Three-dimensional object recognition method and apparatus

Abstract

The disclosure provides a three-dimensional object recognition method and device, the method comprising: projecting a binary stripe to a target three-dimensional object; using a camera to shoot the target three-dimensional object, adjusting the focusing position of the camera to obtain a sinusoidal stripe image corresponding to the binary stripe; using a three-dimensional structured light method to determine the depth data of the target three-dimensional object according to the sinusoidal stripe image; and using a machine learning model to identify the identity of the target three-dimensional object according to the depth data of the target three-dimensional object. The technical problem of how to realize multi-distance recognition of three-dimensional objects and optimize the detection cost and detection accuracy of three-dimensional object information can be solved, the measurement progress and detection distance adaptability can be improved, the hardware cost can be reduced, and the technical effect of how to realize multi-distance recognition of three-dimensional objects and optimize the detection cost and detection accuracy of three-dimensional object information can be achieved.

Description

Technical Field

[0001] This disclosure relates to the field of object recognition technology, and in particular to a three-dimensional object recognition method and apparatus. Background Technology

[0002] When performing 3D object recognition, it is necessary to obtain the object's 3D information. Existing technologies typically use 3D structured light technology to measure the depth information of 3D objects. However, traditional phase-based 3D contour measurement systems suffer from problems such as poor accuracy, limited detection distance, and high detection costs.

[0003] Therefore, how to achieve multi-distance recognition of 3D objects and optimize the detection cost and accuracy of 3D object information is a technical problem that urgently needs to be solved. Summary of the Invention

[0004] The purpose of this disclosure is to provide a three-dimensional object recognition method to at least solve the technical problems in related technologies of how to achieve multi-distance recognition of three-dimensional objects and optimize the detection cost and accuracy of object three-dimensional information.

[0005] According to one aspect of the present disclosure, a three-dimensional object recognition method is provided, comprising:

[0006] Project binarized stripes onto the target 3D object;

[0007] The target 3D object is photographed using a camera, and the focus position of the camera is adjusted to obtain a sinusoidal fringe image corresponding to the binarized fringe.

[0008] Based on the sinusoidal fringe image, the depth data of the target three-dimensional object is determined using a three-dimensional structured light method;

[0009] Based on the depth data of the target 3D object, a machine learning model is used to identify the identity of the target 3D object.

[0010] In one exemplary embodiment, a camera is used to capture an image of the target three-dimensional object, and the focus position of the camera is adjusted to obtain a sinusoidal fringe image corresponding to the binarized fringe, including:

[0011] Obtain the distance information between the target 3D object and the camera;

[0012] Obtain the phase difference of each phase-focused PDAF pixel corresponding to the target 3D object;

[0013] A preset mapping relationship is obtained, which is used to determine the focus position of the camera corresponding to the sinusoidal stripe image based on the distance information;

[0014] Based on the mapping relationship, the distance information, and the phase difference of each PDAF pixel corresponding to the target 3D object, the camera lens is driven to the focusing position that forms the sinusoidal fringe image.

[0015] In one exemplary embodiment, obtaining a preset mapping relationship includes:

[0016] At different shooting distances, PDAF technology is used to focus on the calibration object to obtain the phase difference of each PDAF pixel corresponding to the calibration object and the corresponding focus position of the camera.

[0017] The mapping relationship between the calibrated shooting distance, the phase difference of the PDAF pixels, and the focus position of the camera.

[0018] In one exemplary embodiment, identifying the identity of the target 3D object using a machine learning model based on the object's depth data includes:

[0019] The spectral information of the reflected light from the target three-dimensional object is obtained, wherein the reflected light is formed by the reflection of the multispectral light source by the target three-dimensional object;

[0020] Based on the depth data of the target 3D object and the spectral information of the reflected light from the target 3D object, a machine learning model is used to identify the identity of the target 3D object.

[0021] In one exemplary embodiment, based on the depth data of the target three-dimensional object and the spectral information of the reflected light from the target three-dimensional object, a machine learning model is used to identify the identity of the target three-dimensional object, including:

[0022] By comparing the spectral information of the multispectral light source with the spectral information of the reflected light, the spectral absorption information of the target three-dimensional object is determined.

[0023] Based on the spectral absorption information of the target three-dimensional object, determine the material information of the target three-dimensional object;

[0024] Based on the depth data and spectral absorption information of the target 3D object, a machine learning model is used to identify the identity of the target 3D object.

[0025] In one exemplary embodiment, it further includes:

[0026] The machine learning model is generated based on the training samples;

[0027] The training samples are a set of pre-annotated training data consisting of depth data and spectral information of reflected light from several three-dimensional objects; the annotation content of the training samples includes: several facial identities and object attributes.

[0028] In one exemplary embodiment, projecting binarized fringes onto a target three-dimensional object includes: projecting the binarized fringes onto the target three-dimensional object using a polarized light source.

[0029] In one exemplary embodiment, projecting binarized stripes onto a target three-dimensional object includes:

[0030] The binarized stripes are projected using a MEMS galvanometer projector module with an integrated light source.

[0031] In one exemplary embodiment, projecting binarized stripes onto a target three-dimensional object includes:

[0032] A stripeless pattern is projected onto the target 3D object, and the size of the target 3D object is pre-determined using a target detection model.

[0033] Obtain the distance information between the target 3D object and the camera;

[0034] Based on the distance information and the size of the target 3D object, the period width and the number of stripe groups of the binarized stripes are set;

[0035] Binary stripes are projected onto the target 3D object.

[0036] In one exemplary embodiment, projecting binarized stripes onto a target three-dimensional object includes:

[0037] If the distance information and the size of the target 3D object are within a predetermined range, the encoding rule for the binarized stripes is set to Gray code.

[0038] According to another aspect of the present disclosure, a three-dimensional object recognition device is also provided, comprising:

[0039] The projection module is used to project binary stripes onto the target 3D object;

[0040] The focusing module is used to capture the target three-dimensional object with a camera, adjust the focusing position of the camera, and obtain a sinusoidal fringe image corresponding to the binarized fringe.

[0041] The depth module is used to determine the depth data of the target three-dimensional object using a three-dimensional structured light method based on the sinusoidal fringe image.

[0042] The identification module is used to identify the identity of the target 3D object using a machine learning model based on the depth data of the target 3D object.

