A biometric feature collection system

Abstract

The present application provides a biometric feature collection system, including a feature collection module; a time management module, which allows the user to set one or more time windows, and the time window is the time period when the user intends to use the smart door lock when returning home; an intelligent management module, which is used to control the working state of the feature collection module; a main wake-up module, which is used to decide whether to activate the facial feature collection unit within the set time window; and an auxiliary wake-up module, which is used to decide whether to activate the facial feature collection unit outside the set time window. The present application configures a time window in the smart door lock system. During the time period within the time window, the collection system of the smart door lock can respond quickly to meet the needs of the user and ensure the convenience of the door lock. During the time period outside the time window, the false touch rate of the smart door lock feature collection module by non-user mobile personnel is reduced, and unnecessary energy consumption of the smart door lock is avoided.

Description

Technical Field

[0001] The present invention relates to the field of smart door lock feature collection, and in particular to a biometric feature collection system. Background Art

[0002] With the advancement of technology, various smart door locks based on face recognition technology have appeared on the market. As an important part of smart home, smart door locks are increasingly favored by users. This type of door lock uses advanced technologies, such as sensors, high-definition cameras, etc., to provide users with a more convenient and secure way to manage access. One of the core functions of smart door locks is that when someone stands in a specific collection area in front of the door lock, it can automatically sense and trigger the door lock's working machine, and its built-in high-definition camera begins to capture the target's biometric features for comparison with the registration information in the system database, thereby confirming the user's identity and deciding whether to unlock the door. This automated recognition process greatly improves the convenience and safety of door locks.

[0003] However, with the widespread application of smart door locks, some challenges encountered in actual use have gradually emerged. In actual applications, the collection areas of smart door locks are mostly non-enclosed, so there is a risk of non-user members entering the collection area by mistake. When non-user members accidentally enter the collection area, the sensor will also send a signal to trigger the door lock, causing the high-definition camera to start invalid collection work. In this case, the door lock will continue to start and shut down due to frequent false triggers, which will not only increase the energy consumption of the door lock system and shorten the battery life, but also increase the cost and frequency of battery replacement for users. What is more serious is that frequent system startup and shutdown may also cause certain damage to the door lock hardware, reducing the overall performance and life of the smart door lock. If the sensor's sensing mode is turned off, the user needs to actively wake it up, which reduces the user experience.

[0004] Therefore, we make improvements to this and propose a biometric collection system. Summary of the invention

[0005] The purpose of the present invention is to address the problem that when the smart sensing function of existing smart door locks is turned on, false triggering by non-user members will cause frequent false opening and closing of face recognition, increase the energy consumption of the door lock system, and reduce the performance and life of the smart door lock. When the smart sensing function is turned off, the user needs to actively wake it up, which reduces the user experience.

[0006] In order to achieve the above-mentioned purpose of the invention, the present invention provides the following biometric feature collection system to improve the above-mentioned problem.

[0007] The specific application is as follows:

[0008] A biometric feature collection system, comprising:

[0009] A feature acquisition module, comprising a facial feature acquisition unit and a facial feature extraction unit, wherein the facial feature acquisition unit is used to acquire facial image information of a living object standing in front of the lock body, and the facial feature extraction unit is used to extract feature points of the facial image information;

[0010] The time management module is used to provide real-time time information and allow the user to set one or more time windows, which are the time periods when the user intends to use the smart door lock when returning home;

[0011] The intelligent management module is used to control the working state of the feature collection module, control the feature collection module to be in an active state within the time window, and control the feature collection module to be in a dormant state outside the time window;

[0012] The main wake-up module is used to decide whether to activate the facial feature acquisition unit within a set time window;

[0013] The auxiliary wake-up module is used to decide whether to activate the facial feature collection unit outside the set time window.

[0014] As a preferred technical solution of the present application, it also includes a data storage module for storing the user's biometric data, wherein the biometric data includes the user's facial feature samples and voice information samples.

[0015] As the preferred technical solution of the present application, the main wake-up module includes an infrared sensing unit and a timing unit. The infrared sensing unit senses whether a living body approaches the smart door lock; the timing unit is used to time the residence time of the sensed living object. When the residence time exceeds the threshold, the facial feature acquisition unit is immediately started to collect facial image information of the living object.

[0016] As a preferred technical solution of the present application, the timing unit includes a time threshold T1 within the time window and a time threshold T2 outside the time window, and T1 is less than T2.

