WO2021007760A1 - 身份识别方法、终端设备及计算机存储介质 - Google Patents

身份识别方法、终端设备及计算机存储介质 Download PDF

Info

Publication number
WO2021007760A1
WO2021007760A1 PCT/CN2019/096053 CN2019096053W WO2021007760A1 WO 2021007760 A1 WO2021007760 A1 WO 2021007760A1 CN 2019096053 W CN2019096053 W CN 2019096053W WO 2021007760 A1 WO2021007760 A1 WO 2021007760A1
Authority
WO
WIPO (PCT)
Prior art keywords
surface energy
target object
point cloud
signal
point
Prior art date
Application number
PCT/CN2019/096053
Other languages
English (en)
French (fr)
Inventor
刘建华
周安福
马华东
杨宁
唐海
张治�
Original Assignee
Oppo广东移动通信有限公司
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by Oppo广东移动通信有限公司 filed Critical Oppo广东移动通信有限公司
Priority to CN201980092409.1A priority Critical patent/CN113439274A/zh
Priority to PCT/CN2019/096053 priority patent/WO2021007760A1/zh
Publication of WO2021007760A1 publication Critical patent/WO2021007760A1/zh

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V10/00—Arrangements for image or video recognition or understanding
    • G06V10/40—Extraction of image or video features

Definitions

  • the present invention relates to the field of communication technology, in particular to an identity recognition method, terminal equipment and computer storage medium.
  • common identity authentication technologies can be divided into two categories, one is based on input character strings (such as password) for identity authentication, and the other is based on biological characteristics (such as fingerprints or iris, etc.).
  • input character strings such as password
  • biological characteristics such as fingerprints or iris, etc.
  • biometric fingerprints if a legitimate user's fingerprint is obtained from an object such as a glass or glasses, you can pour liquid glue on the side of the above-mentioned article that is stained with fingerprints, and after the liquid glue is solidified, stick to the article The fingerprint grease will solidify to form a fingerprint model, which can then be used for identity authentication. It can be seen that when biometrics such as fingerprints or iris are acquired by illegal users, this technology cannot effectively distinguish between legal users and illegal users, which causes great security risks in identity authentication. In other words, the current identity authentication technology cannot effectively distinguish legitimate users from illegal users, and the reliability of the identity authentication technology is low.
  • the embodiments of the present application provide an identity recognition method, terminal equipment, and computer storage medium, which can improve the reliability of identity authentication.
  • an identity recognition method including:
  • the reflection signal is a reflection signal received by a signal transmission device after transmitting a wireless signal to a target object in a moving state
  • the point cloud data is input to a trained residual network model to identify the identity of the target object.
  • an embodiment of the present application provides a terminal device, which has the function of implementing the foregoing method.
  • the function can be realized by hardware, or by hardware executing corresponding software.
  • the hardware or software includes one or more units corresponding to the above functions.
  • an embodiment of the present application provides a terminal device, the terminal device includes a processor, and the processor is coupled with the memory, wherein:
  • the memory is configured to store instructions
  • the processor is configured to obtain a reflected signal, wherein the reflected signal is a reflected signal received by a signal transmission device after transmitting a wireless signal to a target object in a moving state; the reflected signal is processed to obtain the Point cloud data of the target object; the point cloud data is input to a trained residual network model to identify the identity of the target object.
  • an embodiment of the present application provides a computer storage medium, wherein the computer-readable storage medium stores a computer program or instruction, and when the program or instruction is executed by a processor, the processor executes The identity recognition method as described in the first aspect.
  • the embodiments of the present application provide a computer program product, wherein the computer program product includes a non-transitory computer-readable storage medium storing a computer program, and the computer program is operable to cause a computer to execute Part or all of the steps described in the first aspect of the application embodiment.
  • the computer program product may be a software installation package.
  • the embodiment of the present application obtains the reflected signal, where the reflected signal is the reflected signal received by the signal transmission device after transmitting the wireless signal to the target object in a moving state, and processes the reflected signal to obtain the point cloud of the target object
  • the point cloud data is input to the trained residual network model to identify the identity of the target object.
  • the embodiment of the present application recognizes the gait through the reflected signal, thereby identifying the identity of the target object, and improving the reliability of identity authentication .
  • FIG. 1 is a schematic diagram of a scenario provided by an embodiment of the present application.
  • Figure 2 is a schematic diagram of another scenario provided by an embodiment of the present application.
  • FIG. 3 is an example diagram of an identity recognition method provided by an embodiment of the present application.
  • FIG. 4 is a schematic diagram of a sequence of identity recognition provided by an embodiment of the present application.
  • FIG. 5 is a schematic diagram of a FM continuous wave signal provided by an embodiment of the present application.
  • FIG. 6 is a schematic diagram of signals of a range-Doppler imaging algorithm provided by an embodiment of the present application.
  • FIG. 7 is a schematic flowchart of a method for acquiring gait features according to an embodiment of the present application.
  • FIG. 8 is a schematic structural diagram of a trained residual network model provided by an embodiment of the present application.
  • FIG. 9 is a schematic structural diagram of a terminal device provided by an embodiment of the present application.
  • FIG. 10 is a block diagram of functional units of a terminal device provided by an embodiment of the present application.
  • FIG. 11 is a schematic structural diagram of another terminal device provided by an embodiment of the present application.
  • Figure 1 shows a schematic diagram of a scenario involved in this application.
  • Signal transmission equipment can be integrated in terminal equipment, which can be smart door locks, attendance machines, safes, health management equipment, computers, handheld computers, or smart phones.
  • the signal transmission device may be a millimeter wave transmission device, an ultrasonic transmission device, or a laser transmission device.
  • the signal transmission device can transmit a wireless signal. If the target object moves in front of the terminal device, the wireless signal will be transmitted to the target object. Then the signal transmission device can receive the reflected signal that the wireless signal is transmitted to the moving target object and then reflected back. The terminal device can obtain the reflected signal received by the signal transmission device and process the reflected signal to obtain the point cloud data of the target object. The point cloud data is input to the trained residual network model to identify the identity of the target object.
  • the signal transmission device is a millimeter wave transmission device
  • the wireless signal transmitted by the signal transmission device is a millimeter wave
  • the reflected signal received by the signal transmission device is a millimeter wave.
  • the signal transmission device is an ultrasonic transmission device
  • the wireless signal transmitted by the signal transmission device is ultrasonic
  • the reflected signal received by the signal transmission device is ultrasonic.
  • the signal transmission device is a laser transmission device
  • the wireless signal emitted by the signal transmission device is laser
  • the reflected signal received by the signal transmission device is laser.
  • the millimeter wave transmission equipment since the millimeter wave is almost unaffected by optical fibers and heat radiation sources, the millimeter wave has a longer propagation distance and is suitable for long-distance identification.
  • the target object may be an object that can walk on feet, such as a human or an animal.
  • the gait characteristics of at least one subject may be pre-stored in the preset database of the terminal device.
  • the terminal device obtains the reflected signal and processes the reflected signal to obtain the point cloud data of the target object.
  • the point cloud data is input to the trained residual network model to identify the identity of the target object.
  • the terminal device inputs the point cloud data into the trained residual network model to obtain the gait characteristics of the target object, and the trained residual network model combines the gait characteristics of the target object with at least one gait in the preset database Feature comparison.
  • the trained residual network model can identify the target object, that is, the target gait feature The corresponding object, and then the terminal device can open the smart door lock or safe, or unlock the computer, palmtop or smart phone, or determine the target object's attendance success, etc.
  • identity recognition is performed by gait, and the user can be recognized without seeing the face, which has high privacy protection.
  • At least one gait feature may be pre-stored in the preset database of the terminal device.
  • the terminal device obtains the reflected signal, processes the reflected signal, and obtains the point cloud data of the target object, and the point cloud data is input To the trained residual network model to identify the identity of the target object.
  • the terminal device inputs the point cloud data into the trained residual network model to obtain the gait characteristics of the target object, and the trained residual network model combines the gait characteristics of the target object with at least one gait in the preset database Feature comparison. If the similarity between the gait feature of the target object and the target gait feature in the preset database is greater than the preset ratio threshold, the trained residual network model can identify the target object, that is, the target gait feature The corresponding patient.
  • FIG. 2 shows a schematic diagram of another scenario involved in this application.
  • the terminal device can establish a communication connection with at least one signal transmission device, and the terminal device can be a security device, a computer, a palmtop computer, or a smart phone.
