

[0020] In another aspect, the present application provides a stranger face recognition method applied to a legged humanoid robot, comprising:

[0021] capturing an environment video stream in a public environment;

[0022] using a lightweight quality restoration model to perform quality restoration on the environment images in the environment video stream to obtain target environment images;

[0023] using a multi-task face recognition model to perform face bounding box prediction and face key point extraction on the target environment images to obtain a face bounding box set and a face key point set in each face bounding box in the face bounding box set;

[0024] aligning the face key point set in each face bounding box to a standard template face image based on a standard template key point set to obtain a non-distorted face image corresponding to each face bounding box;

[0025] The multi-task feature extraction model is used to extract a global face visual feature vector and a multi-dimensional face attribute feature vector of the face to be recognized from a non-distorted face image corresponding to each face bounding box as a face feature vector.

[0026] The similarity between the face feature vector of the face to be recognized in each face bounding box and the face feature vector of each temporary reference face dynamically stored in the local lightweight face database is calculated respectively.

[0027] Based on the similarity, it is determined whether the face to be recognized in each face bounding box is a stranger face; wherein the face feature vector of each temporary reference face dynamically stored in the local lightweight face database is the face feature vector of each stranger face recognized in the last set time range.

[0028] On the other hand, the application provides a legged humanoid robot, comprising a memory, a processor and a computer program stored on the memory and executable on the processor, wherein the processor implements the above-mentioned stranger face recognition method when executing the computer program.

[0029] On the other hand, the application provides a computer readable storage medium, which stores computer instructions, and the computer instructions are executed by a processor to implement the above-mentioned stranger face recognition method.

[0030] The beneficial effects of the application are as follows:

[0031] By deploying a lightweight quality repair model, a multi-task face recognition model, a multi-task feature extraction model and a local lightweight face database on the legged humanoid robot side, the image quality repair of the lightweight quality repair model can be used to solve the problem of low-quality conditions such as overexposure or underexposure, blur and the like caused by complex public environment lighting and motion blur caused by walking of the legged humanoid robot, to ensure the reliability of subsequent face recognition. The face bounding box recognition of the multi-task face recognition model, the face feature extraction of the multi-task feature extraction model and the stranger face discrimination of the local lightweight face database can be used to realize stranger face recognition on the legged humanoid robot side, avoid network delay introduced by cloud comparison, and thus realize real-time interaction response in a public environment, improve the understanding and service ability of the legged humanoid robot to the interactive object, and significantly improve the human-computer interaction experience in public places.

[0032] Other features and advantages of the application will be set forth in the following description, and in part will become apparent to those skilled in the art from the description, or can be learned by practice of the application. The objects and other advantages of the application can be realized and attained by the structure particularly pointed out in the written description and the accompanying drawings. BRIEF DESCRIPTION OF DRAWINGS

[0033] The accompanying drawings, which are included to provide a further understanding of the application and are incorporated in and constitute a part of this application, illustrate embodiments of the application and serve to explain the principles of the application. In the drawings:

[0034] Figure 1 The system framework schematic diagram of the stranger face recognition system in the present application;

[0035] Figure 2 The general flow schematic diagram of the stranger face recognition method in the present application;

[0036] Figure 3 The hardware structure schematic diagram of the foot-type humanoid robot in the present application. DETAILED DESCRIPTION

[0037] In order to make the purpose, technical scheme and beneficial effects of the present application more clear, the technical scheme of the present application will be described clearly and completely in combination with embodiments and drawings. Obviously, the described embodiments are only part of the embodiments of the present application, not all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor are within the scope of protection of the present application.

[0038] The stranger face recognition system provided by the embodiments of the present application is deployed in a foot-type humanoid robot. As shown in Figure 1 The stranger face recognition system 100 provided by the embodiments of the present application includes a video stream acquisition module 110, an image quality optimization module 120, a face feature extraction module 130 and a face identity determination module 140.

[0039] The video stream acquisition module 110 is integrated on the body of the foot-type humanoid robot, and is used to acquire the environment video stream in the public environment.

[0040] The image quality optimization module 120 is connected with the video stream acquisition module 110, and is used to adopt a lightweight quality repair model to perform quality repair on the environment image in the environment video stream to obtain a target environment image.

[0041] The face feature extraction module 130 is connected with the image quality optimization module 120, and is configured to adopt a multi-task face recognition model to perform face bounding box prediction and face key point extraction on the target environment image to obtain a face bounding box set and a face key point set in each face bounding box in the face bounding box set; based on a standard template key point set, align each face key point set in each face bounding box to a standard template face image to obtain a non-distortion face image corresponding to each face bounding box; and adopt a multi-task feature extraction model to extract a global face visual feature vector and a multi-dimensional face attribute feature vector of the to-be-identified face from each non-distortion face image corresponding to each face bounding box as a face feature vector.

[0042] The face identity determination module 140 is connected with the face feature extraction module 130, and is configured to calculate a similarity between the face feature vector of the to-be-identified face in each face bounding box and the face feature vector of each temporary reference face dynamically stored in the local lightweight face database, and determine whether the to-be-identified face in each face bounding box is a stranger face based on the similarity; wherein the face feature vector of each temporary reference face dynamically stored in the local lightweight face database is a face feature vector of each stranger face identified in a latest set time range.