[0043] In one exemplary embodiment, the focusing module includes:

[0044] The distance information submodule is used to obtain distance information between the target 3D object and the camera;

[0045] The phase difference submodule is used to obtain the phase difference of each phase-focused PDAF pixel corresponding to the target 3D object;

[0046] The mapping submodule is used to obtain a preset mapping relationship, which is used to determine the focus position of the camera corresponding to the sinusoidal stripe image based on the distance information.

[0047] The driving submodule is used to drive the lens of the camera to the focusing position that forms the sinusoidal fringe image, based on the mapping relationship, the distance information, and the phase difference of each PDAF pixel corresponding to the target three-dimensional object.

[0048] In one exemplary embodiment, the mapping subunit includes:

[0049] The acquisition unit is used to focus on the calibration object using PDAF technology at different shooting distances, and to acquire the phase difference of each PDAF pixel corresponding to the calibration object and the corresponding focus position of the camera.

[0050] The calibration unit is used to calibrate the mapping relationship between the shooting distance, the phase difference of the PDAF pixels, and the focus position of the camera.

[0051] In one exemplary embodiment,

[0052] The identification module includes:

[0053] The spectral submodule is used to acquire the spectral information of the reflected light from the target three-dimensional object, wherein the reflected light is formed by the reflection of the multispectral light source from the target three-dimensional object;

[0054] The identification submodule is used to identify the target 3D object by using a machine learning model based on the depth data of the target 3D object and the spectral information of the reflected light of the target 3D object.

[0055] In one exemplary embodiment, the identification submodule includes:

[0056] An absorption determination unit is used to compare the spectral information of the multispectral light source with the spectral information of the reflected light to determine the spectral absorption information of the target three-dimensional object.

[0057] The material determination unit is used to determine the material information of the target three-dimensional object based on the spectral absorption information of the target three-dimensional object.

[0058] The identification unit is used to identify the identity of the target three-dimensional object using a machine learning model based on the depth data and spectral absorption information of the target three-dimensional object.

[0059] In one exemplary embodiment, the three-dimensional object recognition device further includes:

[0060] The training module is used to train and generate the machine learning model based on the training samples;

[0061] The training samples are a set of pre-annotated training data consisting of depth data and spectral information of reflected light from several three-dimensional objects; the annotation content of the training samples includes: several facial identities and object attributes.

[0062] In one exemplary embodiment, the projection module includes:

[0063] A polarizer module is used to project the binarized stripes onto the target 3D object using a polarized light source.

[0064] In one exemplary embodiment, the projection module includes:

[0065] A galvanometer module is used to project the binarized stripes using a MEMS galvanometer projector module with an integrated light source.

[0066] In one exemplary embodiment, the projection module includes:

[0067] The pre-judgment submodule is used to project a stripeless pattern onto the target 3D object and use a target detection model to pre-judge the size of the target 3D object.

[0068] The distance information submodule is used to obtain distance information between the target 3D object and the camera;

[0069] The stripe setting submodule is used to set the period width and the number of stripe groups of the binarized stripes based on the distance information and the size of the target three-dimensional object;

[0070] The projection submodule is used to project binary stripes onto the target 3D object.

[0071] In one exemplary embodiment, the projection module includes:

[0072] The Gray code submodule is used to set the encoding rule of the binarized stripes to Gray code when the distance information and the size of the target three-dimensional object are within a predetermined range.

[0073] According to another aspect of the present disclosure, a computer device is also provided, the computer device comprising: a processor adapted to implement various instructions and a storage device storing a plurality of instructions adapted to be loaded by the processor and executed any of the above-described three-dimensional object recognition methods.

[0074] According to another aspect of the present disclosure, a computer-readable storage medium is also provided, the computer-readable storage medium storing a computer program for performing any of the above-described three-dimensional object recognition methods.

[0075] According to another aspect of the present disclosure, there is also a computer program product comprising a computer program that, when executed by a processor, implements any of the above-described three-dimensional object recognition methods.

[0076] The technical solutions of this disclosure will be further described in detail below with reference to the accompanying drawings and embodiments. Attached Figure Description

[0077] The following figures are intended only to illustrate and explain this disclosure and do not limit the scope of this disclosure. Wherein:

[0078] Figure 1 This is a flowchart of a three-dimensional object recognition method according to embodiments of the present disclosure; and,

[0079] Figure 2 This is a schematic diagram of a hardware solution according to an optional embodiment of the present disclosure; and,

[0080] Figure 3 This is a schematic diagram of a three-dimensional object recognition device according to an embodiment of the present disclosure; and,

[0081] Figure 4 This is a schematic diagram of the structure of a computer device according to an embodiment of this disclosure. Detailed Implementation

[0082] The present application will now be described in further detail with reference to the accompanying drawings and embodiments. Through these descriptions, the features and advantages of the present application will become clearer and more apparent.

[0083] The term “exemplary” as used herein means “serving as an example, embodiment, or illustration.” Any embodiment illustrated herein as “exemplary” is not necessarily to be construed as superior to or better than other embodiments. Although various aspects of embodiments are shown in the accompanying drawings, the drawings are not necessarily drawn to scale unless specifically indicated otherwise.

[0084] Furthermore, the technical features involved in the different embodiments of this application described below can be combined with each other as long as they do not conflict with each other.

[0085] This embodiment provides a method for three-dimensional object recognition. Figure 1 This is a flowchart of a three-dimensional object recognition method according to an embodiment of the present disclosure, such as... Figure 1 As shown, it includes:

[0086] Step 102: Project binarized stripes onto the target 3D object.

[0087] The aforementioned target 3D object is the 3D object that needs to be identified, which can be a human face or other objects. The aforementioned binarized stripes can be periodic stripes with alternating bright and dark areas.

[0088] Step 104: Use a camera to photograph the target 3D object, adjust the focus position of the camera, and obtain the sinusoidal fringe image corresponding to the binarized fringe.

[0089] Step 104 above can be achieved by, when the target 3D object with the aforementioned binarized stripes is captured by the aforementioned camera, obtaining the aforementioned sinusoidal stripe image with the binarized stripes blurred at the receiving end of the camera by driving the lens to defocus or focus. The aforementioned focus position can be the focal point of the lens. The aforementioned sinusoidal stripe image can be stripes with continuous brightness changes.

[0090] Step 106: Based on the above sinusoidal fringe image, use the three-dimensional structured light method to determine the depth data of the target three-dimensional object.

