Biological characteristic detection model determination method and device and electronic equipment

By acquiring and annotating biometric image data sets, the initial detection model is trained, and the model parameters are updated using real-life data, the problems of low training efficiency and poor adaptability in the prior art are solved, and a high-accuracy object detection model is achieved.

CN119027973APending Publication Date: 2024-11-26BEIJING CO WHEELS TECH CO LTD
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202310590032.0
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2023-05-24
Publication Date
2024-11-26

AI Technical Summary

Technical Problem

When training a neural network model for portrait detection, the prior art requires a lot of manpower and time cost to obtain sample data, and the setting action type is single, making it difficult to adapt to complex actual driving environments, affecting the training effect of the model.

Method used

By acquiring the first training data set, including multi-frame biometric images and labeled images, training is performed based on the initial detection model, and then obtaining multi-frame biometric images of the vehicle cockpit environment from the real scene shooting data, data cleaning and labeling are performed, and the model is updated The parameters are obtained to obtain the target detection model, which is suitable for actual driving and riding scenarios.

Benefits of technology

It reduces the cost of manually obtaining training samples, improves the training efficiency and accuracy of biometric detection models, and is suitable for complex actual driving and riding environments.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN119027973A_ABST
    Figure CN119027973A_ABST
Patent Text Reader

Abstract

The invention relates to a biological feature detection model determination method and device and electronic equipment, and particularly relates to the technical field of image processing. Comprising the steps that a first training data set is acquired, and the first training data set comprises multiple frames of first biological feature images and annotation images corresponding to each frame of first biological feature image; training the initial detection model based on the first training data set to obtain a biological feature detection model; acquiring multiple frames of second biological characteristic images acquired in the vehicle cabin environment from the live-action shooting data; based on the second training data set, updating model parameters of the biological feature detection model to obtain a target detection model; wherein the second training data set comprises multiple frames of second biological feature images and annotation images corresponding to each frame of second biological feature image. According to the embodiment of the invention, the training efficiency and training effect of the biological feature detection model are improved.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present disclosure relates to the field of image processing technology, and in particular to a method, device and electronic device for determining a biometric detection model. Background Art

[0002] With the learning and development of neural networks, portrait detection based on neural networks is widely used in various fields. In the field of smart cars, the user's travel and safety are facilitated by obtaining images in the car cabin for image recognition processing. For example, there are certain safety hazards when passengers walk around in the cabin while the car is driving. The use of neural networks to detect portraits in the cabin images is used to make safety reminders. At present, when training neural networks, models perform set actions in the cabin and shoot them, and the images obtained are used as training samples for the neural network. However, obtaining training samples in this way consumes a lot of manpower and time costs, and the actual driving environment is relatively complex. The set action type is single and more idealized, which is difficult to adapt to actual needs, thereby affecting the training effect of the neural network. Summary of the invention

[0003] In order to solve the above technical problems or at least partially solve the above technical problems, the present disclosure provides a method, device and electronic device for determining a biometric detection model, which can improve the training efficiency and training effect of the biometric detection model.

[0004] In order to achieve the above objectives, the technical solutions provided by the embodiments of the present disclosure are as follows:

[0005] In a first aspect, the present disclosure provides a method for determining a biometric detection model, comprising:

[0006] Acquire a first training data set, the first training data set comprising: a plurality of frames of first biometric image, and an annotated image corresponding to each frame of the first biometric image, wherein the annotated image is an image annotated with one or more of human body data, head data, and face data based on the biometric image;

[0007] Based on the first training data set, the initial detection model is trained to obtain a biometric detection model, where the biometric detection model is used to detect one or more of human body data, head data, and face data in the image;

[0008] Acquire multiple frames of second biometric feature images collected in the vehicle cabin environment from real-scene shooting data;

[0009] Based on the second training data set, updating the model parameters of the biometric detection model to obtain a target detection model;

[0010] The second training data set includes: multiple frames of second biometric image, and annotated images corresponding to each frame of the second biometric image.

[0011] As an optional implementation of the embodiment of the present disclosure, each frame of the second biometric image is a face coded image; multiple frames of the first biometric image include partial face coded images; wherein the face coded image refers to an image in which the grayscale value of all pixels inside the face external frame is 0.

[0012] As an optional implementation of the embodiment of the present disclosure, the multiple frames of second biometric image include: all second biometric images of the target scene, and at least one frame of second biometric image of other scenes, the target scene includes a scene of a special group of people, and / or a dangerous scene;

[0013] Acquiring multiple frames of second biometric image collected in a vehicle cockpit environment from real-scene shooting data, including: acquiring several frames of second biometric image collected in a vehicle cockpit environment from real-scene shooting data; determining all second biometric images of a target scene and all second biometric images of other scenes from the several frames of second biometric images; and randomly sampling from all second biometric images of other scenes to obtain at least one frame of second biometric image of other scenes.

[0014] As an optional implementation of the embodiment of the present disclosure, the step of acquiring the face coding image includes:

[0015] After acquiring a plurality of frames of first biometric feature images, identifying a face region included in the plurality of frames of first biometric feature images;

[0016] When a face region is identified, the face region is marked to obtain a face external frame;

[0017] The grayscale value of the image within the face frame is adjusted to 0 to obtain a coded face image.

[0018] As an optional implementation of the embodiment of the present disclosure, after updating the model parameters of the biometric detection model based on the second training data set to obtain the target detection model, the method includes:

[0019] Inputting an image to be detected into the target detection model, the image to be detected being different from the biometric image used to train the biometric detection model;

[0020] Obtain one or more of the human body data, head data and face data of the image to be detected output by the target detection model;

[0021] When one or more of the human body data, the human head data and the human face data match the human body key point data, the person position corresponding to the human body key point data is recorded;

[0022] Among them, the human body key point data is obtained after the model to be detected is input into the human body key point detection model, and the human body key point detection model is used to detect the human body key points in the image.

[0023] As an optional implementation of the embodiment of the present disclosure, after acquiring multiple frames of second biometric image collected in the vehicle cabin environment from the real scene shooting data, and before updating the model parameters of the biometric detection model based on the second training data set to obtain the target detection model, the method further includes:

[0024] Performing data cleaning on multiple frames of second biometric feature images;

[0025] On the multiple frames of second biometric image after data cleaning, one or more of the human body data, head data and face data are marked, and the target scene or other scenes are marked to obtain a marked image corresponding to each frame of the second biometric image.

[0026] In a second aspect, the present disclosure provides a device for determining a biometric detection model, comprising:

[0027] An acquisition module is used to acquire a first training data set, wherein the first training data set includes: a plurality of frames of first biometric image, and an annotated image corresponding to each frame of the first biometric image, wherein the annotated image is an image annotated with one or more of human body data, head data, and face data based on the biometric image;

[0028] A training module, used to train the initial detection model based on the first training data set to obtain a biometric detection model, where the biometric detection model is used to detect one or more of human body data, head data, and face data in an image;

[0029] The acquisition module is also used to acquire multiple frames of second biometric feature images collected in the vehicle cabin environment from the real-scene shooting data;

[0030] An updating module, configured to update the model parameters of the biometric detection model based on the second training data set to obtain a target detection model;

[0031] The second training data set includes multiple frames of second biometric image and annotated images corresponding to each frame of the second biometric image.