  • the signal transmission equipment can be installed in public areas such as outdoors, exhibition halls, or conference halls.
  • the signal transmission equipment can specifically be millimeter wave transmission equipment, infrared transmission equipment, ultrasonic transmission equipment, or laser transmission equipment.
  • the signal transmission device can transmit wireless signals. If the target object moves in front of the signal transmission device, the wireless signal will be transmitted to the target object. Then the signal transmission device can receive the reflected signal after the wireless signal is transmitted to the target object in the moving state. The transmission device can send the received reflected signal to the terminal device. The terminal device can process the reflected signal to obtain the point cloud data of the target object, and the point cloud data is input to the trained residual network model to identify the identity of the target object.
  • the signal transmission device is a millimeter wave transmission device
  • the wireless signal transmitted by the signal transmission device is a millimeter wave
  • the reflected signal received by the signal transmission device is a millimeter wave.
  • the signal transmission device is an ultrasonic transmission device
  • the wireless signal transmitted by the signal transmission device is ultrasonic
  • the reflected signal received by the signal transmission device is ultrasonic.
  • the signal transmission device is a laser transmission device
  • the wireless signal emitted by the signal transmission device is laser
  • the reflected signal received by the signal transmission device is laser.
  • the target object may be an object that can walk on feet, such as a human or an animal.
  • the preset database may pre-store the gait characteristics of at least one object, such as the gait characteristics of a criminal or a killer.
  • the trained residual network model compares the gait feature of the target object with at least one gait feature in the preset database. If the gait feature of the target object is similar to the target gait feature in the preset database If the degree is greater than the preset ratio threshold, the trained residual network model can identify the target object, that is, the object corresponding to the target gait feature. If the object corresponding to the target gait feature is a suspect, the terminal device can output prompt information, prompt information It can be used to indicate that the target object is a suspect, so that the user can perform security deployment or arrest in time based on the prompt information.
  • the preset database may pre-store the gait characteristics of at least one object, such as the gait characteristics of the owner of the item.
  • the trained residual network model compares the gait feature of the target object with at least one gait feature in the preset database. If the gait feature of the target object is similar to the target gait feature in the preset database If the degree is less than the preset ratio threshold, the trained residual network model can recognize that the target object is not the owner of the item, then the terminal device can output prompt information, which can be used to indicate that the target object is a thief, so that the user can stop in time based on the prompt information Damage, such as alarms.
  • the suspect may disguise himself to prevent even one hair from falling on the scene of the crime, but the walking posture is difficult to control, and the terminal device can identify the user according to the user’s gait The accuracy of identification is high.
  • Fig. 3 is an identity recognition method provided by an embodiment of the present application. The method includes:
  • S301 Obtain a reflection signal, where the reflection signal is a reflection signal received by the signal transmission device after transmitting a wireless signal to a target object in a moving state.
  • the terminal device can obtain the reflected signal received by the signal transmission device.
  • the signal transmission device transmits a wireless signal to the target object when the target object is moving, and at least one of the target objects After receiving the wireless signal, the surface energy points (SEPs) reflect the wireless signal to the signal transmission device.
  • the wireless signal reflected by the surface energy point is the reflected signal, such as the original signal in Figure 4.
  • the body when the user is walking, the body is irradiated by the wireless signal beam, the body will modulate and reflect the wireless signal beam, and the terminal device can decompose the reflection surface of the body part into at least one surface energy point.
  • the reflected signal generated by the body when it changes dynamically is caused by the superposition of reflections from multiple surface energy points.
  • S302 Process the reflected signal to obtain point cloud data of the target object.
  • the terminal device can deconstruct the reflected signal and generate point cloud data after deconstruction.
  • the point cloud data can be as shown in Figure 4.
  • Point cloud data refers to a set of vectors in a three-dimensional coordinate system. These vectors are usually expressed in the form of X, Y, Z three-dimensional coordinates, and can be used to represent the attribute information of at least one surface energy point of the target object.
  • the attribute information of each surface energy point in at least one surface energy point includes: the moving speed of the surface energy point toward the signal transmission device, the distance between the surface energy point and the signal transmission device, and the reflection signal and the signal transmission device One or more of the angle, the signal-to-noise ratio, the position of the target object, the physical size of the target object, and the point density.
  • S303 The point cloud data is input to the trained residual network model to identify the identity of the target object.
  • the terminal device After the terminal device obtains the point cloud data, it can directly use the trained residual network model to process the point cloud data to obtain the gait characteristics of the target object.
  • the trained residual network model calculates the gait characteristics and prediction of the target object. Set the similarity between at least one gait feature in the database. If the similarity between the gait feature of the target object and the target gait feature in the preset database is greater than the preset ratio threshold, the trained residual network model The target object can be identified as the object corresponding to the target gait feature.
  • the terminal device can input the point cloud data into the trained residual network model for learning, and use the classifier obtained after learning to classify and predict the gait.
  • the trained residual network model may be a residual neural network (Residual Neural Network, resnet) model.
  • the embodiment of this application uses resnet as a deep learning network model, and performs the residual neural network model according to the data characteristics of the point cloud data Adjust, so that the trained residual network model has a better performance in gait recognition using point clouds as raw data. Because the point cloud data obtained by the trained residual network model is relatively sparse, mapping the point cloud to the geometric figure will generate a large amount of redundant data, and the running time of the trained residual network model will be correspondingly longer. In order to make The trained residual network model runs efficiently, without converting the point cloud data into geometric figures, but directly using the trained residual network model to process the point cloud data to obtain the gait characteristics of the target object.
  • the terminal device uses the trained residual network model to process the point cloud data, and before obtaining the gait feature of the target object, the point cloud data may be preprocessed to obtain at least one point cloud sequence. Then, the terminal device can use the trained residual network model to process each point cloud sequence in the at least one point cloud sequence to obtain the gait feature of the target object.
  • the preprocessing may be noise reduction processing, point filling, production point cloud sequence, etc.
  • the terminal device may generate the point cloud sequence by acquiring identification information of at least one surface energy point, the identification information is used to indicate that the first surface energy point of the at least one surface energy point belongs to the same frame, and the The attribute information of the surface energy points of consecutive multiple frames is used as a point cloud sequence, and at least one point cloud sequence is obtained.
  • the number of consecutive frames may be a preset number value, such as 50 or 100, that is, the terminal device regards the attribute information of the surface energy points of every 50 consecutive frames as a point cloud sequence.
  • the attribute information of the surface energy point may also include identification information of the surface energy point, and the identification information may be the frame number of the surface energy point.
  • the terminal device may preprocess the reflected signal to generate point cloud data.
  • the terminal device may use the point cloud data output each time as a frame, and the point cloud data output each time may include attribute information of at least one surface energy point.
  • Gait is a continuous process, which contains the instantaneous and temporal characteristics of gait. Taking into account this characteristic of gait, we combine every 50 frames (30 frames per second, approximately 1.67 seconds) point cloud data into a step State data sample.
  • Each surface energy point includes multiple attributes.
  • RGB represents a single physical characteristic of color
  • attribute information of a surface energy point represents different physical attributes such as space and speed.
  • the terminal device may fill in points by taking the attribute information of the surface energy points of multiple consecutive frames as a point cloud sequence, and before at least one point cloud sequence is obtained, the first point cloud sequence belonging to the same frame
  • the number of surface energy points is compared with the preset number threshold.
  • the second surface energy points, the second surface energy points and the first surface energy points are added
  • the sum of the number reaches the preset number threshold
  • the identification information of the second surface energy point is the same as the identification information of the first surface energy point
  • the attribute information of the second surface energy point is the arithmetic average operation of the attribute information of the first surface energy point owned.
  • the point clouds in each frame can be intercepted or inserted into the average value, so that the number of point clouds in each frame is a preset number threshold, such as 50 Or 70 etc.
  • the terminal device obtains the reflected signal, where the reflected signal is the reflected signal received by the signal transmission device after transmitting the wireless signal to the target object in a moving state, and the reflected signal is processed to obtain the point cloud of the target object Data, point cloud data are input to the trained residual network model to identify the identity of the target object, which can improve the reliability of identity authentication.
  • step S302 in FIG. 3 Based on the schematic diagram of the identity recognition process shown in FIG. 3, the embodiment of the present application specifically describes step S302 in FIG. 3.