[0091] In the measurement system of the aforementioned three-dimensional structured light method, the projection device projects the aforementioned binarized fringes, and the aforementioned camera serves as the receiving device of the measurement system. Furthermore, the depth data of the aforementioned target three-dimensional object can be measured based on the multi-step phase-shifting method and the aforementioned sinusoidal fringe image.

[0092] Step 108: Based on the depth data of the target 3D object, use a machine learning model to identify the identity of the target 3D object.

[0093] The aforementioned machine learning model can be a model trained based on deep learning algorithms. The identity of the target 3D object can be its category, properties, corresponding unique identifier, etc.

[0094] In this embodiment, binarized stripes are projected onto the target 3D object through steps 102 to 108. A camera is used to capture an image of the target 3D object, and the focus position of the camera is adjusted to obtain a sinusoidal stripe image corresponding to the binarized stripes. Based on the sinusoidal stripe image, a 3D structured light method is used to determine the depth data of the target 3D object. Based on the depth data of the target 3D object, a machine learning model is used to identify the identity of the target 3D object. This solves the technical problem in related technologies of how to achieve multi-distance recognition of 3D objects and optimize the detection cost and accuracy of 3D object information. By adjusting the camera focal length to obtain a sinusoidal stripe image, and using this sinusoidal stripe image in a 3D structured light technology depth data measurement scheme for target 3D objects, the measurement progress and detection distance adaptability are improved, hardware costs are reduced, and the technical effect of achieving multi-distance recognition of 3D objects and optimizing the detection cost and accuracy of 3D object information is achieved.

[0095] In one exemplary embodiment, step 104 above includes:

[0096] Obtain the distance information between the aforementioned target 3D object and the aforementioned camera.

[0097] This distance information corresponds to the shooting distance in the calibration process described above and is used to incorporate the mapping relationship.

[0098] Obtain the phase difference of each phase-detection autofocus (PDAF) pixel corresponding to the above-mentioned target 3D object.

[0099] The phase difference of each PDAF pixel corresponding to the target 3D object corresponds to the phase difference in the above calibration process and is used to input the above mapping relationship.

[0100] A preset mapping relationship is obtained, which is used to determine the focus position of the camera corresponding to the sinusoidal stripe image based on the distance information.

[0101] The focus position corresponding to the above sinusoidal fringe image can be predetermined.

[0102] Based on the above mapping relationship, the above distance information, and the phase difference of each PDAF pixel corresponding to the above target 3D object, the lens of the above camera is driven to the focusing position that forms the above sinusoidal fringe image.

[0103] The focal position of the aforementioned sinusoidal fringe image can be based on the focal length difference or lens variation corresponding to a pre-established defocus ratio.

[0104] For example, the correspondence between the phase difference of the PDAF pixels and the liquid crystal zoom lens is pre-calibrated, the ROI (Region of Interest) where the target three-dimensional object is located is detected, the phase difference of the PDAF pixels of each pixel in the region and the distance between the target three-dimensional object and the liquid crystal lens are obtained, and then the lens deformation of the sinusoidal fringe image corresponding to the shooting distance is determined according to the deformation of the liquid crystal zoom lens corresponding to different shooting distances.

[0105] Through the above embodiments, using a pre-calibrated mapping relationship, the focal position of the camera lens is determined when the distance between the target 3D object and the camera is determined, and the above sinusoidal fringe image is formed. This achieves the technical effect of driving the camera to the focal position, obtaining the corresponding defocus / focus image, and obtaining the sinusoidal fringe image at the receiving end.

[0106] In one exemplary embodiment, obtaining a preset mapping relationship includes:

[0107] At different shooting distances, PDAF technology is used to focus on the aforementioned calibration object to obtain the phase difference of each PDAF pixel corresponding to the calibration object and the corresponding focus position of the aforementioned camera.

[0108] The aforementioned shooting distance can be measured by a ranging module. This shooting distance can be the distance between the camera and the target object. Alternatively, it can be measured by a TOF (Time of Flight) sensor. The calibration object can be the target 3D object. The phase difference of the PDAF pixels can be the phase difference between a set of left and right phase detection pixels corresponding to a pointer on the calibration object.

[0109] The mapping relationship between the calibrated shooting distance, the phase difference of PDAF pixels, and the focus position of the aforementioned camera.

[0110] The mapping relationship described above can be the correspondence between the phase difference of the PDAF pixels and the focus position of the camera at different shooting distances. The focus position of the camera can be expressed using the defocus rate.

[0111] Through the above embodiments, the mapping relationship between shooting distance, phase difference of PDAF pixels, and focus position of the camera is pre-calibrated, thereby determining the focus position of the camera when shooting objects at different distances, so as to quickly adjust the camera according to the shooting distance in the future.

[0112] Optionally, the aforementioned camera can be a liquid crystal zoom lens. Furthermore, the deformation of the liquid crystal lens is used as the focal position of the aforementioned camera and the phase difference between the aforementioned shooting distance and the PDAF pixels for calibration.

[0113] In one exemplary embodiment,

[0114] Step 108 above includes:

[0115] The spectral information of the reflected light from the aforementioned three-dimensional target object is obtained. The reflected light is formed by the reflection of the aforementioned multispectral light source by the aforementioned three-dimensional target object.

[0116] The aforementioned multispectral light source can be a projector that emits light at multiple wavelengths.

[0117] The spectral information of the reflected light can be the light intensity information of each wavelength of light after the multispectral light source is reflected by the target three-dimensional object.

[0118] Based on the depth data of the aforementioned target 3D object and the spectral information of the reflected light from the aforementioned target 3D object, a machine learning model is used to identify the identity of the aforementioned target 3D object.

[0119] The identity of the aforementioned target 3D object may also include its material, color, etc.

[0120] Through the above embodiments, by using a multispectral light source to project the aforementioned binarized stripes, while measuring the depth data of the aforementioned target three-dimensional object, the spectral information of the binarized stripes projected by the multispectral light source after reflection by the target three-dimensional object is detected, and then the spectral information is used as a reference factor for identifying the aforementioned target three-dimensional object, thereby achieving the effect of distinguishing the material of the target three-dimensional object.

[0121] In one exemplary embodiment, based on the depth data of the target three-dimensional object and the spectral information of the reflected light from the target three-dimensional object, a machine learning model is used to identify the identity of the target three-dimensional object, including:

[0122] By comparing the spectral information of the multispectral light source and the spectral information of the reflected light, the spectral absorption information of the target three-dimensional object is determined.