[0032] As an optional implementation of the embodiment of the present disclosure, each frame of the second biometric image is a face coded image; multiple frames of the first biometric image include partial face coded images; wherein the face coded image refers to an image in which the grayscale value of all pixels inside the face external frame is 0.

[0033] As an optional implementation of the embodiment of the present disclosure, the multiple frames of second biometric image include: all second biometric images of the target scene, and at least one frame of second biometric image of other scenes, the target scene includes a scene of a special group of people, and / or a dangerous scene;

[0034] The acquisition module acquires multiple frames of second biometric image collected in the vehicle cockpit environment from the real-scene shooting data, and is specifically used to: acquire several frames of second biometric image collected in the vehicle cockpit environment from the real-scene shooting data; determine all second biometric images of the target scene and all second biometric images of other scenes from the several frames of second biometric images; and randomly sample from all second biometric images of other scenes to obtain at least one frame of second biometric image of other scenes.

[0035] As an optional implementation of the embodiment of the present disclosure, the acquisition module, for the step of acquiring the face coding image, is specifically used to: after acquiring multiple frames of first biometric feature images, identify the face area included in the multiple frames of first biometric feature images;

[0036] When a face region is identified, the face region is marked to obtain a face external frame;

[0037] The grayscale value of the image within the face frame is adjusted to 0 to obtain a coded face image.

[0038] As an optional implementation of the embodiment of the present disclosure, an update module is used to input an image to be detected into a target detection model, where the image to be detected is different from a biometric image used to train the biometric detection model; obtain one or more of human body data, head data, and face data of the image to be detected output by the target detection model;

[0039] When one or more of the human body data, the human head data and the human face data match the human body key point data, the person position corresponding to the human body key point data is recorded;

[0040] Among them, the human body key point data is obtained after the model to be detected is input into the human body key point detection model, and the human body key point detection model is used to detect the human body key points in the image.

[0041] As an optional implementation of the embodiment of the present disclosure, after acquiring multiple frames of second biometric feature images collected in the vehicle cockpit environment from the real-scene shooting data, the acquisition module is further used to:

[0042] Performing data cleaning on multiple frames of second biometric feature images;

[0043] On the multiple frames of second biometric image after data cleaning, one or more of the human body data, head data and face data are marked, and the target scene or other scenes are marked to obtain a marked image corresponding to each frame of the second biometric image.

[0044] In a third aspect, the present disclosure provides an electronic device comprising: a processor, a memory, and a computer program stored in the memory and executable on the processor, wherein when the computer program is executed by the processor, the method for determining a biometric detection model as described in the first aspect or any one of its optional embodiments is implemented.

[0045] In a fourth aspect, the present disclosure provides a computer-readable storage medium, comprising: a computer program is stored on the computer-readable storage medium, and when the computer program is executed by a processor, the method for determining a biometric detection model as described in the first aspect or any one of its optional embodiments is implemented.

[0046] In a fifth aspect, a computer program product is provided, comprising: when the computer program product is run on a computer, the computer implements the method for determining a biometric detection model as described in the first aspect or any one of its optional embodiments.

[0047] Compared with the prior art, the technical solution provided by the embodiments of the present disclosure has the following advantages:

[0048] The disclosed embodiments provide a method, device and electronic device for determining a biometric detection model, wherein the method first obtains a first training data set, wherein the first training data set includes: multiple frames of first biometric images, and annotated images corresponding to each frame of the first biometric image, wherein the annotated images are images annotated with one or more of the following data based on the biometric images: human body data, head data and face data; then, an initial detection model is trained based on the first training data set to obtain a biometric detection model, and then multiple frames of second biometric images collected in a vehicle cabin environment are obtained from real-scene shooting data; the multiple frames of second biometric images and the annotated images corresponding to each frame of the second biometric image are used as a second training data set, and further based on the second training data set, model parameters of the biometric detection model are updated to obtain a target detection model. The present invention obtains a biometric detection model by training a first training data set obtained by manual collection, and then obtains a second training data set using real and diverse real-life shooting data, and updates the model parameters of the biometric detection model, thereby obtaining a target detection model that meets actual needs, is suitable for actual driving scenarios, has high accuracy, reduces the cost of manually obtaining training samples, and improves the training efficiency of the biometric detection model. BRIEF DESCRIPTION OF THE DRAWINGS

[0049] The accompanying drawings, which are incorporated in and constitute a part of this specification, illustrate embodiments consistent with the present disclosure and, together with the description, serve to explain the principles of the present disclosure.

[0050] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, for ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative labor.

[0051] Figure 1 A schematic diagram of a flow chart of a method for determining a biometric detection model provided in an embodiment of the present disclosure;

[0052] Figure 2 A schematic diagram of the structure of a device for determining a biometric detection model is provided for an embodiment of the present disclosure;

[0053] Figure 3 A schematic structural diagram of an electronic device is provided in accordance with an embodiment of the present disclosure. DETAILED DESCRIPTION

[0054] In order to more clearly understand the above-mentioned objectives, features and advantages of the present disclosure, the scheme of the present disclosure will be further described below. It should be noted that the embodiments of the present disclosure and the features in the embodiments can be combined with each other without conflict.

[0055] In the following description, many specific details are set forth to facilitate a full understanding of the present disclosure, but the present disclosure may also be implemented in other ways different from those described herein; it is obvious that the embodiments in the specification are only part of the embodiments of the present disclosure, rather than all of the embodiments.

[0056] At present, in order to train the neural network model for portrait detection, when constructing the sample data set, most of the time, special models are hired to perform set actions to obtain sample images. On the one hand, the amount of data required for training is huge, and obtaining the sample data set through this manual collection method requires a lot of manpower and time costs, which affects the efficiency of model training; on the other hand, the set actions are the types of actions that people can imagine that may occur in the driving environment, which is idealized, fixed and of few types, and difficult to adapt to the complex actual driving environment. Therefore, the model trained based on such a sample data set is not accurate.

[0057] In order to solve the above problems, the embodiments of the present disclosure provide a method, device and electronic device for determining a biometric detection model, wherein the method first obtains a first training data set, wherein the first training data set includes: multiple frames of first biometric images, and annotated images corresponding to each frame of the first biometric image, wherein the annotated images are images annotated with one or more of the following data based on the first biometric image: human body data, head data and face data; then, an initial detection model is trained based on the first training data set to obtain a biometric detection model, and then multiple frames of second biometric images collected in a vehicle cabin environment are obtained from real-life shooting data; the multiple frames of second biometric images and the annotated images corresponding to each frame of the second biometric image are used as a second training data set, and further based on the second training data set, the model parameters of the biometric detection model are updated to obtain a target detection model. The present invention obtains a biometric detection model by training a first training data set obtained by manual collection, and then obtains a second training data set using real and diverse real-life shooting data, and updates the model parameters of the biometric detection model, thereby obtaining a target detection model that meets actual needs, is suitable for actual driving scenarios, has high accuracy, reduces the cost of manually obtaining training samples, and improves the training efficiency of the biometric detection model.