  • the terminal device may process the reflected signal according to the Doppler effect or the FM continuous wave principle to obtain the attribute information of at least one surface energy point.
  • the terminal device can obtain the attribute information by: obtaining the Doppler redness of the reflected signal According to the Doppler red shift and the Doppler blue shift, the movement speed of at least one surface energy point in the direction of the signal transmission device is calculated.
  • the Doppler effect means that the wavelength of the object's radiation changes due to the relative motion of the light source and the observer.
  • the wave In front of the moving wave source, the wave is compressed, the wavelength becomes shorter, and the frequency becomes higher; behind the moving wave source, the opposite effect occurs, the wavelength becomes longer, and the frequency becomes lower.
  • the higher the velocity of the wave source the greater the effect produced. According to the degree of red/blue shift of the light wave, the velocity of the wave source moving in the observation direction can be calculated.
  • the terminal device can obtain the attribute information by: obtaining the receiving time of the signal transmission device receiving the reflected signal The difference between the transmission time and the transmission time of the wireless signal transmitted by the signal transmission device, the frequency difference when the target object is in a stationary state, and the Doppler frequency shift when the target object is in a moving state, according to the frequency when the target object is in a stationary state The difference and Doppler frequency shift are used to obtain the frequency difference when the target object is in the moving state, and the frequency difference when the target object is in the moving state is multiplied by the difference value to obtain the distance between at least one surface energy point and the signal transmission device.
  • the transmitted signal is a high-frequency continuous wave whose frequency changes with time according to the law of triangular waves.
  • the frequency of the reflected signal received by the signal transmission equipment is the same as the frequency of the transmitted signal. They are both triangular wave laws, but there is a time difference. Using this tiny time difference, the distance between the surface energy point and the signal transmission equipment can be calculated.
  • the frequency difference is shown in FIG. 6.
  • f b is the frequency difference when the target object is stationary
  • the terminal device can obtain the attribute information by: obtaining the signal transmission device According to the phase of the receiving antenna, the angle between the reflected signal and the signal transmission device is calculated.
  • the signal transmission device may include at least one transmitting antenna and at least one receiving antenna.
  • the transmitting antenna is used for transmitting wireless signals
  • the receiving antenna is used for receiving reflected signals.
  • the T1 millimeter wave radar can include 2 transmitting antennas and 4 receiving antennas.
  • the terminal device can obtain the attribute information by obtaining the output power of the reflected signal, and the signal transmission device is receiving the reflected signal.
  • the output power of the noise received in the signal process is calculated based on the output power of the reflected signal and the output power of the noise.
  • FIG. 7 specifically describes step S303 in FIG. 3.
  • the trained residual network model performs feature extraction on the target attribute information of the surface energy points contained in the point cloud sequence to obtain the attribute characteristics of the target attribute indicated by the target attribute information, and the target attribute information is any of the surface energy points Property information.
  • the terminal device uses the trained residual network model to perform feature extraction on the target attribute information of the surface energy points contained in the point cloud sequence to obtain the attribute characteristics of the target attribute indicated by the target attribute information, including : Use the first preset convolution layer to perform convolution operation on the target attribute information of the surface energy points contained in the point cloud sequence to obtain the first convolution operation result; perform the maximum pooling operation on the first convolution operation result to obtain First operation result; use the second preset convolution layer to perform convolution operation on the first operation result to obtain the second convolution operation result; use the second preset convolution layer to perform convolution operation on the second convolution operation result , Obtain the third convolution operation result; use the second preset convolution layer to perform convolution operation on the third convolution operation result to obtain the fourth convolution operation result; use the second preset convolution layer to convolve the fourth The operation result is convolved to obtain the fifth convolution operation result; the third convolution operation result and the first operation result are subjected to the residual operation to obtain the second operation result; the fifth
  • the deep learning network model has 6 layers, and the first 5 layers of neural network structure are shared, but at least one attribute information of the surface energy points is used for network parameters. training.
  • Figure 9 takes at least one attribute information including 4 attribute information as an example.
  • the first layer of the network structure is a convolutional layer with a convolution kernel size of 7x7 and a step size of 2.
  • the terminal device performs a maximum pooling operation with a step size of 2 after the convolutional layer.
  • Layers 2-5 are all convolutional layers with a convolution kernel of 3 and a step size of 1.
  • the residual operation is performed on the features obtained by the maximum pooling operation.
  • the residual operation is performed on the features obtained by the third layer convolution.
  • the feature is averaged and pooled.
  • the terminal device uses a residual module to ensure the smooth backhaul of gradients, and can extract gait features of different scales.
  • S702 The trained residual network model processes the attribute characteristics of each attribute to obtain gait characteristics.
  • the terminal device uses the trained residual network model to process the attribute characteristics of each attribute to obtain gait characteristics, including: for any point cloud sequence, connecting the attribute characteristics of each attribute information , Get the connected attribute feature; use the preset fully connected layer to classify at least one connected attribute feature to obtain the gait feature.
  • the trained residual network model connects the features obtained by convolution of at least one attribute information together, and uses the sixth fully connected layer for classification, and the network model outputs the final judgment result.
  • the neural network will perform the normalization operation and the Relu operation after the convolution operation, so that the neural network model has a better effect and the accuracy of identity recognition is higher.
  • the trained residual network model processes the point cloud data, and the residual network model can be trained before obtaining the gait characteristics of the target object.
  • the residual network model can be trained before obtaining the gait characteristics of the target object.
  • use MATLAB to save the data and perform offline training.
  • offline training use the cross-loss entropy function as the cost function, use the Adam function as the optimizer, use the learning rate as the number of iterations increases and decrease, and adjust the batch size during training to make the deep learning network model learn the best Parameters.
  • the method of data enhancement is also used to expand the data set, so that the deep learning network model has better generalization ability.
  • the model parameters that performed well on the validation set were selected for real-time testing to select model parameters with good expression and generalization capabilities. Real-time testing is performed by using MATLAB to call the selected network model in real time, and finally the trained residual network model is determined.
  • FIG. 9 is a schematic structural diagram of a terminal device provided by an embodiment of this application.
  • the terminal device may include a range processing module (Range Processing) and a beam forming module (Capon Beam former) , Object Detection Module (Object Detection), Doppler Eestimation Module (DopplerEestimation), and Group Tracker (Group Tracker).
  • Range Processing Range Processing
  • beam forming module Capon Beam former
  • Object Detection Object Detection
  • DopplerEestimation Module DopplerEestimation
  • Group Tracker Group Tracker
  • the event enable register (Enhanced Direct Memory Access, EDMA) is used to move the sample from the analog-to-digital converter (Analog-to-Digital Converter, ADC) output buffer to the The local memory of Digital Signal Processing (DSP).
  • DSP Digital Signal Processing
  • FFT Fast algorithm
  • EDMA is used to move the output from the DSP local memory to the radar cube memory in the third layer (L3) memory.
  • Range processing is interleaved with the effective chirp time of the frame. Unless otherwise specified, all other processing occurs during the idle time between the effective chirp time and the end of the frame.
  • s(t) is the input signal after mixing the baseband signal.
  • M is the number of the angle of the receiver
  • y n is normalized by wavelength sensor location.
  • the object detection module can use the Collaborative Forecast And Replenishment (CFAR) detection algorithm to detect the target object.
  • CFAR detection algorithm uses two channels, the first channel is performed in the distance domain, the second channel is performed in the angle domain, and the second channel confirms the results of the first channel to determine the detection point.
  • the Capon beam weighting algorithm is used to filter the range receiver, and then the peak search is performed on the FFT of the filtered range receiver to So estimate Doppler.
  • the group tracker implements positioning processing through tracking algorithms.
  • the group tracker processes the point cloud data from the DSP.
  • the tracker inputs point cloud data, performs target positioning, and reports the results (target list). Therefore, the output of the group tracker is a set of attribute information (such as the movement speed of the surface energy point in the direction of the signal transmission device, the distance between the surface energy point and the signal transmission device, the reflected signal and the signal transmission device) One or more of angle, signal-to-noise ratio, target object location, target object's physical size, and point density).
  • Each frame signal processed by the above modules includes point cloud data composed of attribute information of at least one surface energy point.
  • the terminal device includes a hardware structure and/or software module corresponding to each function.
  • the present application can be implemented in the form of hardware or a combination of hardware and computer software. Whether a certain function is executed by hardware or computer software-driven hardware depends on the specific application and design constraint conditions of the technical solution. Professionals and technicians can use different methods for each specific application to implement the described functions, but such implementation should not be considered beyond the scope of this application.