[0017] As a preferred technical solution of the present application, the auxiliary wake-up module includes:

[0018] A voice collection unit is used to collect voice information of a living body when the infrared sensing unit senses that a living body is approaching outside the time window;

[0019] A preliminary recognition unit is used to perform preliminary recognition on the collected voice information to verify whether its text content is consistent with the stored voice information sample;

[0020] The second recognition unit is used to perform a deeper recognition on the voice information that is consistent with the content of the voice information sample, and calculate the voiceprint similarity between it and the voice information sample.

[0021] As a preferred technical solution of the present application, the steps of waking up facial recognition in the time period outside the time window are as follows:

[0022] S101, the infrared sensing unit senses whether there is a living body approaching the smart door lock;

[0023] S102: If a living body is detected, start timing and synchronously collect voice information of the living body;

[0024] S103. When the living body residence time exceeds the threshold T2 or the preliminary recognition unit confirms that the collected voice content is consistent with the sample, the facial feature collection unit is immediately activated.

[0025] As a preferred technical solution of the present application, it also includes a feature extraction decision module, which dynamically adjusts the feature extraction strategy of the facial feature extraction unit according to the result of the second recognition unit, and its working steps are as follows:

[0026] S201, obtaining a speech similarity value calculated by a second recognition unit, and comparing it with a preset similarity threshold;

[0027] S202, if the similarity value is higher than the threshold, the facial feature extraction unit will focus on extracting key features in the facial image of the living object;

[0028] S203: If the similarity value is less than or equal to the threshold, the facial feature extraction unit extracts key features and detailed features in the facial image.

[0029] As a preferred technical solution of the present application, the facial feature extraction unit completes step S202 or S203 by constructing a facial feature extraction model based on a deep learning algorithm. The method for constructing the facial feature extraction model is as follows:

[0030] S301, data collection: collecting face image data in different environments;

[0031] S302, data preprocessing: preprocessing the collected face images, including image cleaning, cropping, scaling and normalization;

[0032] S303, data labeling: labeling the pre-processed face image, marking the position, size, key points and other information of the face, which is used to provide supervision signals when training the model;

[0033] S304, model construction: construct a dual-branch (parallel) facial feature extraction model through a convolutional neural network algorithm;

[0034] S305, model training: dividing the collected face image data into two training sets (a training set, a validation set, and a test set), the two training sets are used for training, validation, and evaluation of two branches of the facial feature extraction model respectively;

[0035] S306, model use: deploying the facial feature extraction model that has passed the evaluation into the facial feature extraction unit.

[0036] As a preferred technical solution of the present application, the structure of the facial feature extraction model includes:

[0037] A shared underlying network, used for preliminarily extracting features of a face image, comprising an input layer, at least one convolutional layer, and a corresponding maximum pooling layer, wherein the input layer receives an original face image of a specific size;

[0038] A first feature extraction branch, connected to the shared underlying network, responsive to the recognition result of the second recognition unit, for detecting key features of the face, comprising at least one convolutional layer, a global average pooling layer, a fully connected layer and an output layer, wherein the output layer is configured to output a feature vector encoding the key features of the face;

[0039] The second feature extraction branch is connected to the shared underlying network and responds to the recognition result of the second recognition unit to extract key features and detailed features of the face. The second feature extraction branch includes at least two convolutional layers, a global average pooling layer, a fully connected layer and an output layer, and the output layer is configured to output a feature vector encoding the key features and detailed features of the face.

[0040] As a preferred technical solution of the present application, the number of convolution kernels of the convolution layer in the first feature extraction branch is greater than the number of convolution kernels of the convolution layer in the shared underlying network; the number of convolution kernels of the convolution layer in the second feature extraction branch is greater than the number of convolution kernels of the convolution layer in the first feature extraction branch.