[0123] The spectral absorption information of the aforementioned target three-dimensional object can be the intensity difference between the spectral information of the aforementioned multispectral light source and the spectral information of the aforementioned reflected light at various wavelengths.

[0124] Based on the spectral absorption information of the target three-dimensional object, the material information of the target three-dimensional object is determined.

[0125] It can be done by pre-calibrating the absorption of each material object for each wavelength of light, and then determining the corresponding material information based on the above spectral absorption information.

[0126] Based on the depth data and spectral absorption information of the aforementioned target 3D object, a machine learning model is used to identify the identity of the aforementioned target 3D object.

[0127] Through the above embodiments, by comparing the changes in spectral information of the multispectral light source before and after absorption by the target three-dimensional object, and based on the changes in spectral information before and after absorption by the target three-dimensional object, the identity of the target three-dimensional object is identified, thereby achieving the technical effect of distinguishing the material of the target three-dimensional object.

[0128] In one exemplary embodiment, it further includes:

[0129] The aforementioned machine learning model was generated based on the training samples.

[0130] The aforementioned machine learning model can be trained based on deep learning algorithms.

[0131] The training samples mentioned above are a set of pre-annotated training data consisting of depth data and spectral information of reflected light from several three-dimensional objects; the annotation content of the training samples includes: several facial identities and object attributes.

[0132] Optionally, without performing object material recognition, the above machine learning model can be pre-trained using spectral information of unlabeled reflected light and training samples whose labels do not include object attributes.

[0133] Through the above embodiments, the machine learning model is trained by using a training data set containing the depth data and spectral information of reflected light of several pre-annotated three-dimensional objects as samples. This reduces the false detection rate of the machine learning model and further improves the accuracy of the three-dimensional object recognition results.

[0134] In an exemplary embodiment, step 102 includes: projecting the binarized stripes onto the target three-dimensional object using a polarized light source.

[0135] Through the above embodiments, by using a polarized light source to project the aforementioned binarized stripes, the technical effects of enhancing texture and removing reflections can be achieved, further reducing the impact of the reflective factors of the surface material of the aforementioned three-dimensional object on the recognition results.

[0136] In one exemplary embodiment, step 102 above includes:

[0137] The aforementioned binarized stripes are projected using a MEMS (Micro-Electro-Mechanical System) galvanometer projector module with an integrated light source.

[0138] Through the above embodiments, the aforementioned binary stripes are generated by the MEMS galvanometer projector module, which can replace the floodlight in the prior art, thereby further reducing hardware costs.

[0139] In one exemplary embodiment, step 102 above includes:

[0140] A stripeless pattern is projected onto the aforementioned target 3D object, and the size of the target 3D object is pre-determined using a target detection model.

[0141] The aforementioned target detection model can be a pre-trained model capable of recognizing objects of the same type as the aforementioned 3D target object. The size of the aforementioned 3D target object can be determined based on a pre-defined relationship between the type and size of the 3D target object.

[0142] Obtain the distance information between the aforementioned target 3D object and the aforementioned camera.

[0143] Based on the distance information and the size of the target 3D object, the period width and the number of stripe groups of the binarized stripes are set.

[0144] The period width of the aforementioned binarized fringes can be the total width of a set of fringes that alternate between bright and dark. The number of fringe sets can be the total number of fringe sets projected onto the aforementioned target 3D object.

[0145] Binary stripes are projected onto the aforementioned target 3D object.

[0146] For example, the target 3D object is a human face, approximately 20cm wide. To achieve a stripe density (i.e., number of stripe groups) of approximately 20 groups at different distances, binarized stripes with different periods are needed at different distances. Optionally, the stripe period and number of groups can be specifically set according to the required accuracy for the scene.

[0147] Through the above embodiments, the stripe density of the binarized stripes is specifically set according to the size of the target 3D object and the shooting distance, ensuring that there are a sufficient number of binarized stripes on the target 3D object, thereby further improving the distance adaptability in the 3D object recognition process.

[0148] In one exemplary embodiment, step 102 above includes:

[0149] If the distance information and the size of the target 3D object are within a predetermined range, the encoding rule for the binarized stripes is set to Gray code.

[0150] For example, by pre-setting the accuracy requirements for recognition, as well as the applicable object size and shooting distance, the aforementioned binarized stripes can be set as Gray code stripes, and the step of blurring the binarized stripes to obtain a sinusoidal stripe image can be omitted.

[0151] Through the above embodiments, under the premise that the accuracy requirements, shooting distance and object size meet the predetermined range, the use of Gray code stripes for three-dimensional object recognition simplifies the recognition scheme and achieves the technical effect of improving recognition speed.

[0152] Figure 2 This is a schematic diagram of a hardware solution according to an optional embodiment of the present disclosure, such as... Figure 2 As shown, it includes: a main control chip 201, a MEMS galvanometer projector module 202, a camera 203 with a PDAF sensor, and a linear array ranging TOF module 204.

[0153] Among them, the MEMS galvanometer projector module can use a polarized light source to reduce the specular flooding on the surface of the target 3D object and avoid the loss of surface information of the target 3D object; it can also use a multi-wavelength light source according to the needs of the target 3D object and the scene to obtain the spectral absorption information of the target 3D object.

[0154] Among them, the camera with a PDAF sensor can be used to obtain the phase difference information of the ROI where the target 3D object is located.

[0155] Among them, the linear array ranging TOF module can be used to obtain distance information between the target 3D object and the camera.

[0156] The main control chip receives and processes the above information, controls the camera to defocus / focus based on the above method to obtain a sinusoidal fringe image, measures the depth data of the target three-dimensional object, and uses a machine learning model to perform three-dimensional object recognition.

[0157] According to another aspect of the embodiments of this disclosure, a three-dimensional object recognition device is also provided. Figure 3 This is a schematic diagram of the structure of a three-dimensional object recognition device according to an embodiment of the present disclosure, such as... Figure 3 As shown, it includes:

[0158] Projection module 301 is used to project binary stripes onto the target 3D object.

[0159] The aforementioned target 3D object is the 3D object that needs to be identified, which can be a human face or other objects. The aforementioned binarized stripes can be periodic stripes with alternating bright and dark areas.

[0160] The focusing module 302 is used to capture the above-mentioned target three-dimensional object with a camera, adjust the focusing position of the camera, and obtain the sinusoidal fringe image corresponding to the above-mentioned binarized fringe.