[0043] In the embodiment of the present application, the color camera on the body (for example, the front of the chest) of the biped humanoid robot collects the environmental video stream in the public environment through the color camera during the walking process of the biped humanoid robot. A lightweight quality repair model, a multi-task face recognition model, a multi-task feature extraction model and a local lightweight face database are deployed on the side of the biped humanoid robot; the image quality optimization module 120 uses the lightweight quality repair model to repair the quality of the environmental image in the environmental video stream to obtain a target environmental image; after the face feature extraction module 130 uses the multi-task face recognition model to predict the face bounding box and extract the face key points of the target environmental image to obtain the face bounding box set and the face key point set in each face bounding box in the face bounding box set, based on the standard template key point set, the face key point set in each face bounding box is aligned to the standard template face image to obtain the non-distortion face image corresponding to each face bounding box, and the multi-task feature extraction model is used to extract the global face visual feature vector and multi-dimensional face attribute feature vector of the face to be recognized from the non-distortion face image corresponding to each face bounding box as the face feature vector; the face identity determination module 140 calculates the similarity between the face feature vector of the face to be recognized in each face bounding box and the face feature vector of each temporary reference face dynamically stored in the local lightweight face database, and determines whether the face to be recognized in each face bounding box is a stranger face based on the similarity. In this way, the image quality repair of the lightweight quality repair model can be used to solve the problem of low-quality conditions such as overexposure or underexposure, blur and the like of the environmental image caused by complex public environment lighting and motion blur caused by the walking of the biped humanoid robot, ensuring the reliability of subsequent face recognition, and the face boundary box recognition of the multi-task face recognition model, the face feature extraction of the multi-task feature extraction model and the stranger face discrimination of the local lightweight face database can be used to realize stranger face recognition on the side of the biped humanoid robot, thereby avoiding the network delay introduced by cloud comparison, realizing real-time interactive response in a public environment, improving the understanding and service ability of the biped humanoid robot to the interactive object, and significantly improving the human-computer interaction experience in public places.

[0044] In a possible implementation, the image quality optimization module 120 is configured to calculate the image quality score of each frame of environmental image in the environmental video stream using an image quality evaluation function; when the image quality score is less than a first quality threshold, the environmental image is discarded; when the image quality score is greater than or equal to the first quality threshold and less than a second quality threshold, the lightweight quality repair model is used to repair the motion blur, image noise and lighting abnormalities of the environmental image to obtain a target environmental image; when the image quality score is greater than or equal to the second quality threshold, the environmental image is determined as the target environmental image.

[0045] In the embodiments of the present application, the image quality evaluation function is defined as:

[0046]

[0047] wherein, I is an environment image; Q(I) is the image quality score of the environment image I; S(I) is the image sharpness index of the environment image I, which is used to quantify the sharpness of image details; C(I) is the image contrast index of the environment image I, which is used to measure the image gray distribution range; E(I) is the image exposure evaluation index of the environment image I, which is used to quantify the image brightness deviation; and a, b, and g are weighting coefficients.

[0048] The sharpness index S(I) is calculated in the following manner: the environment image I is converted into a luminance image Y∈ (H×W); a 3x3 convolution is performed on the luminance image Y using horizontal and vertical Sobel operators Gx and Gy to obtain gradient images Gx(i,j) and Gy(i,j); the gradient amplitude G(i,j) is calculated pixel by pixel; and the image sharpness index S(I) is obtained by averaging and normalizing the gradient amplitudes of the entire frame. The normalized image sharpness index S(I)∈[0,1].

[0049] The contrast index C(I) is calculated in the following manner: the environment image I is converted into a luminance image Y∈ (H×W); the global mean of the luminance image Y is calculated. The image contrast index C(I) is obtained by calculating the luminance standard deviation based on the global mean and normalizing it. The normalized image contrast index C(I)∈[0,1].

[0050] The exposure evaluation index E(I) is calculated in the following manner: the environment image I is converted into a luminance image Y∈ (H×W); the luminance image Y is quantized into a 64-bin histogram H[k], k=0…63; an ideal Gaussian weight is constructed as 31.5 is the center position of the 64-bin histogram (k=0…63), i.e., the mean μ of the Gaussian distribution, corresponding to the middle gray level, 2 is a fixed constant to ensure that the probability density function integrates to 1, and 8 is the standard deviation σ of the Gaussian distribution, which controls the weight decline speed and makes the luminance obtain the maximum weight near the center; the weighted histogram sum is calculated and normalized to obtain the image exposure index E(I). The normalized image exposure index E(I)∈[0,1], and the larger the value, the more concentrated the luminance distribution is in the middle gray, and the better the exposure.

[0051] Wherein, a, b, g are dynamically adjusted according to the light condition of the public environment and the motion state of the biped humanoid robot, and specifically include:

[0052] By setting the inertial measurement unit in the body of the biped humanoid robot and the force sensor on the sole, the motion acceleration time sequence signal of the biped humanoid robot is obtained in real time, and the motion impact spectrum energy in the predetermined frequency band is calculated to generate the image sharpness confidence at the current time;

[0053] By setting the ambient light sensor array beside the color camera, the ambient illumination jump strength and color temperature drift amount are obtained in real time to generate the image exposure confidence at the current time;

[0054] The image sharpness confidence and the image exposure confidence are input into the pre-trained lightweight mapping network, and the initial weighting coefficients a', b', g' are output after Softmax normalization, wherein, when the face detection recall rate is lower than the recall rate threshold and the motion impact spectrum energy exceeds the energy threshold or the illumination jump strength exceeds the strength threshold, a gradient descent update is triggered once, so that the network parameters of the lightweight mapping network are adapted to the new environment, and the initial weighting coefficients a', b', g' are output again through the lightweight mapping network;

[0055] The initial weighting coefficients a', b', g' output by the lightweight mapping network are subjected to inertia damping filtering to obtain the final weighting coefficients a, b, g to suppress the oscillation of the weighting coefficients, wherein a+b+g=1.