[0058] The method for determining a biometric detection model provided in the embodiments of the present disclosure can be applied to a device or electronic device for determining a biometric detection model, and the electronic device includes but is not limited to a vehicle terminal. The operating system of the vehicle terminal may include Android, a mobile operating system developed by Apple (iOS), an operating system developed by Microsoft (Windows), etc., and the embodiments of the present disclosure do not limit this.

[0059] It should be noted that the protection scope of the method for determining a biometric detection model described in the embodiment of the present disclosure is not limited to the execution order of the steps listed in this embodiment. All solutions implemented by adding, reducing or replacing steps in the prior art based on the principles of the present disclosure are included in the protection scope of the present disclosure.

[0060] like Figure 1 As shown, Figure 1 A schematic flow chart of a method for determining a biometric detection model provided in an embodiment of the present disclosure, the method comprising the following steps S101 to S104:

[0061] S101: Obtain a first training data set.

[0062] The first training data set includes: multiple frames of first biometric images, and annotated images corresponding to each frame of the first biometric image. The annotated image is an image that is annotated with one or more of human body data, head data, and face data based on the biometric image. To simplify the description and facilitate understanding, the "annotated image corresponding to each frame of the first biometric image" will be referred to as the "first annotated image" in the following.

[0063] In some embodiments, human body data includes: a human body bounding box annotated on the basis of the image to be detected, and coordinate information of the human body bounding box. Human head data includes: a human head bounding box annotated on the basis of the image to be detected, and coordinate information of the human head bounding box. Human face data includes: a human face bounding box annotated on the basis of the image to be detected, and coordinate information of the human face bounding box.

[0064] In some embodiments, in order to improve the training accuracy, the multiple frames of first biometric images provided by the present disclosure include partially coded faces. The present disclosure provides an implementation method, after capturing multiple frames of first biometric images, a random algorithm is used to randomly extract partial biometric images from the multiple frames of first biometric images, and these partial biometric images are coded to obtain partially coded faces. The other first biometric images in the multiple frames of first biometric images except the partially coded faces can be the original first biometric images.

[0065] Optionally, the face coding image refers to an image in which the grayscale value of all pixels within the face external frame is 0. After acquiring multiple frames of the first biometric image, for any one of the multiple frames of the first biometric image, the face is identified as the region of interest. After the face region is identified, the face region is marked in the first biometric image to obtain the face external frame. At this time, the first biometric image is divided into two parts: the image within the face external frame and the image outside the face external frame. Then, the grayscale value of the image within the face external frame is adjusted to 0, so that the grayscale values ​​of all pixels within the face external frame in the first biometric image are 0, thereby achieving face coding.

[0066] Optionally, after randomly extracting a partial biometric image, facial image features of the partial biometric image are extracted, including but not limited to: facial area features, eye area features, and lip area features. The disclosed embodiment may encode one or more of the above facial image features. Exemplarily, the disclosed embodiment encodes the extracted eye area features to obtain a partially encoded face image.

[0067] Optionally, after randomly extracting some biometric images, when coding the part of the biometric images, the face area of ​​the part of the biometric images can be covered by a pre-set mask image to obtain a part of the face coded image, thereby enhancing the diversity of sample data and improving the robustness of model training. The parameters such as the size and grayscale value of the mask image are not specifically limited in this disclosure.

[0068] The above optional implementation modes may be arbitrarily combined. For example, after randomly extracting a partial biometric image, the eye area features of the partial biometric image are extracted, and the eye area features in the partial biometric image are covered by a mask image corresponding to a pre-set eye area to obtain a partial face coded image.

[0069] The above embodiment codes part of the first biometric image included in the acquired multiple frames of the first biometric image, thereby laying a foundation for subsequent training and updating of the model using real-scene shooting data, thereby improving the robustness and accuracy of the model training.

[0070] In some embodiments, the multiple frames of first biometric image provided by the present disclosure include a partial head coded image. The present disclosure provides an implementation method. After the multiple frames of first biometric image are captured, a random algorithm is used to randomly extract a partial biometric image from the multiple frames of first biometric image, and these partial biometric images are coded to obtain a partial head coded image. The other first biometric images in the multiple frames of first biometric image except the partial head coded image can be the original first biometric image and / or a partial face coded image.

[0071] Optionally, the head coding image refers to an image in which the grayscale values ​​of all pixels inside the head external frame are 0. After acquiring multiple frames of the first biometric image, for any one of the multiple frames of the first biometric image, the head is identified as the region of interest. After the region of interest is identified, the region of interest is marked in the first biometric image to obtain the head external frame. At this time, the first biometric image is divided into two parts: the image inside the head external frame and the image outside the head external frame. Then, the grayscale value of the image inside the head external frame is adjusted to 0, so that the grayscale values ​​of all pixels inside the head external frame in the first biometric image are 0, thereby achieving head coding.

[0072] Optionally, after randomly extracting a partial biometric image, extract head image features of the partial biometric image, including but not limited to: head region features, ear region features, and facial region features. The disclosed embodiment may code one or more of the above head image features. Exemplarily, the disclosed embodiment codes the extracted head region features to obtain a partially coded head image.

[0073] Optionally, after randomly extracting a portion of the biometric image, when coding the portion of the biometric image, the head region of the portion of the biometric image may be covered by a pre-set mask image to obtain a portion of the head coded image. The parameters such as the size and grayscale value of the mask image are not specifically limited in the present disclosure. It is understood that the mask image may partially or completely cover the head region.

[0074] The above optional implementation modes may be combined arbitrarily. For example, after randomly extracting a partial biometric image, the ear area features of the partial biometric image are extracted, and the ear area features in the partial biometric image are covered by a mask image corresponding to a preset ear area to obtain a partial head coded image.

[0075] In some embodiments, the first biometric image frames provided by the present disclosure include human body coding images. In the images captured in the actual driving environment, there are cases where the cabin interior obscures the occupants, such as the seat headrest obscures the rear passengers. Human body coding is adapted to such actual obstruction scenarios, and the trained model has strong applicability and high accuracy.

[0076] The disclosed embodiment provides an implementation method, after capturing multiple frames of first biometric feature images, a random algorithm is used to randomly extract part of the biometric feature images from the multiple frames of first biometric feature images, and these part of the biometric feature images are coded to obtain part of the human body coded images. The other first biometric feature images in the multiple frames of first biometric feature images except for the part of the human body coded image can be at least one of: the original first biometric feature image, the part of the human face coded image, and the part of the human head coded image.