[0041] Compared with the prior art, the present invention has the following beneficial effects:

[0042] In the scheme of this application:

[0043] By configuring a time window in the smart door lock system, the time window is the time period when the user intends to use the smart door lock when returning home, that is, the time period within the time window, the smart door lock acquisition system can respond quickly without contact, meet the needs of users, and ensure the convenience of using the door lock. In the time period outside the time window, the conditions for triggering the smart door lock face recognition unit are added. The system is relatively more conservative, ensuring that the smart door lock feature acquisition module is triggered only after there is a clear intention to unlock the door, reducing the false touch rate of the smart door lock feature acquisition module by non-user mobile personnel, and avoiding unnecessary energy consumption of the smart door lock. At the same time, through the cooperation of the set main wake-up module and the auxiliary wake-up module, the user's needs for contactless wake-up face recognition unlocking in the time period outside the time window are met, providing user convenience. BRIEF DESCRIPTION OF THE DRAWINGS

[0044] Figure 1 System block diagram of the biometric feature collection system provided for this application;

[0045] Figure 2 A schematic diagram of the process of starting the facial feature collection unit outside the time window of the biometric feature collection system provided by the present application;

[0046] Figure 3 A schematic diagram of the operation flow of the external feature extraction unit outside the time window of the biometric feature collection system provided in this application. DETAILED DESCRIPTION

[0047] In order to enable those skilled in the art to better understand the scheme of the present invention, the technical scheme in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work should fall within the scope of protection of the present invention.

[0048] As described in the background technology, when the smart sensing function is turned on, most smart door locks in the prior art will be mistakenly triggered by non-user members, which will cause frequent false opening and closing of face recognition, increase the energy consumption of the door lock system, and reduce the performance and life of the smart door lock. When the smart sensing function is turned off, the user needs to actively wake it up, which reduces the user experience.

[0049] In order to solve this technical problem, the present invention provides a biometric feature collection system, which is mainly designed for the feature collection stage of face recognition of smart door locks.

[0050] Specifically, please refer to Figure 1 , the biometric collection system specifically includes:

[0051] The feature acquisition module includes a facial feature acquisition unit and a facial feature extraction unit. The facial feature acquisition unit is used to collect facial image information of a living object standing in front of the lock body. Specifically, the facial image of the user is captured by a camera configured on the smart door lock. The facial feature extraction unit is used to extract feature points of the facial image information for subsequent feature verification and comparison in the unlocking link of the smart door lock.

[0052] The time management module is used to provide real-time time information and allow users to set one or more time windows. The time window is the time period when the user intends to use the smart door lock when returning home. The user sets it according to his or her own habits.

[0053] The intelligent management module is used to control the working state of the feature collection module. Specifically, the feature collection module is controlled to be in an active state within the time window and to be in a dormant state outside the time window. Since the time window is set subjectively by the user and the time period for returning home is estimated, the feature collection module can be prevented from being in an active state all the time, thereby reducing the probability of non-user mobile personnel mistakenly entering the sensing range of the feature collection module, resulting in the feature collection module being frequently activated, thereby avoiding unnecessary energy consumption.

[0054] Correspondingly, two non-contact wake-up feature collection module operation modes are provided. One is the main wake-up module, which is used to decide whether to activate the facial feature collection unit within the set time window; the other is the auxiliary wake-up module, which is used to decide whether to activate the facial feature collection unit outside the set time window. The wake-up feature collection module is divided into two categories according to the time window, providing users with the convenience of contactless awakening of smart door lock face recognition by users.

[0055] It also includes a data storage module for storing the user's biometric data. The biometric data includes the user's facial feature samples and voice information samples. The facial feature samples are used for comparison and authentication of smart door lock unlocking, and the voice information samples are used for comparison and authentication of the auxiliary wake-up module.

[0056] In order to enable those skilled in the art to better understand the solutions of the present invention, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the accompanying drawings.

[0057] It should be noted that, in the absence of conflict, the embodiments of the present invention and the features and technical solutions in the embodiments may be combined with each other.

[0058] Example 1, please refer to Figure 1 to Figure 2The contactless wake-up feature collection module provided in the present application operates in a main wake-up module, and the main wake-up module includes an infrared sensing unit and a timing unit. The infrared sensing unit senses whether a living body approaches the smart door lock. Preferably, it can be understood as an infrared sensor, and its sensing range is set to 0.5-1 meters in front of the door lock to ensure accurate sensing of the approach of living bodies; the timing unit is used to time the residence time of the living object, and when the residence time exceeds the threshold, the facial feature collection unit is immediately started to collect the facial image information of the living object; the timing unit includes a time threshold within the time window as T1, and a time threshold outside the time window as T2, and T1 is less than T2.