[0056] Further, in the embodiment of the application, the first quality threshold and the second quality threshold Two thresholds, low-quality images are discarded directly, that is, the frame is skipped, avoiding the entry of low-quality faces or false recognition, medium-quality images Perform enhancement processing, use a lightweight quality repair model trained based on the adversarial generation network technology to repair motion blur, noise, overexposure or underexposure and other abnormal conditions to obtain the target environment image, high-quality images Keep it as it is and determine it as the target environment image directly. The processed target environment image set is represented as The image repair of the lightweight quality repair model can effectively filter low-quality frames and improve medium-quality images, so that subsequent face detection and recognition remain stable and efficient in dynamic walking environment and complex lighting.

[0057] In one possible implementation, the multi-task face recognition model includes a first feature map extraction branch and a three-level detection head connected in parallel with the first feature map extraction branch; wherein, the three-level detection head is a face binary classification detection branch, a face bounding box regression branch, and a face key point regression branch;

[0058] The face feature extraction module 130 is used to extract features from the target environment image through the first feature map extraction branch in the multi-task face recognition model to obtain the first shared semantic feature map. Then, through the face binary classification detection branch, face bounding box regression branch, and face key point regression branch in the multi-task face recognition model, face binary classification detection, face bounding box detection, and face key point detection are performed in parallel based on the first shared semantic feature map to obtain face binary classification detection results, face bounding box detection results, and face key point detection results. The face binary classification detection results, face bounding box detection results, and face key point detection results are post-processed to obtain the face bounding box set corresponding to the target environment image and the face key point set within each face bounding box in the face bounding box set.

[0059] In this embodiment of the application, based on image quality optimization, the optimized target environment image set is... Face detection and key point extraction are performed to enable real-time face perception for legged humanoid robots in public environments. The k-th frame is the target environment image. The corresponding set of face bounding boxes is represented as follows: , For the detected first Top-left corner coordinates of the individual face bounding box and width and height ( , ), A face bounding box detection function; the target environment image of the k-th frame. The set of facial key points, including the left and right eyes, nose tip, mouth corner, eyebrows, forehead, and cheeks, within the i-th face bounding box in the corresponding face bounding box set is represented as follows: Keypoint regression is achieved by minimizing the mean squared error loss function. Let m be the total number of facial landmarks within the bounding box of the i-th face. Let J be the predicted coordinates of the j-th facial landmark. Let be the calibration coordinates of the j-th facial landmark. During training or evaluation, the location of the facial landmark given by the calibrator or high-precision equipment is considered to be completely correct and is used to compare with the network prediction value to calculate the loss or accuracy.

[0060] In one possible implementation, the face feature extraction module 130 is used to calculate the affine transformation matrix between the set of facial key points within the face bounding box and the set of standard template key points for each face bounding box, and then align the set of facial key points within the face bounding box to the calibration template face image based on the affine transformation matrix to obtain the undeformed face image corresponding to the face bounding box.

[0061] In this embodiment of the application, to improve the accuracy of face recognition, it is necessary to calibrate the set of facial key points within the detected face bounding box. Calibration aims to align the set of facial key points within the face bounding box to a unified standard coordinate system, eliminating pose deviations, rotations, and slight perspective distortions, thereby ensuring the stability of subsequent feature extraction and comparison. Specifically, this includes:

[0062] Based on the set of facial key points within the i-th face bounding box and standard template key point set Perform the following affine transformations or similarity transformations:

[0063]

[0064] in, Let be a rotation matrix. This is the scaling factor. It is a translation vector. The predicted coordinates of facial landmarks before calibration. These are the standard coordinates of the key points in the standard template.

[0065] The affine transformation matrix is ​​obtained by minimizing the keypoint alignment error and solving the above affine transformation or similarity transformation. :

[0066]

[0067] in, Let m be the keypoint alignment error loss function, and m be the total number of facial keypoints within the bounding box of the i-th face. Let J be the predicted coordinates of the j-th facial landmark. Let be the standard coordinates of the standard template keypoint corresponding to the j-th face keypoint.

[0068] Based on affine transformation matrix , will the Personal face bounding box image A calibrated, undistorted face image is obtained by aligning a standard template face image to a standard size and pose. :

[0069]

[0070] in, is a non-distorted face image, is an affine transformation matrix, is the i-th personal face bounding box image, T(·) is an image geometric transformation function (derivable, supporting back propagation) with bilinear interpolation as the sampling kernel.

[0071] In a possible implementation, the multi-task feature extraction model comprises a second feature map extraction branch, and a two-level detection head connected in parallel with the second feature map extraction branch; wherein the two-level detection head is a global visual feature extraction branch and a multi-attribute feature extraction branch;

[0072] The face feature extraction module 130 is configured to: after obtaining the second shared semantic feature map by performing feature extraction on the non-distorted face image corresponding to each face bounding box through the second feature map extraction branch in the multi-task feature extraction model, perform global visual feature extraction and multi-attribute feature extraction in parallel based on the second shared semantic feature map corresponding to each face bounding box through the global visual feature extraction branch and the multi-attribute feature extraction branch in the multi-task feature extraction model to obtain the global face visual feature vector and the multi-dimensional face attribute feature vector corresponding to each face bounding box; and perform post-processing on the global face visual feature vector and the multi-dimensional face attribute feature vector corresponding to each face bounding box to obtain the face feature vector corresponding to each face bounding box.