[0077] Optionally, the human body coding image refers to an image in which the grayscale values ​​of all pixels inside the human body external frame are 0. After acquiring multiple frames of the first biometric image, for any one of the multiple frames of the first biometric image, the human body is identified as a region of interest. After the region of interest is identified, the region of interest is marked in the first biometric image to obtain the human body external frame. At this time, the first biometric image is divided into two parts: the image inside the human body external frame and the image outside the human body external frame. Then, the grayscale value of the image inside the human body external frame is adjusted to 0, so that the grayscale values ​​of all pixels inside the human body external frame in the first biometric image are 0, thereby achieving human body coding.

[0078] Optionally, after randomly extracting a partial biometric feature image, human image features of the partial biometric feature image are extracted, including but not limited to: human body region features, fingerprint region features, and palm print region features. The disclosed embodiment may code one or more of the above human body image features. Exemplarily, the disclosed embodiment codes the extracted palm print region features to obtain a partially coded human body image.

[0079] Optionally, after randomly extracting a portion of the biometric image, when coding the portion of the biometric image, the human body region of the portion of the biometric image may be covered by a pre-set mask image to obtain a portion of the human body coded image. The parameters such as the size and grayscale value of the mask image are not specifically limited in the present disclosure. It is understood that the mask image may partially or completely cover the human body region.

[0080] The above optional implementation modes may be combined arbitrarily. For example, after randomly extracting a partial biometric image, the palm print area features of the partial biometric image are extracted, and the palm print area features in the partial biometric image are covered by a mask image corresponding to a preset palm print area to obtain a partial human body coded image.

[0081] In some embodiments, the multiple frames of first biometric image provided by the present disclosure include at least one of the above-mentioned coded images, that is, the multiple frames of first biometric image include: at least one of: a face coded image, a head coded image, and a body coded image. Accordingly, the first training data set includes annotated images corresponding to each coded image.

[0082] The above embodiment desensitizes the biometric information based on the first training data set collected manually by coding at least one of the face, head and body, so as to be suitable for desensitizing the real-scene shooting data when the real-scene shooting data is subsequently used for model training and updating.

[0083] S102: Based on the first training data set, train the initial detection model to obtain a biometric detection model.

[0084] The initial detection model is based on at least one neural network of a backbone network, a shoulder network, and a head network. The biometric detection model is used to detect one or more of human body data, head data, and face data in an image. The model parameters of the initial detection model and the biometric detection model include, but are not limited to, connection weights and biases of the neural network. During the training process, multiple frames of the first biometric image in the first training data set are input into the initial detection model, and the initial detection model outputs a predicted annotated image. The predicted annotated image is compared with the first annotated image in the first training data set, and the loss function value is calculated. When the loss function value is greater than the preset loss value, the model parameters of the initial detection model are returned to adjust until the loss function value between the predicted annotated image output by the model and the first annotated image is less than or equal to the preset loss function value, and the model convergence can be determined to obtain a converged biometric detection model.

[0085] S103: Acquire multiple frames of second biometric feature images collected in the vehicle cabin environment from the real-scene shooting data.

[0086] Among them, the real-scene shooting data is image data captured after the user's permission, including several second biometric images collected in the vehicle cabin environment in the actual scene.

[0087] In some embodiments, the plurality of frames of the second biometric image include: all the second biometric images of the target scene, and at least one frame of the second biometric image of other scenes. The target scene includes: a scene of a special group of people, and / or a dangerous scene. The target scene and other scenes can be distinguished by manual annotation. Exemplarily, the target scene can be a scene of passengers moving freely in the cabin aisle, or a scene of an infant leaving a child safety seat.

[0088] The disclosed embodiment provides an implementation method for obtaining multiple frames of second biometric image collected in a vehicle cockpit environment from real-scene shooting data, including: obtaining several frames of second biometric image collected in a vehicle cockpit environment from real-scene shooting data, and then determining all second biometric images of a target scene and all second biometric images of other scenes from these several frames of second biometric images, and then randomly sampling from all second biometric images of other scenes to obtain at least one frame of second biometric image of other scenes. Thus, at least one frame of second biometric image of other scenes is used as a negative sample for model training, and then the biometric detection model is trained subsequently to accurately distinguish the target scene from other scenes.

[0089] In some embodiments, after obtaining multiple frames of second biometric feature images collected in the vehicle cockpit environment from real-life shooting data, data cleaning is first performed on the multiple frames of second biometric feature images. Optionally, feature extraction is performed on the multiple frames of second biometric feature images containing noise based on a neural network model pre-trained on an image dataset (ImageNet). Unsupervised clustering algorithms such as density-based spatial clustering of applications with noise (DBSCAN) are used to cluster the multiple frames of second biometric feature images based on the extracted features to obtain multiple clusters, determine the cluster containing the largest number of image data, and use the image data in the cluster as clean data after cleaning. Further, data annotation is performed on the clean data, that is, the multiple frames of second biometric feature images after data cleaning.

[0090] In some embodiments, data annotation is performed on multiple frames of second biometric images, including annotating one or more of human body data, head data, and face data, as well as target scenes or other scenes. For example, for any second biometric image in the multiple frames of second biometric images, the face is identified as the region of interest. After the region of interest is identified, the region of interest is annotated in the second biometric image to obtain a face bounding box, and the coordinate information of the face bounding box is determined. The annotation of head data and human body data is similar, and the present disclosure will not elaborate on this. For any second biometric image in the multiple frames of second biometric images, if it is determined that it corresponds to the target scene, it is labeled with the target scene; if it is determined that it corresponds to other scenes, it is labeled with other scenes to annotate the multiple frames of second biometric images with target scenes or other scenes.

[0091] Furthermore, the annotated multiple frames of the second biometric image are coded to protect user privacy. The coding process includes coding the face, coding the head, and coding the body. Accordingly, the multiple frames of the second biometric image can be at least one of the second face coded image, the second head coded image, and the second body coded image. The following will introduce these three coding methods respectively:

[0092] (1) Face coding

[0093] The multiple frames of second biometric feature images include a second face coding image.

[0094] Optionally, after the second biometric image is annotated with a face, the second biometric image is divided into two parts: an image within the face circumference frame and an image outside the face circumference frame. Then, the grayscale value of the image within the face circumference frame is adjusted to 0, so that the grayscale values ​​of all pixels within the face circumference frame in the second biometric image are 0, thereby achieving face coding.

[0095] Optionally, for any second biometric image in the multiple frames of the second biometric image, extract facial image features of the second biometric image, including but not limited to: facial area features, eye area features, and lip area features. The disclosed embodiment may encode one or more of the above facial image features. Exemplarily, the disclosed embodiment encodes the extracted eye area features to obtain a second facial coded image.

[0096] Optionally, for any second biometric image in the multiple frames of second biometric images, the face area of ​​the second biometric image can be masked by a pre-set mask image to obtain a second face coded image, thereby enhancing the diversity of sample data and improving the robustness of model training. The parameters such as the size and grayscale value of the mask image are not specifically limited in this disclosure.

[0097] The above optional implementation modes may be arbitrarily combined. For example, for any second biometric image in multiple frames of second biometric images, eye area features of the second biometric image are extracted, and the eye area features in the second biometric image are covered by a mask image corresponding to a preset eye area to obtain a second facial coded image.