[0059] Preferably, the time threshold T1 of the timing unit is preferably set to 1 second within the time window to facilitate quick response and ensure user experience. Outside the time window, the sensing time threshold T2 can preferably be set to 3 or 4 seconds to reduce the possibility of false triggering. That is, when the user enters the sensing range of the infrared sensing unit and the stay time exceeds the time threshold T2, it can be determined that the user intends to unlock the door, and the feature acquisition module can be awakened to run. Correspondingly, if the stay time is less than the time threshold T2, the target can be determined to have accidentally entered the sensing range, and the feature acquisition module will not respond, thereby reducing the false trigger rate of the feature acquisition module and avoiding unnecessary energy consumption.

[0060] Since a time window is set, it is inevitable that the user returns home during a time period outside the time window during actual use. For this reason, the present application also provides a contactless wake-up solution, including an auxiliary wake-up module. Specifically, the auxiliary wake-up module includes a voice collection unit, which is used to collect voice information of a living body when the infrared sensing unit senses the approach of a living body outside the time window; and a preliminary recognition unit, which is used to perform preliminary recognition on the collected voice information to verify whether its text content is consistent with the content of the stored voice information sample. The voice information sample is a specific wake-up word or wake-up sentence entered by the user in the initial stage.

[0061] In a specific usage scenario, when the user returns home outside the time window, in order to improve the convenience of using the smart door lock, when the infrared sensing unit senses the user, the user can send a pre-stored voice message. When the content is consistent with the stored voice message, the feature acquisition module is awakened to run and start collecting the user's facial image information; providing a convenient wake-up operation method for the user if the throat is intact; correspondingly, when the user has a throat problem or other discomforts, the user can also wait for a while and wake up the feature acquisition module through the main wake-up module.

[0062] Furthermore, the voice collection unit is triggered only when it senses a living body through the infrared sensing unit, avoiding frequent or erroneous collection by the voice collection unit due to environmental noise or voices that are far away from the door lock but similar to the voice information, further reducing the possibility of false triggering of the feature collection module, which is conducive to reducing the energy consumption of the smart door lock;

[0063] As a preferred technical solution of this application, the steps of waking up facial recognition in a time period outside the time window are as follows:

[0064] S101, the infrared sensing unit senses whether there is a living body approaching the smart door lock;

[0065] S102: If a living body is detected, start timing and synchronously collect voice information of the living body;

[0066] S103. When the living body residence time exceeds the threshold T2 or the preliminary recognition unit confirms that the collected voice content is consistent with the sample, the facial feature collection unit is immediately activated.

[0067] It can be understood that although an infrared sensing unit, a timing unit, a voice collection unit, a preliminary comparison unit (a second comparison unit), etc. are provided in the present embodiment, these are all technologies that are easy to implement in the prior art, and they all require certain conditions to operate, such as the preliminary comparison unit being triggered only when the user or other mobile personnel enters the sensing range of the sensing unit and actively generates voice information. Compared with the feature collection module being activated to collect facial information of living objects, and the vital sign extraction and subsequent feature comparison processes when a living object that looks similar to the user is collected, its energy consumption is relatively small, while at the same time providing users with a higher convenience.

[0068] Example 2, please refer to Figures 1 to 3 , the feature acquisition module provided in Example 1 is further optimized. Specifically, the auxiliary wake-up module also includes a second recognition unit, which is used to perform a deeper recognition of the voice information consistent with the content of the voice information sample (such as the recognition of the voiceprint), and calculate the voiceprint similarity between it and the voice information sample (the second recognition unit and the preliminary recognition unit can both realize the collected voice information through existing mature technologies, which will not be elaborated here). The purpose is to determine the identity of the person standing in front of the lock body. When the target person standing in front of the smart door lock is determined as a user through the second recognition unit, a face recognition unlocking method with higher unlocking efficiency can be provided.

[0069] Specifically, it also includes a feature extraction decision module, which dynamically adjusts the feature extraction strategy of the facial feature extraction unit according to the result of the second recognition unit, and its working steps are as follows:

[0070] S201, obtaining a speech similarity value calculated by a second recognition unit, and comparing it with a preset similarity threshold;

[0071] S202, if the similarity value is higher than the threshold, the facial feature extraction unit will focus on extracting key features in the facial image of the living object;

[0072] S203: If the similarity value is less than or equal to the threshold, the facial feature extraction unit extracts key features and detailed features in the facial image.