[0073] In the embodiments of the present application, the target of the multi-task feature extraction model is to obtain the global face visual feature vector and the multi-dimensional face attribute feature vector corresponding to each face bounding box, and specifically includes:

[0074] After obtaining the second shared semantic feature map by performing feature extraction on the non-distorted face image corresponding to the i-th face bounding box through the second feature map extraction branch in the multi-task feature extraction model, the global face visual feature vector is extracted based on the second shared semantic feature map through the global visual feature extraction branch in the multi-task feature extraction model , ) represents the global visual feature extraction branch, is the second shared semantic feature map, is the feature dimension, is the global face visual feature vector of the i-th personal face bounding box, and the multi-dimensional face attribute feature vector is extracted in parallel based on the second shared semantic feature map through the multi-attribute feature extraction branch in the multi-task feature extraction model , is age, is gender, is skin color, is expression,​ for eye state, for mouth state, for head pose, for whether wearing glasses, for whether wearing a mask. Multi-dimensional facial attribute feature prediction is realized by classification and regression models. For example, age is adopted by a regression function: ; binary or multi-class attributes such as gender, expression, whether wearing a mask, and whether wearing glasses are adopted by a cross-entropy loss function: ; the final multi-task joint loss function is: , is a global facial visual feature loss, is the jth facial attribute feature loss, and are dynamic weights, which can be automatically learned by Uncertainty Weighting or GradNorm. Through the multi-task feature extraction model, comprehensive extraction of facial features and attributes can be realized, ensuring that not only strangers can be recognized, but also rich attribute information can be obtained in a dynamic public environment.

[0075] In one possible implementation, the face identity determination module 140 is configured to calculate the Euclidean distance between the face feature vector of each face bounding box and the face feature vector of each temporary reference face dynamically stored in the local lightweight face database by using a similarity calculation method based on the Euclidean distance; for each face bounding box, if the Euclidean distance between the face feature vector of the face to be recognized and the face feature vector of any temporary reference face dynamically stored in the local lightweight face database is less than the distance threshold, it is determined that the face to be recognized and any temporary reference face are the same face, and if the Euclidean distance between the face feature vector of the face to be recognized and the face feature vector of all temporary reference faces dynamically stored in the local lightweight face database is greater than or equal to the distance threshold, it is determined that the face to be recognized is a stranger.

[0076] In this embodiment, the local lightweight face database adopts a structured storage method. Each record contains a unique face ID (i.e., a unique temporary identity identifier), a global face visual feature vector, a multi-dimensional face attribute feature vector, and the entry time. As shown in Table 1, a local lightweight face database with 12 fields is initialized. Among them, the unique face ID (id) is a unique index key, used by the legged humanoid robot to achieve short-term, anonymous, and collision-free real-time interaction. For example, if the same person enters the field of vision multiple times within 5 minutes and is identified as the same person through ID consistency, the human-computer interaction can be skipped, avoiding repeated greetings, repeated reminders to wear masks, and other repetitive interaction operations. Moreover, only the unique face ID is generated as a unique temporary identity identifier, without involving real identity information, reducing the risk of privacy leakage from the source. For ease of demonstration, Table 1 only uses two example data entries with id=1 and id=2.

[0077] Table 1.

[0078]

[0079] in, Used to record the global facial visual feature vector of a temporary reference face. The temporary reference face is used to record the entry time. Other fields are used to record the age, gender, skin color, expression, eye state, mouth state, head posture, whether glasses are worn, whether masks are worn, and other multi-dimensional facial attribute feature vectors of the temporary reference face, so as to facilitate subsequent identification of unfamiliar faces and short-term data management and periodic cleanup.

[0080] In this embodiment, the face identification module 140 identifies the face feature vector of the face to be identified within the face bounding box. By comparing the facial feature vectors with those of various temporary baseline faces dynamically stored in a local lightweight face database, stranger identification and identity matching can be achieved. In practical implementation, feature storage, retrieval, and similarity calculation are implemented based on SQL technology, specifically including:

[0081] The facial feature vector of the face to be identified within the bounding box of the i-th face. The facial feature vectors of all temporary baseline faces in the local lightweight face database were obtained through SQL queries. Calculate the Euclidean distances one by one:

[0082]

[0083] in, For Euclidean distance, Let be the facial feature vector of the face to be identified within the bounding box of the i-th face. is the face feature vector of the jth temporary reference face in the local lightweight face database, k is the feature dimension index, k = 1, 2, 3, …, d (d = 128 or 256).

[0084] If the Euclidean distance between the face feature vector of the jth temporary reference face and the face feature vector of the to-be-identified face in the ith face bounding box satisfies: , then it is determined that the to-be-identified face in the ith face bounding box is the same identity as the temporary reference face with the unique face number (that is, the unique temporary identity) of in the local lightweight face database; if all Euclidean distances are greater than the distance threshold θ, it is determined that the to-be-identified face in the ith face bounding box is a stranger face.

[0085] In a possible implementation, the stranger face recognition system 100 provided by the embodiment of the present application further includes:

[0086] The data management module 150 is connected with the face identity determination module 140, and is configured to, for the to-be-identified face in each face bounding box, if the to-be-identified face is the same face as any temporary reference face in the local lightweight face database, update the entry time of any temporary reference face to the current time, if the to-be-identified face is a stranger face, assign a unique temporary identity to the stranger face, store the stranger face, the face feature vector and the unique temporary identity of the stranger face as a temporary reference face to the local lightweight face database, and take the current time as the entry time; automatically delete the unique temporary identity and the face feature vector of the temporary reference face in the local lightweight face database whose entry duration exceeds the storage duration threshold.