[0098] The above embodiment avoids leakage of user's facial privacy information during model training by coding the face of multiple frames of the acquired second biometric feature image, so that the effect of desensitization detection can be achieved when the model is subsequently applied for portrait detection.

[0099] (2) Head coding

[0100] The multiple frames of second biometric feature images include: a second head coded image, and in addition, may also include a second face coded image.

[0101] Optionally, after the second biometric image is labeled with the head, the second biometric image is divided into two parts: an image within the head external frame and an image outside the head external frame. Then, the grayscale value of the image within the head external frame is adjusted to 0, so that the grayscale values ​​of all pixels within the head external frame in the second biometric image are 0, thereby achieving head coding.

[0102] Optionally, for any second biometric image in the multiple frames of second biometric images, head image features of the second biometric image are extracted, including but not limited to: head region features, ear region features, and facial region features. The disclosed embodiment may code one or more of the above head image features. Exemplarily, the disclosed embodiment codes the extracted head region features to obtain a second head coded image.

[0103] Optionally, for any second biometric image in the multiple frames of second biometric images, the head area of ​​the second biometric image can be covered by a pre-set mask image to obtain a second head coded image. The parameters such as the size and grayscale value of the mask image are not specifically limited in the present disclosure. It is understood that the mask image can cover the head area or the entire head area.

[0104] The above optional implementation modes may be arbitrarily combined. For example, for any second biometric image in multiple frames of second biometric images, the ear area features of the second biometric image are extracted, and the ear area features in the second biometric image are covered by a mask image corresponding to a preset ear area to obtain a second head coded image.

[0105] (3) Human body coding

[0106] The multiple frames of second biometric feature images include a second human body coding image, and in addition may also include a second face coding image and / or a second head coding image.

[0107] Optionally, after the second biometric image is annotated with human body, the second biometric image is divided into two parts: an image within the human body external frame and an image outside the human body external frame. Then, the gray value of the image within the human body external frame is adjusted to 0, so that the gray values ​​of all pixels within the human body external frame in the second biometric image are 0, thereby realizing human body coding.

[0108] Optionally, after randomly extracting the second biometric feature image, human image features of the second biometric feature image are extracted, including but not limited to: human body region features, fingerprint region features, and palm print region features. The disclosed embodiment may encode one or more of the above human body image features. Exemplarily, the disclosed embodiment encodes the extracted palm print region features to obtain a second human body coded image.

[0109] Optionally, after randomly extracting the second biometric image, when coding the second biometric image, the human body area of ​​the second biometric image can be covered by a pre-set mask image to obtain the second human body coded image. The parameters such as the size and grayscale value of the mask image are not specifically limited in the present disclosure. It is understood that the mask image can cover the second human body area or cover the entire human body area.

[0110] The above optional implementation modes may be combined arbitrarily. For example, after randomly extracting the second biometric image, the palm print area features of the second biometric image are extracted, and the palm print area features in the second biometric image are covered by a mask image corresponding to a preset palm print area to obtain a second human body coding image.

[0111] Furthermore, multiple frames of the second biometric image after coding and the annotated image corresponding to each frame of the second biometric image are used as the second training data set, wherein the annotated image corresponding to each frame of the second biometric image is annotated with one or more of the human body data, head data and face data on the basis of the second biometric image, as well as an image of the target scene or other scenes, and one or more of the specific human body features, head features and face features have been coded and desensitized. For the purpose of simplifying the explanation and distinguishing it from the aforementioned "first annotated image", the "annotated image corresponding to each frame of the second biometric image" will be referred to as the "second annotated image" in the following.

[0112] S104: Based on the second training data set, update the model parameters of the biometric detection model to obtain a target detection model.

[0113] The second training data set includes multiple frames of second biometric images, and a second annotated image corresponding to each frame of the second biometric image. Each frame of the second biometric image is a biometric image after coding, and can be at least one of a second face coding image, a second head coding image, and a second body coding image. Optionally, each frame of the second biometric image is a second face coding image; the biometric detection model is obtained by training based on a portion of the face coding image. Similarly, each frame of the second biometric image is a second head coding image or a second body coding image, and the corresponding biometric detection model is obtained by training based on a portion of the head coding image or a portion of the body coding image.

[0114] In some embodiments, based on the second training data set, the model parameters of the biometric detection model are updated to obtain a target detection model, including using multiple frames of second biometric images in the second training data set as input of the biometric detection model to obtain a predicted annotated image output by the biometric detection model, comparing the predicted annotated image output by the biometric detection model with the second annotated image, calculating the loss function value between the predicted annotated image and the second annotated image according to a loss function, and when the loss function value is greater than a preset loss value, returning to adjust the model parameters of the biometric detection model until the loss function value between the predicted annotated image output by the biometric detection model and the second annotated image is less than or equal to the preset loss value, determining that the biometric detection model converges, thereby achieving an update of the biometric detection model to obtain a converged target detection model.

[0115] The target detection model can perform portrait detection on the coded image. When the second biometric image of the target scene is included in the second training data set, the target detection model can be used to identify the target scene, thereby providing a safety reminder to the user. In some embodiments, after the biometric detection model is updated based on the second training data set to obtain the target detection model, the method further includes: inputting the image to be detected into the target detection model, the image to be detected is different from the biometric image used for model training; and obtaining one or more of the human body data, head data, and face data of the image to be detected output by the target detection model.

[0116] By using the target detection model to perform portrait detection, one or more of the human body data, head data and face data can be identified from the coded image to be detected, thereby identifying the target scene. For example, if the image to be detected is an image of a child leaving the safety seat, the target detection model can accurately identify the facial features of the child, and thus determine that the child is in a dangerous situation of leaving the safety seat. Further combined with other components in the vehicle cabin, such as audio, a safety reminder is output to warn other people in the car that the child has left the safety seat.

[0117] In some embodiments, the trained target detection model can be applied, and the image to be detected can be used as the input of the target detection model to obtain one or more of the human body data, head data and face data output by the target detection model. Optionally, the image to be detected can also be input into a human key point detection model. The human key point detection model is a pre-trained and converged model for detecting human body key points in an image. The human body key points output by the human body key point detection model can be obtained. The human body key point data includes the human body key points marked out on the basis of the image to be detected, as well as the position coordinates of the human body key points.

[0118] It should be emphasized that the target detection model is used to detect at least one of the human body data, head data and face data in the image, and the human body key point detection model is used to detect the human body key points in the image. These two models are trained independently and are also relatively independent when applied. The embodiment of the present disclosure uses the output of the target detection model and the output of the human body key point detection model for comparison and matching, thereby accurately predicting the position of the driver and passenger in the vehicle cabin environment, thereby improving the accuracy of personnel detection. The output of the target detection model and the output of the human body key point detection model are used to predict the position of the driver and passenger in the vehicle cabin environment. Specifically, the human body data, head data and / or face data output by the target detection model are obtained, and the human body key point data output by the human body key point detection model are obtained. The human body key point data, human body data and head data and / or face data are matched, and the matching results include the following three situations: ① Whether the human body data matches the head data; ② Whether the human body data matches the face data; ③ Whether the human body data matches the head data and the face data. It should be emphasized that the human body key point data is specifically used in the different matching results of the above three situations.