[0073] It should be noted that the process of calculating the similarity value of the voice by the second recognition unit is processed in parallel with the facial feature acquisition unit collecting the user's facial image, and the judgment is made before the facial feature extraction unit works. (It can be understood that in actual use, voice recognition is not used for authentication and unlocking, but only for determining the identity of the user. The actual unlocking needs to be coordinated with face recognition. The preset threshold can be set not too high, and it depends on the specific usage scenario);

[0074] In the specific implementation, when the similarity value is higher than the threshold, the system determines that the living object is a legitimate user. At this time, the key features of the user can be extracted through detection. The key features include but are not limited to eyebrows, eyes, nose, mouth, etc., so as to improve the efficiency of smart door lock unlocking and improve the user experience; correspondingly, if the similarity value is less than or equal to the threshold, it is judged that there is uncertainty in the user's identity. Then, on the basis of extracting the key features, it is necessary to extract the user's detailed features for face recognition unlocking. The detailed features include but are not limited to micro facial features, skin texture, pupil shape, etc., so as to improve the recognition accuracy and security; correspondingly, within the time window, when unlocking by face recognition, the key features and detailed features in the facial image are also extracted to ensure the consistent security of smart door lock face recognition unlocking.

[0075] Furthermore, the facial feature extraction unit constructs a facial feature extraction model based on a deep learning algorithm to complete step S202 or S203. Specifically, the method for constructing the facial feature extraction model is as follows:

[0076] S301. Data collection: Collect facial image data under different environments, such as facial image data under different angles, lighting, expressions, etc., to ensure the diversity of the data set.

[0077] S302, data preprocessing: preprocessing the collected face images, including image cleaning, cropping, scaling and normalization;

[0078] S303, data labeling: labeling the pre-processed face image, marking the position, size, key points and other information of the face, which is used to provide supervision signals when training the model;

[0079] S304, model construction: construct a dual-branch (parallel) facial feature extraction model through a convolutional neural network algorithm;

[0080] S305, model training: the collected face image data is divided into two training sets (training set, validation set and test set), the two training sets are respectively used for training, validation and evaluation of two branches of the facial feature extraction model, the two branches are respectively the key feature detection branch and the detailed facial feature extraction branch, the mean square error (MSE) can be used as the loss function for the training of the key feature detection branch, and the cross entropy loss or triplet loss can be used for optimization for the detailed facial feature extraction branch;

[0081] S306, model use: deploying the facial feature extraction model that has passed the evaluation into the facial feature extraction unit.

[0082] The above model construction method is an existing mature CNN model construction technology, which will not be described in detail here.

[0083] Furthermore, the structure of the facial feature extraction model mentioned above includes:

[0084] The shared underlying network is used to initially extract features of facial images, such as edge contours and simple textures of the face. Features are crucial for subsequent recognition and classification. It includes an input layer, at least one convolutional layer, and a corresponding maximum pooling layer. The input layer receives an original facial image of a specific size.

[0085] The first feature extraction branch, connected to the shared underlying network, responds to the recognition result of the second recognition unit (when the similarity is greater than a threshold, the feature point extraction branch is executed), further extracts and condenses the features output by the shared underlying network, focuses on information that is critical for rapid recognition, such as the position and shape of key facial parts such as eyes, nose and mouth, and includes at least one convolutional layer, a global average pooling layer, a fully connected layer and an output layer, and the output layer is configured to output a feature vector encoding key facial features;

[0086] The second feature extraction branch, connected to the shared underlying network, responds to the recognition result of the second recognition unit (when the similarity is less than or equal to the threshold, the feature point extraction branch is executed) and is used to further explore the detailed features of the face image such as skin texture, pupil shape, etc., and includes at least two convolutional layers, a global average pooling layer, a fully connected layer and an output layer. The output layer is configured to output a feature vector encoding key facial features and detailed features.

[0087] Among them, the number of convolution kernels of the convolution layer in the first feature extraction branch is greater than the number of convolution kernels of the convolution layer in the shared underlying network; the number of convolution kernels of the convolution layer in the second feature extraction branch is greater than the number of convolution kernels of the convolution layer in the first feature extraction branch.

[0088] The following is a brief description of the specific dual-branch CNN facial recognition network structure for reference:

[0089] Specifically, they share the underlying network, including:

[0090] Input layer: receives preprocessed face images and resizes them to a size suitable for CNN processing, such as 224x224 pixels;

[0091] Convolutional layer (Conv1): Convolution operation is performed using 32 3x3 convolution kernels, the activation function is ReLU, the stride is 1, and the padding is 1 to keep the output size consistent with the input;

[0092] MaxPooling1: 2x2 pooling window with a step size of 2, used to reduce the size of the feature map;

[0093] Convolutional layer (Conv2): uses 64 3x3 convolution kernels, the activation function is ReLU, the step size is 1, and the padding is 1;

[0094] MaxPool2: 2x2 pooling window with a stride of 2.