[0087] In the embodiment of the present application, in the case that the to-be-identified face matches any temporary reference face in the local lightweight face database, the entry time of the any temporary reference face is updated to the current time based on the unique face number (that is, the unique temporary identity) of the any temporary reference face. This process ensures that the local lightweight face database can always reflect the latest identification event, thereby realizing dynamic management based on time. Specifically, the entry time updating operation is represented as:

[0088]

[0089] , wherein, represents the entry time, represents the unique face number (that is, the unique temporary identity) of the any temporary reference face, represents the current time (system time).

[0090] ​For example: in a local lightweight face database The temporary baseline face was first entered on 2025-09-04 at 10:24:55. When the temporary baseline face is detected and recognized again on 2025-09-04 at 10:30:10, only record_time=2025-09-04 10:30:10 is updated, without adding a new temporary baseline face. Thus, by updating the entry time, the consistency and timeliness of the local lightweight face database can be maintained, providing a reliable basis for subsequent short-term management, activity statistics or trajectory tracking, while also reducing storage and computing pressure.

[0091] In this embodiment, when the face to be identified fails to match any of the temporary reference faces in the local lightweight face database, that is, when the Euclidean distance between the face to be identified and any of the temporary reference faces in the local lightweight face database is greater than a distance threshold... At this point, the face to be identified is determined to be an unfamiliar face. A unique face ID is then assigned to this unfamiliar face as a unique temporary identifier. This unique face ID is generated in an auto-incrementing sequence. When the unique face ID reaches its upper limit (e.g., 999), it restarts from 0 and is numbered again using slot detection. For example, if the recording time of the temporary baseline face with unique face ID 0 is still within the active period of the storage time threshold (e.g., 5 minutes), then 0 is skipped and the auto-increment continues until a unique face ID that has not been within the active period of the storage time threshold (e.g., 5 minutes) and has been automatically deleted is found as the unique temporary identifier for the newly recorded unfamiliar face. The unfamiliar face, its facial feature vector, recording time, and unique temporary identifier are then stored in a local lightweight face database. The storage of unfamiliar face data includes: a global facial visual feature vector, multi-dimensional facial attribute feature vectors (e.g., age, gender, skin color, expression, eye state, mouth state, head posture, whether glasses are worn, whether a mask is worn, etc.), a unique temporary identifier, and the recording time. For example, if an unfamiliar face is detected at 11:02:33 on 2025-09-04 and a unique temporary identifier id=3 is assigned, then a temporary baseline face and its related data are added to the local lightweight face database as shown in Table 2:

[0092] Table 2.

[0093]

[0094] In this way, by introducing the timing deletion and empty slot detection self-increment mechanism, the number of temporary reference faces in the local lightweight face database can be kept within the set number range (such as 1000) at all times, and the numbering cycle will not cause conflicts, so that only active and recently detected stranger face data is saved in the local lightweight face database as temporary reference face data, thereby realizing real-time entry and effective management of temporary reference face data under the condition of limited storage space, and ensuring space controllability and high real-time performance.

[0095] In the embodiments of the present application, the timing deletion strategy based on entry time is to directly complete the expired data screening and removal at the database level through SQL query, thereby significantly reducing the system burden. Specifically, the following operations are periodically performed, a storage time threshold is set, and any temporary reference face and its global face visual feature vector, multi-dimensional face attribute feature vector, and unique temporary identity data whose storage time exceeds the storage time threshold are determined to be expired and automatically removed. Specifically, the expired determination condition is expressed as:

[0096] record_time(id)< τ

[0097] wherein, record_time represents the entry time, id represents the unique face number (i.e., the unique temporary identity) of the temporary reference face, current_time represents the current time, and τ represents the storage time threshold.

[0098] For example, if the current time is =2025-09-04 11:10:00, then all temporary reference faces and their global face visual feature vectors, multi-dimensional face attribute feature vectors, and unique temporary identity data whose record_time is earlier than 2025-09-04 11:05:00 will be deleted at once, thereby avoiding the redundant overhead of calculating the time difference one by one and improving the running efficiency.

[0099] In a possible implementation, the data management module 150 is also connected to the face feature extraction module 130, and is configured to count the number of faces to be recognized based on the number of face bounding boxes in the set of face bounding boxes corresponding to the target environment image recognized by the multi-task face recognition model; based on the crowd density coefficients corresponding to different crowd density intervals, determine the crowd density coefficient corresponding to the crowd density interval where the number of faces to be recognized is located as the current crowd density coefficient, where the higher the crowd density interval, the greater the crowd density coefficient; adjust the storage duration threshold based on the current crowd density coefficient; where the storage duration threshold τ_base is adjusted to τ_dynamic = τ_base × ρ, and ρ is the current crowd density coefficient.

[0100] For example: Define the crowd density coefficient ρ. Low crowd density interval: The number of faces to be recognized < the first density threshold (e.g., 5), and the corresponding crowd density coefficient is the first value (e.g., ρ = 0.8); Medium crowd density interval: The first density threshold (e.g., 5) ≤ the number of faces to be recognized ≤ the second density threshold (e.g., 15), and the corresponding crowd density coefficient is the second value (e.g., ρ = 1.0); High crowd density interval: The number of faces to be recognized > the second density threshold (e.g., 15), and the corresponding crowd density coefficient is the second value (e.g., ρ = 1.2). If τ_base is the storage duration threshold = 5 min, then in the low crowd density interval: τ_dynamic = 4 min (shorten the storage time and reduce redundancy); in the medium crowd density interval: τ_dynamic = 5 min (maintain the original benchmark and balance efficiency); in the high crowd density interval: τ_dynamic = 6 min (lengthen the storage time to avoid premature deletion of effective interaction data, such as when the data is cleared before the robot completes the guiding service).