  • the embodiment of the present application may divide the terminal device into functional units according to the foregoing method examples.
  • each functional unit may be divided corresponding to each function, or two or more functions may be integrated into one processing unit.
  • the above-mentioned integrated unit can be realized in the form of hardware or software program module. It should be noted that the division of units in the embodiments of the present application is illustrative, and is only a logical function division, and there may be other division methods in actual implementation.
  • FIG. 10 shows a block diagram of a possible functional unit composition of the terminal device involved in the foregoing embodiment, and the terminal device includes:
  • the communication unit 1001 is configured to obtain a reflection signal, where the reflection signal is a reflection signal received by a signal transmission device after transmitting a wireless signal to a target object in a moving state;
  • the processing unit 1002 is configured to process the reflected signal to obtain point cloud data of the target object; the point cloud data is input to the trained residual network model to identify the identity of the target object.
  • the processing unit 1002 may be a processor or a controller, and the communication unit 1001 may be a transceiver, a transceiver circuit, a radio frequency chip, or the like.
  • the processing unit 1002 processes the reflected signal to obtain the point cloud data of the target object, including:
  • the reflected signal is processed according to the Doppler effect or the principle of frequency-modulated continuous wave to obtain attribute information of at least one surface energy point.
  • the trained residual network model before the point cloud data is input to the trained residual network model, it further includes:
  • the processing unit 1002 obtains identification information of at least one surface energy point, where the identification information is used to indicate that the first surface energy point of the at least one surface energy point belongs to the same frame;
  • the processing unit 1002 uses the attribute information of the surface energy points of multiple consecutive frames as a point cloud sequence to obtain at least one point cloud sequence;
  • the point cloud data is input to the trained residual network model, including:
  • Each of the at least one point cloud sequence is input to the trained residual network model.
  • the processing unit 1002 uses the attribute information of the surface energy points of multiple consecutive frames as a point cloud sequence, and before obtaining at least one point cloud sequence, the method further includes:
  • the processing unit 1102 increases the second surface energy points, and the sum of the number of the second surface energy points and the first surface energy points reaches The preset number threshold, the identification information of the second surface energy point and the identification information of the first surface energy point are the same, and the attribute information of the second surface energy point is relative to the first surface energy point Attribute information is obtained by arithmetic average operation;
  • the processing unit 1102 deletes the third surface energy point in the first surface energy point, and the first surface energy point is divided by The sum of the number of surface energy points other than the third surface energy point reaches the preset number threshold.
  • the trained residual network model compares the target attribute information of the surface energy points contained in the point cloud sequence Performing feature extraction to obtain the attribute feature of the target attribute indicated by the target attribute information, where the target attribute information is any attribute information of the surface energy point;
  • the attribute characteristics of each attribute are processed to obtain the gait characteristics.
  • the trained residual network model performs feature extraction on the target attribute information of the surface energy points contained in the point cloud sequence to obtain the attribute characteristics of the target attribute indicated by the target attribute information, including :
  • the processing unit 1002 processes the attribute characteristics of each attribute to obtain the gait characteristics, including:
  • the attribute information of each surface energy point in the at least one surface energy point includes: a moving speed of the surface energy point in the direction of the signal transmission device, and the surface energy point One of the distance between the signal transmission equipment, the angle between the reflected signal and the signal transmission equipment, the signal-to-noise ratio, the position of the target object, the physical size of the target object, the point density, or Many kinds.
  • the processing unit 1002 processes the reflected signal according to the Doppler effect to obtain attribute information of at least one surface energy point, including:
  • the movement speed of the at least one surface energy point in the direction of the signal transmission device is calculated.
  • the processing unit 1002 processes the reflected signal according to the principle of frequency modulated continuous wave to obtain attribute information of at least one surface energy point, including:
  • the frequency difference when the target object is in a moving state is multiplied by the difference value to obtain the distance between the at least one surface energy point and the signal transmission device.
  • the processing unit 1002 processes the reflected signal to obtain the point cloud data of the target object, including:
  • the angle between the reflected signal and the signal transmission device is calculated.
  • the processing unit 1002 processes the reflected signal to obtain the point cloud data of the target object, including:
  • the signal-to-noise ratio is calculated.
  • the wireless signal transmitted by the signal transmission device of the processing unit 1002 is a millimeter wave
  • the reflected signal received by the signal transmission device is a millimeter wave
  • the terminal device involved in the embodiment of the present application may be the terminal device shown in FIG. 11.
  • the communication unit 1001 obtains a reflected signal, where the reflected signal is a reflected signal received by a signal transmission device after transmitting a wireless signal to a target object in a moving state, and the processing unit 1002 processes the reflected signal ,
  • the point cloud data of the target object is obtained, and the point cloud data is input to the trained residual network model pair to identify the identity of the target object, which can improve the security of identity recognition.
  • the embodiment of the present application also provides a computer-readable storage medium, wherein the computer-readable storage medium stores a computer program for electronic data exchange, wherein the computer program causes the computer to execute the terminal in the above method embodiment Part or all of the steps described by the device.
  • the embodiments of the present application also provide a computer program product, wherein the computer program product includes a non-transitory computer-readable storage medium storing a computer program, and the computer program is operable to cause a computer to execute the method embodiments described above Part or all of the steps described in the terminal device.
  • the computer program product may be a software installation package.
  • the steps of the method or algorithm described in the embodiments of the present application may be implemented in a hardware manner, or may be implemented in a manner that a processor executes software instructions.
  • Software instructions can be composed of corresponding software modules, which can be stored in random access memory (Random Access Memory, RAM), flash memory, read-only memory (Read Only Memory, ROM), and erasable programmable read-only memory ( Erasable Programmable ROM (EPROM), Electrically Erasable Programmable Read-Only Memory (Electrically EPROM, EEPROM), register, hard disk, mobile hard disk, CD-ROM or any other form of storage medium known in the art.
  • An exemplary storage medium is coupled to the processor, so that the processor can read information from the storage medium and can write information to the storage medium.
  • the storage medium may also be an integral part of the processor.
  • the processor and the storage medium may be located in the ASIC.
  • the ASIC may be located in an access network device, a target network device, or a core network device.
  • the processor and the storage medium may also exist as discrete components in the access network device, the target network device, or the core network device.
  • the functions described in the embodiments of the present application may be implemented in whole or in part by software, hardware, firmware, or any combination thereof.
  • software it can be implemented in the form of a computer program product in whole or in part.
  • the computer program product includes one or more computer instructions.
  • the computer may be a general-purpose computer, a special-purpose computer, a computer network, or other programmable devices.
  • the computer instructions may be stored in a computer-readable storage medium or transmitted from one computer-readable storage medium to another computer-readable storage medium.
  • the computer instructions may be transmitted from a website, computer, server, or data center. Transmission to another website, computer, server, or data center via wired (for example, coaxial cable, optical fiber, Digital Subscriber Line (DSL)) or wireless (for example, infrared, wireless, microwave, etc.).
  • the computer-readable storage medium may be any available medium that can be accessed by a computer or a data storage device such as a server or a data center integrated with one or more available media.
  • the usable medium may be a magnetic medium (for example, a floppy disk, a hard disk, a magnetic tape), an optical medium (for example, a Digital Video Disc (DVD)), or a semiconductor medium (for example, a Solid State Disk (SSD)) )Wait.