[0095] And the first feature extraction branch includes:

[0096] Convolution layer (Conv3_eff): uses 128 3x3 convolution kernels, the activation function is ReLU, the step size is 1, and the padding is 1, which is used to further extract features related to the key features;

[0097] Global average pooling layer (GlobalAveragePooling_eff): converts the feature map into a one-dimensional feature vector;

[0098] Fully connected layer (FC_eff): maps the one-dimensional feature vector to the classification space or embedding space for face recognition or verification;

[0099] Output layer: can be a softmax classification layer or a feature embedding layer for fast recognition.

[0100] The second feature extraction branch line parallel to the first feature extraction branch line includes:

[0101] Convolutional layer (Conv3_sec): uses 128 3x3 convolution kernels, the activation function is ReLU, the step size is 1, and the padding is 1;

[0102] Convolutional layer (Conv4_sec): uses 256 3x3 convolution kernels, ReLU activation function, step size 1, and padding 1 to extract more detailed features;

[0103] Global average pooling layer (GlobalAveragePooling_sec): also converts the feature map into a one-dimensional feature vector;

[0104] Fully connected layer (FC_sec): more neurons to process more complex features;

[0105] Output layer: softmax classification or feature embedding for high-precision recognition.

[0106] Through the setting of the above-mentioned facial feature extraction model, specifically, the feature extraction strategy of the facial feature extraction model is determined by the second recognition unit. When the model runs the first feature extraction branch, that is, in the scenario where the user voice similarity is greater than the threshold, the first feature extraction branch can quickly make an identification judgment, thereby reducing the user waiting time. Specifically, in terms of feature extraction, the first feature extraction branch extracts the user's key features, such as basic facial structure, main contours, etc., through relatively fewer convolutional layers and fully connected layers, which improves efficiency, but the features it extracts still have a certain uniqueness, and are implemented on the basis of cooperation with the voice recognition results, and still ensure the security of unlocking;

[0107] When the model runs the second feature extraction branch, the second feature extraction branch ensures a high-precision recognition. That is, in the scenario where the user voice similarity is less than or equal to the threshold, there is uncertainty in the user identity. The second feature extraction branch will perform in-depth feature extraction and comparison. For example, based on key features, it will extract skin texture, pupil shape, etc. to improve the accuracy and security of recognition. Specifically, the second feature extraction branch extracts the user's fine facial features through more convolutional layers and fully connected layers. These features make the recognition results more accurate and reliable.

[0108] In the present invention, unless otherwise clearly specified and limited, the terms "installed", "connected", "connected", "fixed" and the like should be understood in a broad sense, for example, it can be a fixed connection, a detachable connection, or an integral one; it can be a mechanical connection, an electrical connection, or communication with each other; it can be a direct connection, or an indirect connection through an intermediate medium, it can be the internal connection of two elements or the interaction relationship between two elements, unless otherwise clearly defined. For ordinary technicians in this field, the specific meanings of the above terms in the present invention can be understood according to specific circumstances.

[0109] Obviously, the embodiments described above are only some embodiments of the present invention, rather than all embodiments. The preferred embodiments of the present invention are given in the accompanying drawings, but they do not limit the patent scope of the present invention. The present invention can be implemented in many different forms. On the contrary, the purpose of providing these embodiments is to make the understanding of the disclosure of the present invention more thorough and comprehensive. Although the present invention has been described in detail with reference to the aforementioned embodiments, for those skilled in the art, it is still possible to modify the technical solutions recorded in the aforementioned specific embodiments, or to perform equivalent replacements for some of the technical features therein. Any equivalent structure made using the contents of the specification and drawings of the present invention, directly or indirectly used in other related technical fields, is also within the scope of patent protection of the present invention.