[0161] The aforementioned focusing module can be used to, when photographing the aforementioned three-dimensional target object with the aforementioned binarized stripes using the aforementioned camera, drive the lens to defocus or focus, and obtain the aforementioned sinusoidal stripe image with blurred binarized stripes at the receiving end of the camera. The aforementioned focusing position can be the focal point of the lens. The aforementioned sinusoidal stripe image can be stripes with continuous brightness changes.

[0162] The depth module 303 is used to determine the depth data of the target three-dimensional object using a three-dimensional structured light method based on the sinusoidal fringe image described above.

[0163] In the aforementioned three-dimensional structured light measurement system, the projection device projects the aforementioned binary stripes, and the aforementioned camera serves as the receiving device of the measurement system. Furthermore, the depth data of the aforementioned target three-dimensional object can be measured based on a multi-step phase-shifting method and the aforementioned sinusoidal fringe image.

[0164] The identification module 304 is used to identify the identity of the target 3D object by using a machine learning model based on the depth data of the target 3D object.

[0165] The aforementioned machine learning model can be a model trained based on deep learning algorithms. The identity of the target 3D object can be its category, properties, corresponding unique identifier, etc.

[0166] In this embodiment, the aforementioned device projects binarized stripes onto the target 3D object. A camera is used to capture an image of the target 3D object, and the camera's focus position is adjusted to obtain a sinusoidal stripe image corresponding to the binarized stripes. Based on the sinusoidal stripe image, a 3D structured light method is used to determine the depth data of the target 3D object. Based on the depth data of the target 3D object, a machine learning model is used to identify the identity of the target 3D object. This method solves the technical problem of how to achieve multi-distance recognition of 3D objects and optimize the detection cost and accuracy of 3D object information in related technologies. By adjusting the camera's focal length to obtain a sinusoidal stripe image, and using this image in a 3D structured light technology depth data measurement scheme for target 3D objects, the measurement progress and detection distance adaptability are improved, hardware costs are reduced, and the technical effect of achieving multi-distance recognition of 3D objects and optimizing the detection cost and accuracy of 3D object information is achieved.

[0167] In one exemplary embodiment, the above-described three-dimensional object recognition device further includes:

[0168] The acquisition module is used to focus on the calibration object using PDAF technology at different shooting distances, and to acquire the phase difference of each PDAF pixel corresponding to the calibration object and the corresponding focus position of the camera.

[0169] The aforementioned shooting distance can be measured by a ranging module. This shooting distance can be the distance between the camera and the target object. Alternatively, it can be measured by a TOF (Time of Flight) sensor. The calibration object can be the target 3D object. The phase difference of the PDAF pixels can be the phase difference between a set of left and right phase detection pixels corresponding to a pointer on the calibration object.

[0170] The calibration module is used to calibrate the mapping relationship between the shooting distance, the phase difference of the PDAF pixels, and the focus position of the aforementioned camera.

[0171] The mapping relationship described above can be the correspondence between the phase difference of the PDAF pixels and the focus position of the camera at different shooting distances. The focus position of the camera can be expressed using the defocus rate.

[0172] The above mapping relationship is used to determine the focus position of the camera corresponding to the above sinusoidal fringe image at different shooting distances.

[0173] The focus position corresponding to the above sinusoidal fringe image can be predetermined.

[0174] Through the above embodiments, the mapping relationship between shooting distance, phase difference of PDAF pixels, and focus position of the camera is pre-calibrated, thereby determining the focus position of the camera when shooting objects at different distances, so as to quickly adjust the camera according to the shooting distance in the future.

[0175] Optionally, the aforementioned camera can be a liquid crystal zoom lens. Furthermore, the deformation of the liquid crystal lens is used as the focal position of the aforementioned camera and the phase difference between the aforementioned shooting distance and the PDAF pixels for calibration.

[0176] In one exemplary embodiment, the focusing module includes:

[0177] The distance information submodule is used to obtain the distance information between the aforementioned target 3D object and the aforementioned camera.

[0178] This distance information corresponds to the shooting distance in the calibration process described above and is used to incorporate the mapping relationship.

[0179] The phase difference submodule is used to obtain the phase difference of each phase-focused autofocus (PDAF) element corresponding to the above-mentioned target 3D object.

[0180] The phase difference of each PDAF pixel corresponding to the target 3D object corresponds to the phase difference in the above calibration process and is used to input the above mapping relationship.

[0181] The mapping submodule is used to obtain a preset mapping relationship, which is used to determine the focus position of the camera corresponding to the sinusoidal stripe image based on the distance information.

[0182] The focus position corresponding to the above sinusoidal fringe image can be predetermined.

[0183] The driving submodule is used to drive the lens of the camera to the focusing position that forms the sinusoidal fringe image, based on the above mapping relationship, the above distance information, and the phase difference of each PDAF pixel corresponding to the above target three-dimensional object.

[0184] The focal position of the aforementioned sinusoidal fringe image can be based on the focal length difference or lens variation corresponding to a pre-established defocus ratio.

[0185] For example, the correspondence between the phase difference of the PDAF pixels and the liquid crystal zoom lens is pre-calibrated, the ROI (Region of Interest) where the target three-dimensional object is located is detected, the phase difference of the PDAF pixels of each pixel in the region and the distance between the target three-dimensional object and the liquid crystal lens are obtained, and then the lens deformation of the sinusoidal fringe image corresponding to the shooting distance is determined according to the deformation of the liquid crystal zoom lens corresponding to different shooting distances.

[0186] Through the above embodiments, using a pre-calibrated mapping relationship, the focal position of the camera lens is determined when the distance between the target 3D object and the camera is determined, and the above sinusoidal fringe image is formed. This achieves the technical effect of driving the camera to the focal position, obtaining the corresponding defocus / focus image, and obtaining the sinusoidal fringe image at the receiving end.

[0187] In one exemplary embodiment, the above-mentioned mapping subunit includes:

[0188] The acquisition unit is used to focus on the calibration object using PDAF technology at different shooting distances, and to acquire the phase difference of each PDAF pixel corresponding to the calibration object and the corresponding focus position of the camera.

[0189] The aforementioned shooting distance can be measured by a ranging module. This shooting distance can be the distance between the camera and the target object. Alternatively, it can be measured by a TOF (Time of Flight) sensor. The calibration object can be the target 3D object. The phase difference of the PDAF pixels can be the phase difference between a set of left and right phase detection pixels corresponding to a pointer on the calibration object.