  • a magnetic medium for example, a floppy disk, a hard disk, a magnetic tape
  • an optical medium for example, a Digital Video Disc (DVD)
  • DVD Digital Video Disc
  • SSD Solid State Disk

Landscapes

  • Engineering & Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • General Physics & Mathematics (AREA)
  • Multimedia (AREA)
  • Theoretical Computer Science (AREA)
  • Radar Systems Or Details Thereof (AREA)

Abstract

本申请实施例公开了身份识别方法、终端设备及计算机存储介质，包括：获取反射信号，其中反射信号是信号传输设备在发射无线信号至处于移动状态的目标对象后接收到的反射信号；对反射信号进行处理，得到目标对象的点云数据；点云数据被输入至经训练的残差网络模型对以识别目标对象的身份。本申请实施例可提高身份认证的可靠性。

Description

身份识别方法、终端设备及计算机存储介质 技术领域
本发明涉及通信技术领域，尤其涉及身份识别方法、终端设备及计算机存储介质。
背景技术
目前，常见的身份认证技术可以分为两类，一类是基于输入字符串(如密码)进行身份认证的技术，另一类是基于生物特征(如指纹或虹膜等)进行身份认证的技术。在实践中发现，在上述的基于输入字符串进行身份认证的技术中，当输入字符串被非法用户获取时，该技术无法有效地区别出合法用户和非法用户，从而使得身份认证存在很大的安全隐患。另外，以生物特征为指纹为例，如果从玻璃杯或者眼镜等物体上获取到合法用户的指纹，可以向上述物品中沾有指纹的一面倒入液体胶，等液体胶凝固之后，粘在物品上的指纹油脂就会固化，形成指纹模型，然后可以使用该指纹模型进行身份认证。由此可见，指纹或虹膜等生物特征被非法用户获取时，该技术无法有效地区别出合法用户和非法用户，从而使得身份认证存在很大的安全隐患。也就是说，目前的身份认证技术无法有效地区别出合法用户和非法用户，身份认证技术的可靠性较低。
发明内容
本申请的实施例提供身份识别方法、终端设备及计算机存储介质，可提高身份认证的可靠性。
第一方面，本申请实施例提供一种身份识别方法，包括：
获取反射信号，其中所述反射信号是信号传输设备在发射无线信号至处于移动状态的目标对象后接收到的反射信号；
对所述反射信号进行处理，得到所述目标对象的点云数据；
所述点云数据被输入至经训练的残差网络模型以识别所述目标对象的身份。
第二方面，本申请实施例提供一种终端设备，该终端设备具有实现上述方法的功能。所述功能可以通过硬件实现，也可以通过硬件执行相应的软件实现。所述硬件或软件包括一个或多个与上述功能相对应的单元。
第三方面，本申请实施例提供一种终端设备，该终端设备包括处理器，所述处理器与所述存储器耦合，其中：
所述存储器，被配置成存储指令；
所述处理器，被配置成获取反射信号，其中所述反射信号是信号传输设备在发射无线信号至处于移动状态的目标对象后接收到的反射信号；对所述反射信号进行处理，得到所述目标对象的点云数据；所述点云数据被输入至经训练的残差网络模型以识别所述目标对象的身份。
第四方面，本申请实施例提供了一种计算机存储介质，其中，所述计算机可读存储介质存储有计算机程序或指令，当所述程序或指令被处理器执行时，使所述处理器执行如第一方面所述的身份识别方法。
第五方面，本申请实施例提供了一种计算机程序产品，其中，所述计算机程序产品包括存储了计算机程序的非瞬时性计算机可读存储介质，所述计算机程序可操作来使计算机执行如本申请实施例第一方面所描述的部分或全部步骤。该计算机程序产品可以为一个软 件安装包。
可以看出，本申请实施例通过获取反射信号，其中反射信号是信号传输设备在发射无线信号至处于移动状态的目标对象后接收到的反射信号，对反射信号进行处理，得到目标对象的点云数据，所述点云数据被输入至经训练的残差网络模型以识别目标对象的身份，相对基于输入字符串进行身份认证或者基于指纹等生物特征进行身份认证，由于步态受性别、体重、年龄或者性格类型等影响，不同用户的步态是不相同的，且步态很难伪造，因此本申请实施例通过反射信号识别步态，进而识别目标对象的身份，可提高身份认证的可靠性。
附图说明
下面将对实施例或现有技术描述中所需要使用的附图作简单地介绍。
图1是本申请实施例提供的一种场景示意图；
图2是本申请实施例提供的另一种场景示意图；
图3是本申请实施例提供的一种身份识别方法示例图；
图4是本申请实施例提供的一种身份识别的时序示意图；
图5是本申请实施例提供的一种调频连续波的信号示意图；
图6是本申请实施例提供的一种距离-多普勒成像算法的信号示意图；
图7是本申请实施例提供的一种步态特征获取方法的流程示意图；
图8是本申请实施例提供的一种经训练的残差网络模型的结构示意图；
图9是本申请实施例提供的一种终端设备的结构示意图；
图10是本申请实施例提供的一种终端设备的功能单元组成框图；
图11是本申请实施例提供的另一种终端设备的结构示意图。
具体实施方式
下面将结合附图对本申请实施例中的技术方案进行描述。
图1示出了本申请涉及的一种场景示意图。信号传输设备可以集成在终端设备中，终端设备可以为智能门锁，考勤机，保险柜，健康管理设备，计算机，掌上电脑或者智能手机等。信号传输设备可以为毫米波传输设备、超声波传输设备或者激光传输设备等。
信号传输设备可发射无线信号，如果目标对象在终端设备前方移动，无线信号将发射至目标对象，那么信号传输设备可接收无线信号发射至处于移动状态的目标对象后反射回来的反射信号。终端设备可以获取信号传输设备接收到的反射信号，对反射信号进行处理，得到目标对象的点云数据，点云数据被输入至经训练的残差网络模型以识别目标对象的身份。
示例性的，如果信号传输设备为毫米波传输设备，那么信号传输设备发射的无线信号为毫米波，信号传输设备接收到的反射信号为毫米波。如果信号传输设备为超声波传输设备，那么信号传输设备发射的无线信号为超声波，信号传输设备接收到的反射信号为超声波。如果信号传输设备为激光传输设备，那么信号传输设备发射的无线信号为激光，信号传输设备接收到的反射信号为激光。
以信号传输设备为毫米波传输设备为例，由于毫米波几乎不受光纤和热辐射源等影响，所以毫米波的传播距离更远，可适用于远距离身份识别。
其中，目标对象可以为可用脚行走的对象，例如人或者动物等。
在一种实施例中，终端设备的预设数据库中可以预先存储有至少一个对象的步态特征。终端设备获取到反射信号，对反射信号进行处理，得到目标对象的点云数据，点云数据被输入至经训练的残差网络模型以识别目标对象的身份。例如，终端设备将点云数据输入至 经训练的残差网络模型，得到目标对象的步态特征，经训练的残差网络模型将目标对象的步态特征和预设数据库中的至少一个步态特征进行比对，如果目标对象的步态特征和预设数据库中的目标步态特征之间的相似度大于预设比例阈值，那么经训练的残差网络模型可以识别目标对象即目标步态特征对应的对象，进而终端设备可以开启智能门锁或者保险柜，或者解锁计算机，掌上电脑或者智能手机，或者确定目标对象考勤成功，等等。
在该实施例中，通过步态进行身份识别，不需要看到人脸就能识别出用户，具有很高的隐私保护性。
在一种实现方式中，终端设备的预设数据库中可以预先存储有至少一个步态特征，终端设备获取到反射信号，对反射信号进行处理，得到目标对象的点云数据，点云数据被输入至经训练的残差网络模型以识别目标对象的身份。例如，终端设备将点云数据输入至经训练的残差网络模型，得到目标对象的步态特征，经训练的残差网络模型将目标对象的步态特征和预设数据库中的至少一个步态特征进行比对，如果目标对象的步态特征和预设数据库中的目标步态特征之间的相似度大于预设比例阈值，那么经训练的残差网络模型可以识别目标对象即目标步态特征对应的病患。
图2示出了本申请涉及的另一种场景示意图。终端设备可以和至少一个信号传输设备建立通信连接，终端设备可以为安防设备，计算机，掌上电脑或者智能手机等。信号传输设备可以安装在室外，展厅，或者大会堂等公共区域，信号传输设备具体可以为毫米波传输设备、红外线传输设备、超声波传输设备或者激光传输设备等。