[0119] The following will introduce these three situations:

[0120] ①Whether the human body data matches the head data

[0121] In some embodiments, after obtaining the human body data and human head data output by the target detection model, the human body data and the human head data are matched using the Hungarian algorithm. Specifically, the human body external bounding box and the head external bounding box are matched using the Hungarian algorithm to correspond the human body data and the human head data.

[0122] If the human body data and the head data match successfully, the confidence of the head data is further determined. The confidence of the head data is used to indicate the probability that the image to be detected contains the head data. It is calculated and output by the target detection model. Then, it is determined whether the confidence of the head data is greater than or equal to a preset threshold. The preset threshold is the preset minimum probability of determining that the image contains the head data.

[0123] If the confidence of the head data is greater than or equal to the preset threshold, it means that the head data is contained in the image to be detected. It can be understood that there is a valid human body corresponding to the head data and the matching human body data, then the person position corresponding to the human body key point data is recorded.

[0124] If the confidence of the head data is less than the preset threshold, it is preliminarily determined that the image to be detected does not contain the head data, then the first external bounding box corresponding to the head area key point and the head external bounding box corresponding to the head data are determined according to the head area key point in the human key point data output by the human key point detection model, and then the head external bounding box is matched with the first external bounding box by intersection and union ratio. If the intersection and union ratio match between the head external bounding box and the first external bounding box is successful, it means that after the secondary processing, it is determined that the image to be detected contains the head data, which means that there is a valid human body corresponding to the head data and the human body data matched therewith; further, the position of the person corresponding to the human key point data is recorded.

[0125] If the human body data and the human head data fail to match, it is determined that the detected human body data is a misjudgment and the human body data is deleted.

[0126] In other embodiments, the human body data, human head data and human body key point data are matched. Specifically, the human body external frame is matched with the minimum external frame of the human body key point data, and the human head external frame is matched with the first external frame corresponding to the key points of the head area in the human body key point data. If all matches are successful, it is determined that there is a valid human body corresponding to the human body data, and the person position corresponding to the human body key point data is recorded, that is, the coordinates of the human body key point data are used as the person position coordinates.

[0127] ②Whether the human body data matches the face data

[0128] In some embodiments, after obtaining the human body data and facial data output by the target detection model, the human body data and facial data are matched using the Hungarian algorithm. Specifically, the human body bounding box and the facial bounding box are matched using the Hungarian algorithm to correspond the human body data and the facial data.

[0129] If the human body data and the face data match successfully, the confidence of the face data is further determined. The confidence of the face data is used to indicate the probability that the image to be detected contains face data, which is calculated and output by the target detection model. Then, it is determined whether the confidence of the face data is greater than or equal to a preset threshold, which is a preset minimum probability for determining that face data is contained.

[0130] If the confidence of the facial data is greater than or equal to the preset threshold, it means that the facial data is contained in the image to be detected. It can be understood that there is a valid human body corresponding to the facial data and the matching human body data, then the person position corresponding to the human body key point data is recorded.

[0131] If the confidence of the face data is less than the preset threshold, it is preliminarily determined that the image to be detected does not contain the face data, then the second external bounding box corresponding to the face area key point and the face external bounding box corresponding to the face data are determined according to the face area key point in the human key point data output by the human key point detection model, and then the face external bounding box and the second external bounding box are matched by intersection and union ratio. If the intersection and union ratio of the face external bounding box and the second external bounding box is successfully matched, it means that after the secondary processing, it is determined that the image to be detected contains face data, which means that there is a valid human body corresponding to the face data and the human body data matched therewith; further, the position of the person corresponding to the human body key point data is recorded.

[0132] If the body data and face data fail to match, it is determined that the detected body data is a misjudgment and the body data is deleted.

[0133] In other embodiments, the human body data, facial data and human body key point data are matched. Specifically, the human body bounding box is matched with the minimum bounding box of the human body key point data, and the facial bounding box is matched with the second bounding box corresponding to the key points of the facial area in the human body key point data. If all matches are successful, it is determined that there is a valid human body corresponding to the human body data, and the person position corresponding to the human body key point data is recorded, that is, the coordinates of the human body key point data are used as the person position coordinates.

[0134] ③Whether the human body data matches the head data and face data

[0135] In some embodiments, after obtaining the human body data, head data and face data output by the target detection model, the human body data is matched with the head data and the face data through the Hungarian algorithm. Specifically, the human body external frame, the head external frame and the face external frame are matched through the Hungarian algorithm to correspond the human body data, head data and face data.

[0136] If the human body data, head data and face data match successfully, the confidence of the head data and face data is further determined, and then it is determined whether the confidence of the head data and face data is greater than or equal to a preset threshold.

[0137] If the confidence of the head data and face data is greater than or equal to the preset threshold, it means that the image to be detected contains the head data and face data. It can be understood that there is a valid human body corresponding to the head data and face data and the human body data matching them, then the person position corresponding to the human body key point data is recorded.

[0138] If the confidence of the head data and face data is less than the preset threshold, it is preliminarily determined that the image to be detected does not contain the head data and face data, then the first external bounding box corresponding to the head area key point and the second external bounding box corresponding to the face area key point, as well as the head external bounding box corresponding to the head data and the face external bounding box corresponding to the face data are determined according to the head area key points and the face area key points in the human key point data output by the human key point detection model, and then the head external bounding box is matched with the first external bounding box by intersection and union ratio, and the face external bounding box is matched with the second external bounding box by intersection and union ratio. If the head external bounding box successfully matches the first external bounding box by intersection and union ratio, and the face external bounding box successfully matches the second external bounding box by intersection and union ratio, it means that after the secondary processing, it is determined that the image to be detected contains the head data and face data, which means that there is a valid human body corresponding to the head data, face data and the human body data matched therewith; further, the position of the person corresponding to the human body key point data is recorded.

[0139] If the human body data fails to match the head data and the face data, it is determined that the detected human body data is a misjudgment and the human body data is deleted.

[0140] In other embodiments, the human body data, human head data, human face data and human body key point data are matched. Specifically, the human body external frame is subjected to intersection and union matching with the minimum external frame of the human body key point data, the human head external frame is subjected to intersection and union matching with the first external frame, and the human head external frame is subjected to intersection and union matching with the second external frame of the key points of the facial area in the human body key point data. If all the matches are successful, it is determined that there is a valid human body corresponding to the human body data, and the person position corresponding to the human body key point data is recorded, that is, the coordinates of the human body key point data are used as the person position coordinates.

[0141] In some embodiments, in the process of recording the position of a person corresponding to the key point data of the human body, the key points of the hip region in the key point data of the human body are first determined, and the key points of the hip region are used to represent the position of the effective human body, thereby recording the position of the person corresponding to the key points of the hip region. By using the key points of the hips to represent the position of the person, the interference of complex sitting postures on the detection of the person is eliminated, so that not only the position of the person sitting in a correct posture can be recorded, but also the position of other people with complex sitting postures can be recorded, thereby avoiding misjudgment and improving the accuracy of the detection of the person.