[0101] In a possible implementation, the data management module 150 is configured to, when performing timed deletion, call the real-time calculated τ_dynamic, and update the expiration determination condition to record_time(id) < T_now - τ_dynamic; at the same time, when the crowd density switches from high → medium → low, automatically trigger a screening of redundant data and delete the expired data that exceeds the current τ_dynamic.

[0102] Based on the above embodiments, the embodiments of the present application provide a method for recognizing strange faces, which is applied to a legged humanoid robot. Refer to Figure 2 As shown, the general process of the method for recognizing strange faces provided by the embodiments of the present application is as follows:

[0103] Step 200: Collect the environmental video stream in the public environment.

[0104] Step 201: Use a lightweight quality repair model to perform quality repair on the environmental images in the environmental video stream to obtain the target environmental images.

[0105] Step 202: Using a multi-task face recognition model, face boundary box prediction and face key point extraction are performed on the target environment image to obtain a face boundary box set and a face key point set in each face boundary box.

[0106] Step 203: Based on the standard template key point set, the face key point set in each face boundary box is aligned to the standard template face image to obtain a non-distorted face image corresponding to each face boundary box.

[0107] Step 204: Using a multi-task feature extraction model, a global face visual feature vector and a multi-dimensional face attribute feature vector of the face to be recognized are extracted from the non-distorted face image corresponding to each face boundary box as a face feature vector.

[0108] Step 205: The similarity between the face feature vector of the face to be recognized in each face boundary box and the face feature vector of each temporary reference face dynamically stored in the local lightweight face database is calculated.

[0109] Step 206: Based on the similarity, it is determined whether the face to be recognized in each face boundary box is a stranger face; wherein the face feature vector of each temporary reference face dynamically stored in the local lightweight face database is the face feature vector of each stranger face recognized within a certain time range.

[0110] Step 207: For the face to be recognized in each face boundary box, if the face to be recognized is determined to be a stranger face, a unique temporary identity is assigned to the stranger face, the stranger face is stored in the local lightweight face database as a temporary reference face together with its face feature vector and unique temporary identity, and the current time is recorded as the entry time; if the face to be recognized is determined to be the same person as any temporary reference face in the local lightweight face database, the entry time of any temporary reference face is updated to the current time.

[0111] Step 208: The unique temporary identity and face feature vector of the temporary reference face in the local lightweight face database whose entry duration exceeds the storage duration threshold are automatically deleted.

[0112] In the embodiment of the present application, by deploying a lightweight quality repair model, a multi-task face recognition model, a multi-task feature extraction model and a local lightweight face database on the end side of the legged humanoid robot, the image quality repair of the lightweight quality repair model can be used to solve the problem of low-quality conditions such as overexposure or underexposure and blur of environmental images caused by motion blur due to complex public environmental light and walking of the legged humanoid robot, and ensure the reliability of subsequent face recognition. In addition, the face boundary box recognition of the multi-task face recognition model, the face feature extraction of the multi-task feature extraction model and the stranger face discrimination of the local lightweight face database can be used to realize stranger face recognition on the end side of the legged humanoid robot, avoid network delay introduced by cloud comparison, and thus realize real-time interaction response in a public environment, improve the understanding and service ability of the legged humanoid robot for interactive objects, and significantly improve the human-computer interaction experience in public places.

[0113] After introducing the stranger face recognition system and method provided by the embodiment of the present application, the legged humanoid robot provided by the embodiment of the present application is briefly introduced.

[0114] Referring to Figure 3 As shown in the figure, the legged humanoid robot 300 provided by the embodiment of the present application at least includes a processor 301, a memory 302 and a computer program stored in the memory 302 and executable on the processor 301, and the processor 301 implements the stranger face recognition method provided by the embodiment of the present application when executing the computer program.

[0115] The legged humanoid robot 300 provided by the embodiment of the present application can further include a bus 303 connecting different components (including the processor 301 and the memory 302). Among them, the bus 303 represents one or more of several types of bus structures, including a memory bus, a peripheral bus, a local bus, etc.

[0116] The memory 302 can include a readable storage medium in the form of a volatile memory, such as a random access memory (RAM) 3021 and / or a cache memory 3022, and can further include a read-only memory (ROM) 3023. The memory 302 can also include a program tool 3025 having a set of (at least one) program modules 3024, including but not limited to an operating system, one or more application programs, other program modules, and program data, each of these examples or some combination thereof can include the implementation of a network environment.

[0117] The processor 301 can be one processing element or a collective term for multiple processing elements. For example, the processor 301 can be a central processing unit (CPU), or one or more integrated circuits configured to implement the method of recognizing a stranger face provided by the embodiments of the present application. Specifically, the processor 301 can be a general-purpose processor, including but not limited to a CPU, an application specific integrated circuit (ASIC), a field programmable gate array (FPGA) or other programmable logic device, a discrete gate or transistor logic device, a discrete hardware component, etc.

[0118] The biped robot 300 can communicate with one or more external devices 304 (such as a keyboard, a remote controller, etc.), and can also communicate with one or more devices (such as a mobile phone, a computer, etc.) that enable a user to interact with the biped robot 300, and / or communicate with a device (such as a router, a modem, etc.) that enables the biped robot 300 to communicate with one or more other biped robots 300. Such communication can be performed through an input / output (I / O) interface 305. In addition, the biped robot 300 can also communicate with one or more networks (such as a local area network (LAN), a wide area network (WAN), and / or a public network such as the Internet) through a network adapter 306. As shown, the network adapter 306 communicates with other modules of the biped robot 300 through the bus 303. It should be understood that although the network adapter 306 is shown as a separate component from the processor 301 in FIG. 3, the network adapter 306 can be an integral part of the processor 301 in some embodiments. Figure 3 As shown, the network adapter 306 communicates with other modules of the biped robot 300 through the bus 303. It should be understood that although the network adapter 306 is shown as a separate component from the processor 301 in FIG. 3, the network adapter 306 can be an integral part of the processor 301 in some embodiments. Figure 3 It should be understood that although other hardware and / or software modules are not shown in FIG. 3, they can be used in conjunction with the biped robot 300, including but not limited to microcode, device drivers, redundant processors, external disk drive arrays, redundant arrays of independent disks (RAID) subsystems, tape drives, and data backup storage subsystems, etc.