信号传输设备可发射无线信号，如果目标对象在信号传输设备前方移动，无线信号将发射至目标对象，那么信号传输设备可接收无线信号发射至处于移动状态的目标对象后反射回来的反射信号，信号传输设备可以将接收到的反射信号发送给终端设备。终端设备可以对反射信号进行处理，得到目标对象的点云数据，点云数据被输入至经训练的残差网络模型以识别目标对象的身份。
示例性的，如果信号传输设备为毫米波传输设备，那么信号传输设备发射的无线信号为毫米波，信号传输设备接收到的反射信号为毫米波。如果信号传输设备为超声波传输设备，那么信号传输设备发射的无线信号为超声波，信号传输设备接收到的反射信号为超声波。如果信号传输设备为激光传输设备，那么信号传输设备发射的无线信号为激光，信号传输设备接收到的反射信号为激光。
其中，目标对象可以为可用脚行走的对象，例如人或者动物等。
例如，预设数据库中可以预先存储有至少一个对象的步态特征，例如罪犯或者杀手的步态特征。经训练的残差网络模型将目标对象的步态特征和预设数据库中的至少一个步态特征进行比对，如果目标对象的步态特征和预设数据库中的目标步态特征之间的相似度大于预设比例阈值，那么经训练的残差网络模型可以识别目标对象即目标步态特征对应的对象，如果目标步态特征对应的对象为嫌疑人，那么终端设备可以输出提示信息，提示信息可以用于指示目标对象为嫌疑人，以便用户基于提示信息及时进行安防部署或者逮捕等。
又如，预设数据库中可以预先存储有至少一个对象的步态特征，例如物品所有者的步态特征。经训练的残差网络模型将目标对象的步态特征和预设数据库中的至少一个步态特征进行比对，如果目标对象的步态特征和预设数据库中的目标步态特征之间的相似度小于预设比例阈值，那么经训练的残差网络模型可以识别目标对象不是物品所有者，那么终端设备可以输出提示信息，提示信息可以用于指示目标对象为小偷，以便用户基于提示信息及时止损，例如报警等。
在该实施例中，嫌疑人或许会给自己化装，不让自己身上的哪怕一根毛发掉在作案现场，但走路的姿势是很难控制的，终端设备可以根据用户的步态去识别出用户的身份，身 份识别的准确度较高。
针对上述场景，本申请实施例提出以下实施例，下面结合附图进行详细描述。
基于图1和图2所示的场景示意图，请参阅图3，图3是本申请实施例提供的一种身份识别方法，该方法包括：
S301，获取反射信号，其中反射信号是信号传输设备在发射无线信号至处于移动状态的目标对象后接收到的反射信号。
终端设备可以获取信号传输设备接收到的反射信号，以图4所示的身份识别的时序示意图为例，目标对象在移动过程中，信号传输设备发射无线信号至目标对象，目标对象上的至少一个表面能量点(SEPs)接收到无线信号之后反射无线信号至信号传输设备，表面能量点反射的无线信号即反射信号，例如图4中的原始信号。
其中，用户在行走时，身体被无线信号波束照射，身体会将该无线信号波束调制并反射，终端设备可以将身体部位的反射面分解成至少一个表面能量点。身体在动态变化时所产生的反射信号就是由多个表面能量点的反射的叠加所造成。
S302，对反射信号进行处理，得到目标对象的点云数据。
终端设备可以对反射信号进行解构，解构后生成点云数据，点云数据可以如图4所示。点云数据是指在一个三维坐标系统中的一组向量的集合。这些向量通常以X，Y，Z三维坐标的形式表示，可用来代表目标对象的至少一个表面能量点的属性信息。至少一个表面能量点中每个表面能量点的属性信息包括：表面能量点向信号传输设备的方向移动的移动速度，表面能量点与信号传输设备之间的距离，反射信号与信号传输设备之间的角度，信噪比，目标对象的位置，目标对象的物理尺寸，点密度中的一种或多种。
S303，点云数据被输入至经训练的残差网络模型以识别目标对象的身份。
终端设备获取到点云数据之后，可以直接使用经训练的残差网络模型对点云数据进行处理，得到目标对象的步态特征，经训练的残差网络模型计算目标对象的步态特征和预设数据库中的至少一个步态特征之间的相似度，如果目标对象的步态特征和预设数据库中的目标步态特征之间的相似度大于预设比例阈值，经训练的残差网络模型可以识别目标对象为目标步态特征对应的对象。
以图4为例，终端设备可以将点云数据输入经训练的残差网络模型进行学习，使用学习后所得的分类器对步态进行分类预测。经训练的残差网络模型可以为残差神经网络(Residual Neural Network，resnet)模型，本申请实施例将resnet作为深度学习网络模型，并且根据点云数据的数据特点对残差神经网络模型进行了调整，从而使得经训练的残差网络模型在以点云为原始数据的步态识别上有更好的表现。由于经训练的残差网络模型获取到的点云数据比较稀疏，把点云映射到几何图形会生成大量的冗余数据，经训练的残差网络模型的运行时间也会相应变长，为了使经训练的残差网络模型高效运行，可以无需将点云数据转换为几何图形，而是直接使用经训练的残差网络模型对点云数据进行处理，得到目标对象的步态特征。
在一种实现方式中，终端设备使用经训练的残差网络模型对点云数据进行处理，得到目标对象的步态特征之前，可以对点云数据进行预处理，得到至少一个点云序列。然后，终端设备可以使用经训练的残差网络模型对至少一个点云序列中的每个点云序列进行处理，得到目标对象的步态特征。示例性的，预处理可以为降噪处理、点数填充以及生产点云序列等。
在一种实现方式中，终端设备生成点云序列的方式可以为：获取至少一个表面能量点的标识信息，标识信息用于指示至少一个表面能量点中的第一表面能量点属于同一帧，将连续多个帧的表面能量点的属性信息作为一个点云序列，得到至少一个点云序列。连续多 个帧的数量可以为预先设定的数量值，例如50或者100，即终端设备将每50个连续帧的表面能量点的属性信息作为一个点云序列。
表面能量点的属性信息还可以包括表面能量点的标识信息，标识信息可以为表面能量点的帧号。终端设备可以将反射信号进行预处理，生成点云数据，终端设备可以将每次输出的点云数据作为一帧，每次输出的点云数据可以包括至少一个表面能量点的属性信息。步态是一个连续的过程，其中包含步态的瞬时特征与时间特征，考虑到步态的这种特性，我们将每50帧(每秒30帧，大约1.67秒)点云数据组合成一个步态数据样本。每个表面能量点包括多个属性，相对于几何图形的RGB，RGB表示的是颜色这个单一物理特性，而表面能量点的属性信息表示的是空间，速度等不同的物理属性。然后，终端设备将不同属性通过相同的子网络分别提取特征，然后再将不同的属性特征连接起来，获得更高层的语义信息，可提高身份识别的准确度。
在一种实现方式中，终端设备进行点数填充的方式可以为：将连续多个帧的表面能量点的属性信息作为一个点云序列，得到至少一个点云序列之前，将属于同一帧的第一表面能量点的数量和预设数量阈值进行比较，当属于同一帧的第一表面能量点的数量小于预设数量阈值时，增加第二表面能量点，第二表面能量点和第一表面能量点的数量总和达到预设数量阈值，第二表面能量点的标识信息和第一表面能量点的标识信息相同，第二表面能量点的属性信息是对第一表面能量点的属性信息进行算术平均运算得到的。当属于同一帧的第一表面能量点的数量大于预设数量阈值时，删除第一表面能量点中的第三表面能量点，第一表面能量点中除第三表面能量点以外的表面能量点的数量总和达到预设数量阈值。
由于终端设备每次输出的点云的数量参差不齐，可以将每一帧中的点云通过截取或插入平均值的方法，使得每一帧的点云的数量为预设数量阈值，例如50或者70等。
在本申请实施例中，终端设备获取反射信号，其中反射信号是信号传输设备在发射无线信号至处于移动状态的目标对象后接收到的反射信号，对反射信号进行处理，得到目标对象的点云数据，点云数据被输入至经训练的残差网络模型以识别目标对象的身份，可提高身份认证的可靠性。
基于图3所示的身份识别的流程示意图，本申请实施例对图3中的步骤S302进行具体描述。
在一种实现方式中，终端设备可以根据多普勒效应或调频连续波原理对反射信号进行处理，得到至少一个表面能量点的属性信息。
在一种实现方式中，如果表面能量点的属性信息为该表面能量点向信号传输设备的方向移动的移动速度，那么终端设备获取该属性信息的方式可以为：获取反射信号的多普勒红移和多普勒蓝移，根据多普勒红移和所述多普勒蓝移，计算得到至少一个表面能量点向信号传输设备的方向移动的移动速度。