Claims

1. A biometric feature collection system, characterized in that: include: A feature acquisition module, comprising a facial feature acquisition unit and a facial feature extraction unit, wherein the facial feature acquisition unit is used to acquire facial image information of a living object standing in front of the lock body, and the facial feature extraction unit is used to extract feature points of the facial image information; The time management module is used to provide real-time time information and allow the user to set one or more time windows, which are the time periods when the user intends to use the smart door lock when returning home; The intelligent management module is used to control the working state of the feature collection module, control the feature collection module to be in an active state within the time window, and control the feature collection module to be in a dormant state outside the time window; The main wake-up module is used to decide whether to activate the facial feature acquisition unit within a set time window; Auxiliary wake-up module, used to decide whether to activate the facial feature acquisition unit outside the set time window; The auxiliary wake-up module includes: A voice collection unit is used to collect voice information of a living body when the infrared sensing unit senses that a living body is approaching outside the time window; A preliminary recognition unit is used to perform preliminary recognition on the collected voice information to verify whether its text content is consistent with the stored voice information sample; The second recognition unit is used to perform a deeper recognition of the voice information that is consistent with the voice information sample content, and calculate the voiceprint similarity between the voice information sample and the voice information sample; It also includes a feature extraction decision module, which dynamically adjusts the feature extraction strategy of the facial feature extraction unit according to the result of the second recognition unit, and its working steps are as follows: S201, obtaining a speech similarity value calculated by a second recognition unit, and comparing it with a preset similarity threshold; S202, if the similarity value is higher than the threshold, the facial feature extraction unit will focus on extracting key features in the facial image of the living object; S203, if the similarity value is less than or equal to the threshold, the facial feature extraction unit extracts key features and detailed features in the facial image; The facial feature extraction unit completes step S202 or S203 by constructing a facial feature extraction model based on a deep learning algorithm, and the structure of the facial feature extraction model includes: A shared underlying network, used for preliminarily extracting features of a face image, comprising an input layer, at least one convolutional layer, and a corresponding maximum pooling layer, wherein the input layer receives an original face image of a specific size; A first feature extraction branch, connected to the shared underlying network, responsive to the recognition result of the second recognition unit, for detecting key features of the face, comprising at least one convolutional layer, a global average pooling layer, a fully connected layer and an output layer, wherein the output layer is configured to output a feature vector encoding the key features of the face; The second feature extraction branch is connected to the shared underlying network and responds to the recognition result of the second recognition unit to extract key features and detailed features of the face. The second feature extraction branch includes at least two convolutional layers, a global average pooling layer, a fully connected layer and an output layer, and the output layer is configured to output a feature vector encoding the key features and detailed features of the face.

2. A biometric feature collection system according to claim 1, characterized in that: It also includes a data storage module for storing the user's biometric data, wherein the biometric data includes the user's facial feature samples and voice information samples.

3. A biometric feature collection system according to claim 2, characterized in that: The main wake-up module includes an infrared sensing unit and a timing unit. The infrared sensing unit senses whether a living object approaches the smart door lock; the timing unit is used to time the residence time of the sensed living object. When the residence time exceeds a threshold, the facial feature collection unit is immediately started to collect facial image information of the living object.

4. A biometric feature collection system according to claim 3, characterized in that: The timing unit includes a time threshold T1 within the time window and a time threshold T2 outside the time window, and T1 is smaller than T2.

5. A biometric feature collection system according to claim 4, characterized in that: In the time period outside the time window, the steps of waking up facial recognition are as follows: S101, the infrared sensing unit senses whether there is a living body approaching the smart door lock; S102: If a living body is detected, start timing and synchronously collect voice information of the living body; S103. When the living body residence time exceeds the threshold T2 or the preliminary recognition unit confirms that the collected voice and text content is consistent with the sample content, the facial feature collection unit is immediately activated.

6. A biometric feature collection system according to claim 5, characterized in that: The construction method of the facial feature extraction model is as follows: S301, data collection: collecting face image data in different environments; S302, data preprocessing: preprocessing the collected face images, including image cleaning, cropping, scaling and normalization; S303, data labeling: labeling the preprocessed face image, marking the position, size and key point information of the face, and providing supervision signals for training the model; S304, model construction: construct a double-branch facial feature extraction model through a convolutional neural network algorithm; S305, model training: dividing the collected face image data into two training sets, the two training sets are used for training, verification and evaluation of two branches of the facial feature extraction model respectively; S306, model use: deploying the facial feature extraction model that has passed the evaluation into the facial feature extraction unit.

7. A biometric feature collection system according to claim 6, characterized in that: The number of convolution kernels of the convolution layer in the first feature extraction branch is greater than the number of convolution kernels of the convolution layer in the shared underlying network; the number of convolution kernels of the convolution layer in the second feature extraction branch is greater than the number of convolution kernels of the convolution layer in the first feature extraction branch.

Images