[0190] The calibration unit is used to calibrate the mapping relationship between the shooting distance, the phase difference of the PDAF pixels, and the focus position of the aforementioned camera.

[0191] The mapping relationship described above can be the correspondence between the phase difference of the PDAF pixels and the focus position of the camera at different shooting distances. The focus position of the camera can be expressed using the defocus rate.

[0192] Through the above embodiments, the mapping relationship between shooting distance, phase difference of PDAF pixels, and focus position of the camera is pre-calibrated, thereby determining the focus position of the camera when shooting objects at different distances, so as to quickly adjust the camera according to the shooting distance in the future.

[0193] Optionally, the aforementioned camera can be a liquid crystal zoom lens. Furthermore, the deformation of the liquid crystal lens is used as the focal position of the aforementioned camera and the phase difference between the aforementioned shooting distance and the PDAF pixels for calibration.

[0194] In one exemplary embodiment, the identification module includes:

[0195] The spectral submodule is used to acquire the spectral information of the reflected light from the aforementioned target three-dimensional object, which is formed by the reflection of the aforementioned multispectral light source by the aforementioned target three-dimensional object.

[0196] The aforementioned multispectral light source can be a projector that emits light at multiple wavelengths.

[0197] The spectral information of the reflected light can be the light intensity information of each wavelength of light after the multispectral light source is reflected by the target three-dimensional object.

[0198] The identification submodule is used to identify the identity of the target 3D object by using a machine learning model based on the depth data of the target 3D object and the spectral information of the reflected light of the target 3D object.

[0199] The identity of the aforementioned target 3D object may also include its material, color, etc.

[0200] Through the above embodiments, by using a multispectral light source to project the aforementioned binarized stripes, while measuring the depth data of the aforementioned target three-dimensional object, the spectral information of the binarized stripes projected by the multispectral light source after reflection by the target three-dimensional object is detected, and then the spectral information is used as a reference factor for identifying the aforementioned target three-dimensional object, thereby achieving the effect of distinguishing the material of the target three-dimensional object.

[0201] In one exemplary embodiment, the identification module includes:

[0202] The absorption determination unit is used to compare the spectral information of the multispectral light source and the spectral information of the reflected light to determine the spectral absorption information of the target three-dimensional object.

[0203] The spectral absorption information of the aforementioned target three-dimensional object can be the intensity difference between the spectral information of the aforementioned multispectral light source and the spectral information of the aforementioned reflected light at various wavelengths.

[0204] The material determination unit is used to determine the material information of the target three-dimensional object based on the spectral absorption information of the target three-dimensional object.

[0205] It can be done by pre-calibrating the absorption of each material object for each wavelength of light, and then determining the corresponding material information based on the above spectral absorption information.

[0206] The identification unit is used to identify the identity of the target three-dimensional object by using a machine learning model based on the depth data and spectral absorption information of the target three-dimensional object.

[0207] Through the above embodiments, by comparing the changes in spectral information of the multispectral light source before and after absorption by the target three-dimensional object, and based on the changes in spectral information before and after absorption by the target three-dimensional object, the identity of the target three-dimensional object is identified, thereby achieving the technical effect of distinguishing the material of the target three-dimensional object.

[0208] In one exemplary embodiment, the above-described three-dimensional object recognition device further includes:

[0209] The training module is used to train and generate the aforementioned machine learning model based on the training samples.

[0210] The aforementioned machine learning model can be trained based on deep learning algorithms.

[0211] The training samples mentioned above are a set of pre-annotated training data consisting of depth data and spectral information of reflected light from several three-dimensional objects; the annotation content of the training samples includes: several facial identities and object attributes.

[0212] Optionally, without performing object material recognition, the above machine learning model can be pre-trained using spectral information of unlabeled reflected light and training samples whose labels do not include object attributes.

[0213] Through the above embodiments, the machine learning model is trained by using a training data set containing the depth data and spectral information of reflected light of several pre-annotated three-dimensional objects as samples. This reduces the false detection rate of the machine learning model and further improves the accuracy of the three-dimensional object recognition results.

[0214] In one exemplary embodiment, the projection module includes:

[0215] The polarization module is used to project the aforementioned binary stripes onto the aforementioned target 3D object using a polarized light source.

[0216] Through the above embodiments, by using a polarized light source to project the aforementioned binarized stripes, the technical effects of enhancing texture and removing reflections can be achieved, further reducing the impact of the reflective factors of the surface material of the aforementioned three-dimensional object on the recognition results.

[0217] In one exemplary embodiment, the above

[0218] The projection module includes:

[0219] The galvanometer module is used to project the aforementioned binarized stripes using a MEMS galvanometer projector module with an integrated light source.

[0220] Through the above embodiments, the aforementioned binary stripes are generated by the MEMS galvanometer projector module, which can replace the floodlight in the prior art, thereby further reducing hardware costs.

[0221] In one exemplary embodiment, the projection module includes:

[0222] The pre-judgment submodule is used to project a stripeless pattern onto the aforementioned target 3D object and use the target detection model to pre-judge the size of the aforementioned target 3D object.

[0223] The aforementioned target detection model can be a pre-trained model capable of recognizing objects of the same type as the aforementioned 3D target object. The size of the aforementioned 3D target object can be determined based on a pre-defined relationship between the type and size of the 3D target object.

[0224] The distance information submodule is used to obtain the distance information between the aforementioned target 3D object and the aforementioned camera.

[0225] The stripe setting submodule is used to set the period width and the number of stripe groups of the binarized stripes based on the distance information and the size of the target 3D object.

[0226] The period width of the aforementioned binarized fringes can be the total width of a set of fringes that alternate between bright and dark. The number of fringe sets can be the total number of fringe sets projected onto the aforementioned target 3D object.

[0227] The projection submodule is used to project binarized stripes onto the aforementioned target 3D object.

[0228] For example, the target 3D object is a human face, approximately 20cm wide. To achieve a stripe density (i.e., number of stripe groups) of approximately 20 groups at different distances, binarized stripes with different periods are needed at different distances. Optionally, the stripe period and number of groups can be specifically set according to the required accuracy for the scene.

[0229] Through the above embodiments, the stripe density of the binarized stripes is specifically set according to the size of the target 3D object and the shooting distance, ensuring that there are a sufficient number of binarized stripes on the target 3D object, thereby further improving the distance adaptability in the 3D object recognition process.