多普勒效应是指物体辐射的波长因为光源和观测者的相对运动而产生变化。在运动的波源前面，波被压缩，波长变得较短，频率变得较高；在运动的波源后面，产生相反的效应，波长变得较长，频率变得较低。波源的速度越高，所产生的效应越大，根据光波红/蓝移的程度，可以计算出波源循着观测方向运动的速度。
在一种实现方式中，如果表面能量点的属性信息为该表面能量点与信号传输设备之间的距离，那么终端设备获取该属性信息的方式可以为：获取信号传输设备接收反射信号的接收时间和信号传输设备发射无线信号的发射时间之间的差值，获取目标对象处于静止状态时的频差，以及目标对象处于移动状态时的多普勒频移，根据目标对象处于静止状态时的频差和多普勒频移，得到目标对象处于移动状态时的频差，将目标对象处于移动状态时的频差与差值相乘，得到至少一个表面能量点与信号传输设备之间的距离。
以图5所示的调频连续波(Frequency Modulated Continuous Wave，FMCW)的信号示意图为例，发射信号为高频连续波，其频率随时间按照三角波规律变化。信号传输设备接收的反射信号的频率与发射信号的频率变化规律相同，都是三角波规律，只是有一个时间差，利用这个微小的时间差可计算出表面能量点与信号传输设备之间的距离。
以图6所示的距离-多普勒成像算法(RangeDoppler)的信号示意图为例，考虑到目标对象的移动，根据多普勒效应，其频差如图6所示。其中f b为目标对象静止时的频差，f d是目标对象移动时的多普勒频移。如果表面能量点反射的无线信号位于波峰之前，那么将目标对象处于静止状态时的频差和多普勒频移相加，得到目标对象处于移动状态时的频差，即f b+f d＝f bu。如果表面能量点反射的无线信号位于波峰之后，那么将目标对象处于静止状态时的频差减去多普勒频移，得到目标对象处于移动状态时的频差，即f b-f d＝f bd。
在一种实现方式中，如果表面能量点的属性信息为该表面能量点反射回来的无线信号与信号传输设备之间的角度，那么终端设备获取该属性信息的方式可以为：获取信号传输设备的接收天线的相位，根据接收天线的相位，计算得到反射信号与信号传输设备之间的角度。
其中，信号传输设备可以包括至少一个发射天线和至少一个接收天线，发射天线用于发射无线信号，接收天线用于接收反射信号。以信号传输设备为TI(Texas Instruments,德州仪器)毫米波雷达为例，T1毫米波雷达可以包括2个发射天线和4个接收天线。
在一种实现方式中，如果表面能量点的属性信息为该表面能量点的信噪比，那么终端设备获取该属性信息的方式可以为：获取反射信号的输出功率，以及信号传输设备在接收反射信号的过程中接收到的噪声的输出功率，根据反射信号的输出功率和噪声的输出功率，计算得到信噪比。
基于图3所示的身份识别的流程示意图，请参阅图7，图7对图3中的步骤S303进行具体描述。
S701，经训练的残差网络模型对点云序列所包含的表面能量点的目标属性信息进行特征提取，得到目标属性信息所指示的目标属性的属性特征，目标属性信息为表面能量点的任一属性信息。
在一种实现方式中，终端设备使用经训练的残差网络模型对将点云序列所包含的表面能量点的目标属性信息进行特征提取，得到目标属性信息所指示的目标属性的属性特征，包括：使用第一预设卷积层对点云序列所包含的表面能量点的目标属性信息进行卷积运算，得到第一卷积运算结果；对第一卷积运算结果进行最大池化操作，得到第一操作结果；使用第二预设卷积层对第一操作结果进行卷积运算，得到第二卷积运算结果；使用第二预设卷积层对第二卷积运算结果进行卷积运算，得到第三卷积运算结果；使用第二预设卷积层对第三卷积运算结果进行卷积运算，得到第四卷积运算结果；使用第二预设卷积层对第四卷积运算结果进行卷积运算，得到第五卷积运算结果；将第三卷积运算结果和第一操作结果进行残差操作，得到第二操作结果；将第五卷积运算结果和第三卷积运算结果进行残差操作，得到第三操作结果；对第二操作结果和第三操作结果进行平均池化操作，得到目标属性的属性特征。
以图8所示的经训练的残差网络模型的结构示意图为例，深度学习网络模型一共有6层，前5层神经网络结构共享，但是表面能量点的至少一个属性信息各自进行网络参数的训练。图9以至少一个属性信息包括4个属性信息为例，网络结构的第一层是一个卷积核大 小7x7，步长为2的卷积层。终端设备在卷积层后进行步长为2的最大池化操作。第2-5层均为卷积核为3，步长为1的卷积层。在第3层卷积操作结束后，将其得到的特征与最大池化操作得到的特征进行残差操作。在第五层卷积操作结束后，将其得到的特征与第三层卷积得到的特征做残差操作。残差操作后对特征进行平均池化操作。
在该实施例中，终端设备运用了残差模块来保证梯度反向回传的流畅，同时可以提取到不同尺度的步态特征。
S702，经训练的残差网络模型将每个属性的属性特征进行处理，得到步态特征。
在一种实现方式中，终端设备使用经训练的残差网络模型将每个属性的属性特征进行处理，得到步态特征，包括：针对任一点云序列，将每个属性信息的属性特征进行连接，得到连接后的属性特征；使用预设全连接层对至少一个连接后的属性特征进行分类，得到步态特征。
以图8为例，经训练的残差网络模型将至少一个属性信息卷积得到的特征连接到一起，使用第六层全连接层进行分类，网络模型输出最后的判断结果。该神经网络在进行卷积操作后都会进行归一化操作和Relu操作，使得该神经网络模型有更好的效果，身份识别的准确率较高。
在该实施例中，采用了一种类似早期融合(Early Fusion)的操作，使模型分类的准确度进一步提高。
在一种实现方式中，经训练的残差网络模型对点云数据进行处理，得到目标对象的步态特征之前，可以对残差网络模型进行训练。例如，使用MATLAB将数据保存下来，并进行离线训练。离线训练时，使用交叉损失熵函数作为代价函数，使用Adam函数作为优化器，使用随着迭代次数变大而变小的学习率并且调整训练时的批大小来使得深度学习网络模型学习到最优的参数。在训练的过程中还使用数据增强的方法扩大数据集，使得深度学习网络模型具有更好的泛化能力。经过数次迭代训练后，选取在验证集上表现良好的模型参数进行实时测试以选取表达能力与泛化能力俱佳的模型参数。通过使用MATLAB实时调用选取的网络模型的方式进行实时测试，最终确定经训练的残差网络模型。
基于上述实施例以及前述内容，请参阅图9，图9为本申请实施例提供的一种终端设备的结构示意图，终端设备可以包括距离处理模块(Range Processing)，波束形成模块(Capon Beam Former)，物体检测模块(Object Detection)，多普勒估计模块(DopplerEestimation)，以及群追踪器(Group Tracker)。
针对距离处理模块，对于信号传输设备的每个天线，事件使能寄存器(Enhanced Direct Memory Access，EDMA)用于将采样从模数转换器(Analog-to-Digital Converter，ADC)输出缓冲区移动到数字信号处理技术(Digital Signal Processing，DSP)的本地存储器。执行16位定点1D窗口和16位定点1D离散傅氏变换的快速算法(Fast Fourier Transformation，FFT)。EDMA用于将输出从DSP本地存储器移动到第三层(L3)存储器中的雷达立方体存储器。范围处理与帧的有效chirp时间交织。除非特殊说明外，其他所有处理都发生在有效chirp时间和帧结束之间的idle时间期间。
针对波束形成模块，s(t)是混合基带信号后的输入信号。传感器阵列信号通过公式X(t)＝A(θ)s(t)+n(t)进行处理。其中A(θ)＝(a(θ 1),…a(θ M))是控制矩阵，a(θ)＝(e j2πy1 sin(θ),…,e j2πyN sin(θ))是控制向量，M是角度接收器的数量，y n是按波长归一化的传感器位置。Capon BF方法的公式是θ capon＝argmin θ{trace(A(θ)×R n -1×A(θ) H}，其中R n是空间协方差矩阵。
另外，静态杂物去除通过移除每个range接收器的直流元件来实现。这消除了静态对象桌子或椅子在感兴趣区域的反射。然后，对于每个范围区间，使用帧内的多个chirps来计算空间协方差矩阵Rn。然后将Rn反转，并将Rn -1的上对角线存储在每个距离箱(range bin) 的存储器中。对于每个距离箱(range bin)，计算Capon波束形成器输出并将角度谱存储在存储器中以构建范围方位热图。