[0119] It should be noted that the biped robot 300 shown in FIG. 3 is merely an example, and should not impose any limitation on the functions and use range of the embodiments of the present application. Figure 3 It should be noted that the biped robot 300 shown in FIG. 3 is merely an example, and should not impose any limitation on the functions and use range of the embodiments of the present application.

[0120] The computer readable storage medium provided by the embodiments of the present application is described as follows. The computer readable storage medium provided by the embodiments of the present application stores computer instructions, and the computer instructions are executed by a processor to implement the stranger face recognition method provided by the embodiments of the present application. Specifically, the computer instructions can be built-in or installed in the processor, so that the processor can implement the stranger face recognition method provided by the embodiments of the present application by executing the built-in or installed computer instructions.

[0121] In addition, the stranger face recognition method provided by the embodiments of the present application can also be implemented as a computer program product, which includes program codes. When the program codes are run on a processor, the stranger face recognition method provided by the embodiments of the present application is implemented.

[0122] The computer program product provided by the embodiments of the present application can adopt one or more computer readable storage media, and the computer readable storage media can be, but are not limited to, an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, device or component, or any appropriate combination of the above. Specifically, more specific examples (non-exhaustive list) of the computer readable storage media include an electrical connection with one or more wires, a portable disc, a hard disk, a RAM, a ROM, an Erasable Programmable Read Only Memory (EPROM), an optical fiber, a portable Compact Disc Read-Only Memory (CD-ROM), an optical storage device, a magnetic storage device, or any appropriate combination of the above.

[0123] The computer program product provided by the embodiments of the present application can adopt a CD-ROM and include program codes, and can also run on a biped robot. However, the computer program product provided by the embodiments of the present application is not limited to this. In the embodiments of the present application, the computer readable storage medium can be any tangible medium containing or storing program codes, which can be used by or in combination with an instruction execution system, device or component.

[0124] It should be noted that although several units or sub-units of the apparatus are mentioned in the above detailed description, such division is only exemplary and not mandatory. In fact, according to the embodiments of the present application, the features and functions of two or more units described above can be embodied in one unit. Conversely, the features and functions of one unit described above can be further divided into units embodied by multiple units.

[0125] Furthermore, although the operations of the method(s) herein can be described in a particular, sequential order, this order is not meant to be a limitation and one or more of the operations described can be performed in parallel, or in a different order, including before or after other operations described. The various steps described can be implemented in hardware, software, or a combination thereof. The subject specification can be implemented by computer software implemented by one or more processors of a computing device.

[0126] While the preferred embodiments of the application have been described above, it should be understood that they have been presented by way of example only, and not limitation. Numerous changes to the embodiments can be made in addition to those described and nevertheless accomplish the same objectives of the application. Thus, while the application is susceptible to various modifications and alternative forms, specific embodiments have been shown by way of example in order to elucidate the application. It should be understood, therefore, that the application is not to be limited to the particular embodiments described but it is intended to cover any and all modifications and equivalents within the scope of the appended claims.

[0127] Obviously, numerous modifications and variations of the present application are possible in light of the above teachings. It is therefore to be understood that within the scope of the appended claims, the application can be practiced otherwise than as specifically described herein.

Claims

1. A stranger face recognition system, characterized by, The robot is deployed on a legged humanoid robot, and includes a video stream acquisition module, an image quality optimization module, a face feature extraction module, and a face identity determination module. The video stream acquisition module is integrated on the body of the legged humanoid robot and is configured to acquire an environment video stream in a public environment. The image quality optimization module is connected to the video stream acquisition module and is configured to use a lightweight quality repair model to perform quality repair on environment images in the environment video stream to obtain target environment images. The face feature extraction module is connected to the image quality optimization module and is configured to use a multi-task face recognition model to perform face bounding box prediction and face key point extraction on the target environment images to obtain a face bounding box set and a face key point set in each face bounding box in the face bounding box set. Based on a standard template key point set, the face key point set in each face bounding box is aligned to a standard template face image to obtain a non-distorted face image corresponding to each face bounding box. A multi-task feature extraction model is used to extract a global face visual feature vector and a multi-dimensional face attribute feature vector of a face to be recognized from each non-distorted face image as a face feature vector. The multi-task face recognition model includes a first feature map extraction branch and a three-level detection head connected in parallel to the first feature map extraction branch. The three-level detection head includes a face binary classification detection branch, a face bounding box regression branch, and a face key point regression branch. The multi-task feature extraction model includes a second feature map extraction branch and a two-level detection head connected in parallel to the second feature map extraction branch. The two-level detection head includes a global visual feature extraction branch and a multi-attribute feature extraction branch. The face identity determination module is connected to the face feature extraction module and is configured to calculate the similarity between the face feature vector of a face to be recognized in each face bounding box and the face feature vector of each temporary reference face dynamically stored in a local lightweight face database, and determine whether the face to be recognized in each face bounding box is a stranger based on the similarity. The face feature vector of each temporary reference face dynamically stored in the local lightweight face database is the face feature vector of each stranger recognized within a specified time range.