[0230] In one exemplary embodiment, the projection module includes:

[0231] The Gray code submodule is used to set the encoding rule of the aforementioned binarized stripes to Gray code, provided that the aforementioned distance information and the size of the aforementioned target 3D object are within a predetermined range.

[0232] For example, by pre-setting the accuracy requirements for recognition, as well as the applicable object size and shooting distance, the aforementioned binarized stripes can be set as Gray code stripes, and the step of blurring the binarized stripes to obtain a sinusoidal stripe image can be omitted.

[0233] Through the above embodiments, under the premise that the accuracy requirements, shooting distance and object size meet the predetermined range, the use of Gray code stripes for three-dimensional object recognition simplifies the recognition scheme and achieves the technical effect of improving recognition speed.

[0234] This disclosure also provides a computer device. Figure 4 This is a schematic diagram of a computer device in an embodiment of the present disclosure. This computer device is capable of implementing all steps in the three-dimensional object recognition method described above. Specifically, the computer device includes the following components:

[0235] Processor 401, memory 402, communications interface 403, and communication bus 404.

[0236] The processor 401, memory 402, and communication interface 403 communicate with each other via the communication bus 404. The communication interface 403 is used to realize information transmission between related devices.

[0237] The processor 401 is used to call the computer program in the memory 402. When the processor executes the computer program, it implements the three-dimensional object recognition method in the above embodiment.

[0238] Optionally, in an embodiment, when the above-described computer program instructions are executed by the processor, the following steps are implemented:

[0239] Step S1: Project binarized stripes onto the target 3D object.

[0240] Step S2: Use a camera to photograph the target 3D object, adjust the focus position of the camera, and obtain the sinusoidal fringe image corresponding to the binarized fringe.

[0241] Step S3: Based on the sinusoidal fringe image described above, use the three-dimensional structured light method to determine the depth data of the target three-dimensional object.

[0242] Step S4: Based on the depth data of the target 3D object, use a machine learning model to identify the identity of the target 3D object.

[0243] This disclosure also provides a computer-readable storage medium storing a computer program that performs the operation of the three-dimensional object recognition method described above in response to the computer program being executed by a processor.

[0244] This disclosure also provides a computer program product, which includes a computer program that, when executed by a processor, implements the above-described three-dimensional object recognition method.

[0245] While this disclosure provides the method operation steps described in the embodiments or flowcharts, more or fewer operation steps may be included based on conventional or non-inventive labor. The order of steps listed in the embodiments is merely one possible execution order among many and does not represent the only execution order. In actual device or client product execution, the method can be executed sequentially as shown in the embodiments or drawings, or in parallel (e.g., using parallel processors or multi-threaded processing).

[0246] Those skilled in the art will understand that the embodiments of this specification can be provided as methods, apparatus (systems), or computer program products. Therefore, the embodiments of this specification can take the form of entirely hardware embodiments, entirely software embodiments, or embodiments combining software and hardware aspects. Furthermore, this disclosure can take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0247] This disclosure is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of this disclosure. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, create a machine for implementing the flowchart illustrations and / or block diagrams. Figure 1One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0248] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0249] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0250] The various embodiments in this specification are described in a progressive manner. Similar or identical parts between embodiments can be referred to mutually. Each embodiment focuses on describing the differences from other embodiments. In particular, the system embodiments are basically similar to the method embodiments, so the description is relatively simple; relevant parts can be referred to the descriptions of the method embodiments. In this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations.

[0251] It should be noted that, unless otherwise specified, the embodiments and features described in this disclosure can be combined with each other. This disclosure is not limited to any single aspect, nor to any single embodiment, nor to any combination and / or substitution of these aspects and / or embodiments. Moreover, each aspect and / or embodiment of this disclosure can be used alone or in combination with one or more other aspects and / or embodiments thereof.

[0252] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of this disclosure, and not to limit them. Although this disclosure has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some or all of the technical features. These modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of this disclosure, and all should be covered within the scope of the claims and specification of this disclosure.

Claims

1. A method for recognizing three-dimensional objects, characterized in that, include: A binary stripe is projected onto a target three-dimensional object, wherein the binary stripe is a periodic stripe with alternating bright and dark areas; Obtain the distance information between the target 3D object and the camera; The target 3D object is photographed using a camera. Based on the distance information, the focus position of the camera is adjusted to blur the binarized stripes, thereby obtaining a sinusoidal stripe image corresponding to the binarized stripes. The sinusoidal stripe image is a stripe with continuous brightness variation. Based on the sinusoidal fringe image, the depth data of the target three-dimensional object is determined using a three-dimensional structured light method; Based on the depth data of the target 3D object, a machine learning model is used to identify the identity of the target 3D object.

2. The three-dimensional object recognition method according to claim 1, characterized in that, Using a camera to capture an image of the target 3D object, and adjusting the camera's focus position based on the distance information to blur the binarized stripes, a sinusoidal stripe image corresponding to the binarized stripes is obtained, including: Obtain the phase difference of each phase-focused PDAF pixel corresponding to the target 3D object; A preset mapping relationship is obtained, which is used to determine the focus position of the camera corresponding to the sinusoidal stripe image based on the distance information; Based on the mapping relationship, the distance information, and the phase difference of each PDAF pixel corresponding to the target 3D object, the camera lens is driven to the focusing position that forms the sinusoidal fringe image.

3. The three-dimensional object recognition method according to claim 2, characterized in that, Obtain the preset mapping relationship, including: At different shooting distances, PDAF technology is used to focus on the calibration object to obtain the phase difference of each PDAF pixel corresponding to the calibration object and the corresponding focus position of the camera; The mapping relationship between the calibrated shooting distance, the phase difference of the PDAF pixels, and the focus position of the camera.

4. The three-dimensional object recognition method according to any one of claims 1-3, characterized in that, Based on the depth data of the target 3D object, a machine learning model is used to identify the identity of the target 3D object, including: The spectral information of the reflected light from the target three-dimensional object is obtained, wherein the reflected light is formed by the reflection of a multispectral light source from the target three-dimensional object; Based on the depth data of the target 3D object and the spectral information of the reflected light from the target 3D object, a machine learning model is used to identify the identity of the target 3D object.