[0142] In some embodiments, after recording the position of the person corresponding to the key point data of the human body, the present disclosure can also obtain the regional position data of the vehicle cabin environment, which is used to indicate the seat area; the seat area where the person is located is determined by comparing the regional position data to obtain the person occupancy result, which can indicate whether there are people and how many people are in each seat in the seat area of ​​the vehicle cabin environment. In this way, combined with the target scene, whether the driving and passengers in the car are safe, it is possible to accurately identify similar dangerous scenes such as passengers leaving their seats during vehicle driving.

[0143] The above-mentioned embodiment combines the output of the target detection model with the output of the human key point detection model to accurately identify the number and positions of drivers and passengers in the vehicle cabin environment, thereby improving the accuracy of personnel detection.

[0144] In summary, the embodiment of the present disclosure provides a method for determining a biometric detection model. The method first obtains a first training data set, which includes: multiple frames of first biometric images, and annotated images corresponding to each frame of the first biometric image, wherein the annotated image is an image annotated with one or more of the following data on the basis of the biometric image: human body data, head data, and face data; then, the initial detection model is trained based on the first training data set to obtain a biometric detection model, and then multiple frames of second biometric images collected in the vehicle cockpit environment are obtained from real-life shooting data; multiple frames of second biometric images, and the annotated images corresponding to each frame of the second biometric image are used as a second training data set, and further based on the second training data set, the model parameters of the biometric detection model are updated to obtain a target detection model. The present disclosure obtains a biometric detection model by training the first training data set obtained by artificial collection, and then obtains the second training data set using real and diverse real-life shooting data, and updates the model parameters of the biometric detection model, thereby obtaining a target detection model that meets actual needs, is suitable for actual driving scenes, has high accuracy, reduces the cost of manually obtaining training samples, and improves the training efficiency of the biometric detection model.

[0145] like Figure 2 As shown, Figure 2 A schematic diagram of a structure of a device for determining a biometric detection model is provided in an embodiment of the present disclosure. The device includes:

[0146] The acquisition module 201 is used to acquire a first training data set, wherein the first training data set includes: a plurality of frames of first biometric image, and an annotated image corresponding to each frame of the first biometric image, wherein the annotated image is an image annotated with one or more of human body data, head data, and face data based on the first biometric image;

[0147] A training module 202, configured to train an initial detection model based on a first training data set to obtain a biometric detection model, wherein the biometric detection model is configured to detect one or more of human body data, head data, and face data in an image;

[0148] The acquisition module 201 is further used to acquire multiple frames of second biometric feature images collected in the vehicle cabin environment from the real scene shooting data;

[0149] An updating module 203, configured to update the model parameters of the biometric detection model based on the second training data set to obtain a target detection model;

[0150] The second training data set includes multiple frames of second biometric image and a second annotated image corresponding to each frame of the second biometric image.

[0151] As an optional implementation of the embodiment of the present disclosure, each frame of the second biometric image is a face coded image; multiple frames of the first biometric image include partial face coded images; wherein the face coded image refers to an image in which the grayscale value of all pixels inside the face external frame is 0.

[0152] As an optional implementation of the embodiment of the present disclosure, the multiple frames of second biometric feature images include: all second biometric feature images of the target scene of interest, and at least one frame of second biometric feature image of other scenes, the target scene includes a scene of a special group of people, and / or a dangerous scene;

[0153] The acquisition module 201 acquires multiple frames of second biometric image collected in the vehicle cockpit environment from the real-scene shooting data, and is specifically used to: acquire several frames of second biometric image collected in the vehicle cockpit environment from the real-scene shooting data; determine all second biometric images of the target scene and all second biometric images of other scenes from the several frames of second biometric images; and randomly sample from all second biometric images of other scenes to obtain at least one frame of second biometric image of other scenes.

[0154] As an optional implementation of the embodiment of the present disclosure, the acquisition module 201, for the step of acquiring the face coding image, is specifically used to: after acquiring multiple frames of first biometric feature images, identify the face area included in the multiple frames of first biometric feature images;

[0155] When a face region is identified, the face region is marked to obtain a face external frame;

[0156] The grayscale value of the image within the face frame is adjusted to 0 to obtain a coded face image.

[0157] As an optional implementation of the embodiment of the present disclosure, the updating module 203 is further used to input the image to be detected into the target detection model, where the image to be detected is different from the biometric image used to train the biometric detection model; obtain one or more of the human body data, head data and face data of the image to be detected output by the target detection model; when one or more of the human body data, head data and face data matches the human body key point data, record the personnel position corresponding to the human body key point data;

[0158] Among them, the human body key point data is obtained after the model to be detected is input into the human body key point detection model, and the human body key point detection model is used to detect the human body key points in the image.

[0159] As an optional implementation of the embodiment of the present disclosure, after acquiring multiple frames of second biometric feature images collected in the vehicle cockpit environment from the real-scene shooting data, the acquisition module 201 is further used to:

[0160] Performing data cleaning on multiple frames of second biometric feature images;

[0161] On the multiple frames of second biometric image after data cleaning, one or more of the human body data, head data and face data are marked, and the target scene or other scenes are marked to obtain a marked image corresponding to each frame of the second biometric image.

[0162] In summary, the disclosed embodiments provide a device for determining a biometric detection model, wherein the device first acquires a first training data set by an acquisition module, wherein the first training data set includes: multiple frames of first biometric images, and an annotated image corresponding to each frame of the first biometric image, wherein the annotated image is an image annotated with one or more of the following data on the basis of the first biometric image: human body data, head data, and face data; then, the training module trains an initial detection model based on the first training data set to obtain a biometric detection model, and then the acquisition module acquires multiple frames of second biometric images collected in a vehicle cabin environment from real-scene shooting data; the multiple frames of second biometric images and the annotated image corresponding to each frame of the second biometric image are used as a second training data set, and further, the updating module updates the model parameters of the biometric detection model based on the second training data set to obtain a target detection model. The present invention obtains a biometric detection model by training a first training data set obtained by manual collection, and then obtains a second training data set using real and diverse real-life shooting data, and updates the model parameters of the biometric detection model, thereby obtaining a target detection model that meets actual needs, is suitable for actual driving scenarios, has high accuracy, reduces the cost of manually obtaining training samples, and improves the training efficiency of the biometric detection model.

[0163] like Figure 3 As shown, Figure 3 The present disclosure provides a schematic diagram of the structure of an electronic device, which includes: a processor, a memory, and a computer program stored in the memory and executable on the processor, and when the computer program is executed by the processor, each process of the method for determining the biometric detection model in the above method embodiment is implemented. The same technical effect can be achieved, and it will not be described here to avoid repetition.

[0164] An embodiment of the present invention provides a computer-readable storage medium, on which a computer program is stored. When the computer program is executed by a processor, the various processes of the method for determining a biometric detection model in the above method embodiment are implemented, and the same technical effect can be achieved. To avoid repetition, it will not be repeated here.