2. The stranger face recognition system of claim 1, wherein, The image quality optimization module is configured to calculate the image quality score of each environment image in the environment video stream using an image quality evaluation function. When the image quality score is less than a first quality threshold, the environment image is discarded. When the image quality score is greater than or equal to the first quality threshold and less than a second quality threshold, the lightweight quality repair model is used to repair the motion blur, image noise, and illumination abnormalities of the environment image to obtain the target environment image. When the image quality score is greater than or equal to the second quality threshold, the environment image is determined as the target environment image.

3. The stranger face recognition system of claim 1, wherein, The face feature extraction module is configured to: after extracting first shared semantic feature maps from the target environment image by the first feature map extraction branch in the multi-task face recognition model, performing face two-classification detection, face bounding box detection, and face key point detection based on the first shared semantic feature maps in parallel by the face two-classification detection branch, the face bounding box regression branch, and the face key point regression branch in the multi-task face recognition model to obtain face two-classification detection results, face bounding box detection results, and face key point detection results; and performing post-processing on the face two-classification detection results, the face bounding box detection results, and the face key point detection results to obtain a face bounding box set corresponding to the target environment image and a face key point set in each face bounding box in the face bounding box set.

4. The stranger face recognition system of claim 1, wherein, The face feature extraction module is configured to: for each face bounding box, calculate an affine transformation matrix between the face key point set in the face bounding box and the standard template key point set, and align the face key point set in the face bounding box to a calibration template face image based on the affine transformation matrix to obtain a non-distortion face image corresponding to the face bounding box.

5. The stranger face recognition system of claim 1, wherein, The face feature extraction module is configured to: after extracting second shared semantic feature maps from each non-distortion face image corresponding to the face bounding box by the second feature map extraction branch in the multi-task feature extraction model, performing global visual feature extraction and multi-attribute feature extraction based on the second shared semantic feature maps corresponding to each face bounding box in parallel by the global visual feature extraction branch and the multi-attribute feature extraction branch in the multi-task feature extraction model to obtain a global face visual feature vector and a multi-dimensional face attribute feature vector corresponding to each face bounding box; and performing post-processing on the global face visual feature vector and the multi-dimensional face attribute feature vector corresponding to each face bounding box to obtain a face feature vector corresponding to each face bounding box.

6. The stranger face recognition system of claim 1, wherein, The face identity determination module is configured to: calculate the Euclidean distance between the face feature vector of each face bounding box and the face feature vector of each temporary reference face dynamically stored in the local lightweight face database by using a similarity calculation method based on the Euclidean distance; and for each face bounding box, if the Euclidean distance between the face feature vector of the face bounding box and the face feature vector of any temporary reference face dynamically stored in the local lightweight face database is less than a distance threshold, it is determined that the face bounding box and the temporary reference face are the same person, and if the Euclidean distance between the face feature vector of the face bounding box and the face feature vector of all temporary reference faces dynamically stored in the local lightweight face database is greater than or equal to the distance threshold, it is determined that the face bounding box is a stranger.

7. The stranger face recognition system of any one of claims 1-5, wherein, Further comprising: The data management module is connected with the face identity determination module, and is configured to, for each to-be-identified face in the face bounding box, if the to-be-identified face is the same person as any temporary reference face in the local lightweight face database, update the recording time of the any temporary reference face to the current time, if the to-be-identified face is a stranger face, assign a unique temporary identity to the stranger face, store the stranger face as a temporary reference face together with the face feature vector and the unique temporary identity of the stranger face to the local lightweight face database, and take the current time as the recording time; and automatically delete the unique temporary identity and the face feature vector of each temporary reference face in the local lightweight face database whose recording duration exceeds a storage duration threshold.

8. A method for recognizing faces of strangers, characterized in that, The application is applied to a legged humanoid robot, comprising: Collecting an environment video stream in a public environment; Using a lightweight quality repair model to perform quality repair on an environment image in the environment video stream to obtain a target environment image; Using a multi-task face recognition model to perform face bounding box prediction and face key point extraction on the target environment image to obtain a face bounding box set and a face key point set in each face bounding box in the face bounding box set; wherein the multi-task face recognition model comprises a first feature map extraction branch and a three-level detection head connected in parallel with the first feature map extraction branch; the three-level detection head is a face two-classification detection branch, a face bounding box regression branch and a face key point regression branch; Based on a standard template key point set, aligning the face key point set in each face bounding box to a standard template face image to obtain a non-distortion face image corresponding to each face bounding box; Using a multi-task feature extraction model to extract a global face visual feature vector and a multi-dimensional face attribute feature vector of a to-be-identified face from the non-distortion face image corresponding to each face bounding box as a face feature vector; wherein the multi-task feature extraction model comprises a second feature map extraction branch and a two-level detection head connected in parallel with the second feature map extraction branch; the two-level detection head is a global visual feature extraction branch and a multi-attribute feature extraction branch; Calculating the similarity between the face feature vector of each to-be-identified face in each face bounding box and the face feature vector of each temporary reference face dynamically stored in a local lightweight face database; Based on the similarity, determining whether the to-be-identified face in each face bounding box is a stranger face; wherein the face feature vector of each temporary reference face dynamically stored in the local lightweight face database is the face feature vector of each stranger face identified within a recent set time range.

9. A legged humanoid robot, characterized by, The application includes a memory, a processor and a computer program stored on the memory and executable on the processor, and the processor implements the stranger face recognition method of claim 8 when executing the computer program.

10. A computer-readable storage medium, characterized in that, The computer readable storage medium stores computer instructions, and the computer instructions are executed by the processor to implement the stranger face recognition method in claim 8.

Images