5. The three-dimensional object recognition method according to claim 4, characterized in that, Based on the depth data of the target 3D object and the spectral information of the reflected light from the target 3D object, a machine learning model is used to identify the identity of the target 3D object, including: By comparing the spectral information of the multispectral light source with the spectral information of the reflected light, the spectral absorption information of the target three-dimensional object is determined. Based on the spectral absorption information of the target three-dimensional object, determine the material information of the target three-dimensional object; Based on the depth data and spectral absorption information of the target 3D object, a machine learning model is used to identify the identity of the target 3D object.

6. The three-dimensional object recognition method according to claim 5, characterized in that, Also includes: The machine learning model is generated based on the training samples; The training samples are a set of pre-labeled training data consisting of depth data and spectral information of reflected light from several three-dimensional objects. The annotations on the training samples include: several facial identities and object attributes.

7. The three-dimensional object recognition method according to any one of claims 1-3, characterized in that, Projecting binarized fringes onto a target 3D object includes: projecting the binarized fringes onto the target 3D object using a polarized light source.

8. The three-dimensional object recognition method according to any one of claims 1-3, characterized in that, Projecting binarized fringes onto a target 3D object includes: The binarized stripes are projected using a MEMS galvanometer projector module with an integrated light source.

9. The three-dimensional object recognition method according to claim 2 or 3, characterized in that, Projecting binarized fringes onto the target 3D object, including: A stripeless pattern is projected onto the target 3D object, and the size of the target 3D object is pre-determined using a target detection model. Obtain the distance information between the target 3D object and the camera; Based on the distance information and the size of the target 3D object, the period width and the number of stripe groups of the binarized stripes are set; Binary stripes are projected onto the target 3D object.

10. The three-dimensional object recognition method according to claim 1, characterized in that, Projecting binarized fringes onto a target 3D object includes: If the distance information and the size of the target 3D object are within a predetermined range, the encoding rule for the binarized stripes is set to Gray code.

11. A three-dimensional object recognition device, characterized in that, include: The projection module is used to project binary stripes onto a target three-dimensional object, wherein the binary stripes are periodic stripes with alternating bright and dark areas. The focusing module is used to acquire distance information between the target 3D object and the camera; to use the camera to photograph the target 3D object, and to adjust the focusing position of the camera according to the distance information, so that the binarized stripes are blurred, thereby obtaining a sinusoidal stripe image corresponding to the binarized stripes, wherein the sinusoidal stripe image is a stripe with continuous brightness variation; The depth module is used to determine the depth data of the target three-dimensional object using a three-dimensional structured light method based on the sinusoidal fringe image. The identification module is used to identify the identity of the target 3D object using a machine learning model based on the depth data of the target 3D object.

12. The three-dimensional object recognition device according to claim 11, characterized in that, The focusing module includes: The phase difference submodule is used to obtain the phase difference of each phase-focused PDAF pixel corresponding to the target 3D object; The mapping submodule is used to obtain a preset mapping relationship, which is used to determine the focus position of the camera corresponding to the sinusoidal stripe image based on the distance information. The driving submodule is used to drive the lens of the camera to the focusing position that forms the sinusoidal fringe image, based on the mapping relationship, the distance information, and the phase difference of each PDAF pixel corresponding to the target three-dimensional object.

13. The three-dimensional object recognition device according to claim 12, characterized in that, The mapping submodule includes: The acquisition unit is used to focus on the calibration object using PDAF technology at different shooting distances, and to acquire the phase difference of each PDAF pixel corresponding to the calibration object and the corresponding focus position of the camera. The calibration unit is used to calibrate the mapping relationship between the shooting distance, the phase difference of the PDAF pixels, and the focus position of the camera.

14. The three-dimensional object recognition device according to any one of claims 11-13, characterized in that, The identification module includes: The spectral submodule is used to acquire the spectral information of the reflected light from the target three-dimensional object, wherein the reflected light is formed by the reflection of a multispectral light source from the target three-dimensional object; The identification submodule is used to identify the target 3D object by using a machine learning model based on the depth data of the target 3D object and the spectral information of the reflected light of the target 3D object.

15. The three-dimensional object recognition device according to claim 14, characterized in that, The identification submodule includes: An absorption determination unit is used to compare the spectral information of the multispectral light source with the spectral information of the reflected light to determine the spectral absorption information of the target three-dimensional object. The material determination unit is used to determine the material information of the target three-dimensional object based on the spectral absorption information of the target three-dimensional object. The identification unit is used to identify the identity of the target three-dimensional object using a machine learning model based on the depth data and spectral absorption information of the target three-dimensional object.

16. The three-dimensional object recognition device according to claim 15, characterized in that, The three-dimensional object recognition device also includes: The training module is used to train and generate the machine learning model based on the training samples; The training samples are a set of pre-annotated training data consisting of depth data and spectral information of reflected light from several three-dimensional objects; the annotation content of the training samples includes: several facial identities and object attributes.

17. The three-dimensional object recognition device according to any one of claims 11-13, characterized in that, The projection module includes: A polarizer module is used to project the binarized stripes onto the target 3D object using a polarized light source.

18. The three-dimensional object recognition device according to any one of claims 11-13, characterized in that, The projection module includes: A galvanometer module is used to project the binarized stripes using a MEMS galvanometer projector module with an integrated light source.

19. The three-dimensional object recognition device according to claim 12 or 13, characterized in that, The projection module includes: The pre-judgment submodule is used to project a stripeless pattern onto the target 3D object and use a target detection model to pre-judge the size of the target 3D object. The distance information submodule is used to obtain distance information between the target 3D object and the camera; The stripe setting submodule is used to set the period width and the number of stripe groups of the binarized stripes based on the distance information and the size of the target three-dimensional object; The projection submodule is used to project binary stripes onto the target 3D object.

20. The three-dimensional object recognition device according to claim 11, characterized in that, The projection module includes: The Gray code submodule is used to set the encoding rule of the binarized stripes to Gray code when the distance information and the size of the target three-dimensional object are within a predetermined range.

21. A computer device, characterized in that, The computer device includes: a processor adapted to implement various instructions and a storage device, the storage device storing multiple instructions adapted to be loaded by the processor and executed by the processor to perform the three-dimensional object recognition method according to any one of claims 1 to 10.

22. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program that, when executed by a processor, implements the three-dimensional object recognition method according to any one of claims 1 to 10.

23. A computer program product, characterized in that, The computer program product includes a computer program that, when executed by a processor, implements the three-dimensional object recognition method according to any one of claims 1 to 10.

Images