[0165] The computer readable storage medium may be a read-only memory (ROM), a random access memory (RAM), a magnetic disk or an optical disk, etc.

[0166] An embodiment of the present invention provides a computer program product, which stores a computer program. When the computer program is executed by a processor, the various processes of the method for determining the biometric detection model in the above method embodiment are implemented, and the same technical effect can be achieved. To avoid repetition, it will not be repeated here.

[0167] Those skilled in the art will appreciate that the embodiments of the present disclosure may be provided as methods, systems, or computer program products. Therefore, the present disclosure may take the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware. Furthermore, the present disclosure may take the form of a computer program product implemented on one or more computer-usable storage media containing computer-usable program code.

[0168] In the present disclosure, the processor may be a central processing unit (CPU), or other general-purpose processors, digital signal processors (DSP), application-specific integrated circuits (ASIC), field-programmable gate arrays (FPGA) or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. A general-purpose processor may be a microprocessor or the processor may also be any conventional processor, etc.

[0169] In the present disclosure, memory may include non-permanent memory in a computer-readable medium, random access memory (RAM) and / or non-volatile memory in the form of read-only memory (ROM) or flash RAM. Memory is an example of a computer-readable medium.

[0170] In the present disclosure, computer-readable media include permanent and non-permanent, removable and non-removable storage media. Storage media can be implemented by any method or technology to store information, and the information can be computer-readable instructions, data structures, modules of programs or other data. Examples of computer storage media include, but are not limited to, phase change memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory or other memory technology, read-only compact disk read-only memory (CD-ROM), digital versatile disk (DVD) or other optical storage, magnetic cassettes, disk storage or other magnetic storage devices or any other non-transmission media that can be used to store information that can be accessed by a computing device. According to the definition in this article, computer-readable media does not include temporary computer-readable media (transitory media), such as modulated data signals and carrier waves.

[0171] It should be noted that, in this article, relational terms such as "first" and "second" are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Moreover, the terms "include", "comprise" or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device. In the absence of further restrictions, the elements defined by the sentence "comprise a ..." do not exclude the existence of other identical elements in the process, method, article or device including the elements.

[0172] The above are only specific embodiments of the present disclosure, so that those skilled in the art can understand or implement the present disclosure. Various modifications to these embodiments will be apparent to those skilled in the art, and the general principles defined herein can be implemented in other embodiments without departing from the spirit or scope of the present disclosure. Therefore, the present disclosure will not be limited to these embodiments herein, but will conform to the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A method for determining a biometric detection model, characterized in that: include: Acquire a first training data set, wherein the first training data set includes: a plurality of frames of first biometric image, and an annotated image corresponding to each frame of the first biometric image, wherein the annotated image is an image annotated with one or more of human body data, head data, and face data based on the biometric image; Based on the first training data set, an initial detection model is trained to obtain a biometric detection model, where the biometric detection model is used to detect one or more of human body data, head data, and face data in an image; Acquire multiple frames of second biometric feature images collected in the vehicle cabin environment from real-scene shooting data; Based on a second training data set, the model parameters of the biometric detection model are updated to obtain a target detection model; wherein the second training data set includes: the multiple frames of second biometric images, and annotated images corresponding to each frame of the second biometric image.

2. The method according to claim 1, characterized in that Each frame of the second biometric image is a face coded image; the multiple frames of the first biometric image include some face coded images; wherein the face coded image refers to an image in which the grayscale value of all pixels inside the face external frame is 0.

3. The method according to claim 1, characterized in that: The multiple frames of second biometric feature images include: all second biometric feature images of target scenes, and at least one frame of second biometric feature image of other scenes, wherein the target scenes include scenes of special groups of people and / or dangerous scenes; The step of acquiring a plurality of frames of second biometric feature images collected in the vehicle cockpit environment from the real-scene shooting data includes: Acquire a plurality of frames of second biometric feature images collected in the vehicle cabin environment from the real-scene shooting data; Determine all second biometric images of the target scene and all second biometric images of other scenes from the plurality of frames of second biometric images; Random sampling is performed from all the second biometric feature images of the other scenes to obtain at least one frame of the second biometric feature image of the other scenes.

4. The method according to claim 2, characterized in that: The step of obtaining the face coding image comprises: After acquiring the plurality of frames of the first biometric feature image, identifying the face area included in the plurality of frames of the first biometric feature image; When the face region is identified, marking the face region to obtain a face circumscribed frame; The grayscale value of the image within the face circumscribed frame is adjusted to 0 to obtain the face coded image.

5. The method according to any one of claims 1 to 4, characterized in that: After updating the model parameters of the biometric detection model based on the second training data set to obtain the target detection model, the method includes: Inputting an image to be detected into the object detection model, wherein the image to be detected is different from a biometric image used to train the biometric detection model; Obtain one or more of the human body data, head data and face data of the image to be detected output by the target detection model; When one or more of the human body data, the head data and the face data match the human body key point data, recording the person position corresponding to the human body key point data; Among them, the human body key point data is obtained after the model to be detected is input into a human body key point detection model, and the human body key point detection model is used to detect human body key points in an image.

6. The method according to claim 1, characterized in that After acquiring a plurality of frames of second biometric feature images collected in the vehicle cabin environment from the real-scene shooting data, and before updating the model parameters of the biometric feature detection model based on the second training data set to obtain the target detection model, the method further includes: Performing data cleaning on the plurality of frames of second biometric feature images; On the multiple frames of the second biometric image after data cleaning, one or more of the human body data, head data and face data are marked, and the target scene or other scenes are marked to obtain a marked image corresponding to each frame of the second biometric image.

7. A device for determining a biometric detection model, characterized in that: include: An acquisition module is used to acquire a first training data set, wherein the first training data set includes: a plurality of frames of first biometric image, and an annotated image corresponding to each frame of the first biometric image, wherein the annotated image is an image annotated with one or more of human body data, head data, and face data based on the biometric image; A training module, used for training an initial detection model based on the first training data set to obtain a biometric detection model, wherein the biometric detection model is used to detect one or more of human body data, head data and face data in an image; The acquisition module is also used to acquire multiple frames of second biometric feature images collected in the vehicle cabin environment from the real-scene shooting data; An updating module, configured to update the model parameters of the biometric detection model based on the second training data set to obtain a target detection model; The second training data set includes the multiple frames of second biometric image and annotated images corresponding to each frame of the second biometric image.

8. An electronic device, characterized in that: include: A processor, a memory, and a computer program stored in the memory and executable on the processor, wherein the computer program, when executed by the processor, implements the method for determining a biometric detection model as described in any one of claims 1 to 6.

9. A computer-readable storage medium, characterized in that: include: The computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the method for determining a biometric detection model according to any one of claims 1 to 6 is implemented.

10. A vehicle, characterized in that: include: The device for determining a biometric detection model as claimed in claim 7, or the electronic device as claimed in claim 8.