物体检测模块可利用恒虚警处理(Collaborative Forecast And Replenishment，CFAR)检测算法进行目标对象的检测。CFAR检测算法使用两个通道，第一通道在距离域中进行，第二通道在角度域中进行，第二通道对第一通道的结果加以确认，以确定检测点。
针对多普勒估计模块，对于从检测模型中检测到的每一[距离，方位]对，利用Capon波束加权算法过滤距离接收器，然后在滤波后的距离接收器的FFT上进行峰值搜索，以此来估计多普勒。
群追踪器通过跟踪算法实现定位处理。群跟踪器处理来自DSP的点云数据。跟踪器输入点云数据，执行目标定位，并报告结果(目标列表)。因此，群跟踪器的输出是一组具有属性信息(如表面能量点向信号传输设备的方向移动的移动速度，表面能量点与信号传输设备之间的距离，反射信号与信号传输设备之间的角度，信噪比，目标对象的位置，目标对象的物理尺寸，点密度中的一种或多种)的可跟踪对象。
通过上述各个模块处理后的每一帧信号包括至少一个表面能量点的属性信息构成的点云数据。
上述主要从各个网元之间交互的角度对本申请实施例的方案进行了介绍。可以理解的是，终端设备为了实现上述功能，其包含了执行各个功能相应的硬件结构和/或软件模块。本领域技术人员应该很容易意识到，结合本文中所公开的实施例描述的各示例的单元及算法步骤，本申请能够以硬件或硬件和计算机软件的结合形式来实现。某个功能究竟以硬件还是计算机软件驱动硬件的方式来执行，取决于技术方案的特定应用和设计约束条件。专业技术人员可以对每个特定的应用使用不同方法来实现所描述的功能，但是这种实现不应认为超出本申请的范围。
本申请实施例可以根据上述方法示例对终端设备进行功能单元的划分，例如，可以对应各个功能划分各个功能单元，也可以将两个或两个以上的功能集成在一个处理单元中。上述集成的单元既可以采用硬件的形式实现，也可以采用软件程序模块的形式实现。需要说明的是，本申请实施例中对单元的划分是示意性的，仅仅为一种逻辑功能划分，实际实现时可以有另外的划分方式。
在采用集成的单元的情况下，图10示出了上述实施例中所涉及的终端设备的一种可能的功能单元组成框图，终端设备包括:
通信单元1001，被配置成获取反射信号，其中所述反射信号是信号传输设备在发射无线信号至处于移动状态的目标对象后接收到的反射信号；
处理单元1002，被配置成对所述反射信号进行处理，得到所述目标对象的点云数据；点云数据被输入至经训练的残差网络模型以识别所述目标对象的身份。
其中，处理单元1002可以是处理器或控制器，通信单元1001可以是收发器、收发电路、射频芯片等。
在一种实现方式中，所述处理单元1002对所述反射信号进行处理，得到所述目标对象的点云数据，包括：
根据多普勒效应或调频连续波原理对所述反射信号进行处理，得到至少一个表面能量点的属性信息。
在一种实现方式中，点云数据被输入至经训练的残差网络模型之前，还包括：
所述处理单元1002获取至少一个表面能量点的标识信息，所述标识信息用于指示所述至少一个表面能量点中的第一表面能量点属于同一帧；
所述处理单元1002将连续多个帧的表面能量点的属性信息作为一个点云序列，得到至 少一个点云序列；
进一步的，点云数据被输入至经训练的残差网络模型，包括：
所述至少一个点云序列中的每个点云序列被输入至经训练的残差网络模型。
在一种实现方式中，所述处理单元1002将连续多个帧的表面能量点的属性信息作为一个点云序列，得到至少一个点云序列之前，还包括：
当属于同一帧的第一表面能量点的数量小于预设数量阈值时，所述处理单元1102增加第二表面能量点，所述第二表面能量点和所述第一表面能量点的数量总和达到所述预设数量阈值，所述第二表面能量点的标识信息和所述第一表面能量点的标识信息相同，所述第二表面能量点的属性信息是对所述第一表面能量点的属性信息进行算术平均运算得到的；
当属于同一帧的第一表面能量点的数量大于预设数量阈值时，所述处理单元1102删除所述第一表面能量点中的第三表面能量点，所述第一表面能量点中除所述第三表面能量点以外的表面能量点的数量总和达到所述预设数量阈值。
在一种实现方式中，在所述点云数据被输入至经训练的残差网络模型时，所述经训练的残差网络模型对所述点云序列所包含的表面能量点的目标属性信息进行特征提取，得到所述目标属性信息所指示的目标属性的属性特征，所述目标属性信息为所述表面能量点的任一属性信息；
将每个属性的属性特征进行处理，得到所述步态特征。
在一种实现方式中，经训练的残差网络模型对所述点云序列所包含的表面能量点的目标属性信息进行特征提取，得到所述目标属性信息所指示的目标属性的属性特征，包括：
使用第一预设卷积层对所述点云序列所包含的表面能量点的目标属性信息进行卷积运算，得到第一卷积运算结果；
对所述第一卷积运算结果进行最大池化操作，得到第一操作结果；
使用第二预设卷积层对所述第一操作结果进行卷积运算，得到第二卷积运算结果；
使用所述第二预设卷积层对所述第二卷积运算结果进行卷积运算，得到第三卷积运算结果；
使用所述第二预设卷积层对所述第三卷积运算结果进行卷积运算，得到第四卷积运算结果；
使用所述第二预设卷积层对所述第四卷积运算结果进行卷积运算，得到第五卷积运算结果；
将所述第三卷积运算结果和所述第一操作结果进行残差操作，得到第二操作结果；
将所述第五卷积运算结果和所述第三卷积运算结果进行残差操作，得到第三操作结果；
对所述第二操作结果和所述第三操作结果进行平均池化操作，得到所述目标属性的属性特征。
在一种实现方式中，所述处理单元1002将每个属性的属性特征进行处理，得到所述步态特征，包括：
针对任一点云序列，将每个属性信息的属性特征进行连接，得到连接后的属性特征；
使用预设全连接层对至少一个连接后的属性特征进行分类，得到所述步态特征。
在一种实现方式中，所述至少一个表面能量点中每个表面能量点的属性信息包括：所述表面能量点向所述信号传输设备的方向移动的移动速度，所述表面能量点与所述信号传输设备之间的距离，所述反射信号与所述信号传输设备之间的角度，信噪比，所述目标对象的位置，所述目标对象的物理尺寸，点密度中的一种或多种。
在一种实现方式中，所述处理单元1002根据多普勒效应对所述反射信号进行处理，得到至少一个表面能量点的属性信息，包括：
获取所述反射信号的多普勒红移和多普勒蓝移；
根据所述多普勒红移和所述多普勒蓝移，计算得到所述至少一个表面能量点向所述信号传输设备的方向移动的移动速度。
在一种实现方式中，所述处理单元1002根据调频连续波原理对所述反射信号进行处理，得到至少一个表面能量点的属性信息，包括：
获取所述信号传输设备接收所述反射信号的接收时间和所述信号传输设备发射所述无线信号的发射时间之间的差值；
获取所述目标对象处于静止状态时的频差，以及所述目标对象处于移动状态时的多普勒频移；
根据所述目标对象处于静止状态时的频差和所述多普勒频移，得到所述目标对象处于移动状态时的频差；
将所述目标对象处于移动状态时的频差与所述差值相乘，得到所述至少一个表面能量点与所述信号传输设备之间的距离。
在一种实现方式中，所述处理单元1002对所述反射信号进行处理，得到所述目标对象的点云数据，包括：
获取所述信号传输设备的接收天线的相位；
根据所述接收天线的相位，计算得到所述反射信号与所述信号传输设备之间的角度。
在一种实现方式中，所述处理单元1002对所述反射信号进行处理，得到所述目标对象的点云数据，包括：
获取所述反射信号的输出功率，以及所述信号传输设备在接收所述反射信号的过程中接收到的噪声的输出功率；
根据所述反射信号的输出功率和所述噪声的输出功率，计算得到信噪比。
在一种实现方式中，所述处理单元1002信号传输设备发射的无线信号为毫米波，所述信号传输设备接收到的反射信号为毫米波。
当处理单元1002为处理器，通信单元1001为通信接口时，本申请实施例所涉及的终端设备可以为图11所示的终端设备。
本申请实施例中，通信单元1001获取反射信号，其中所述反射信号是信号传输设备在发射无线信号至处于移动状态的目标对象后接收到的反射信号，处理单元1002对所述反射信号进行处理，得到所述目标对象的点云数据，点云数据被输入至经训练的残差网络模型对以识别所述目标对象的身份，可提高身份识别的安全性。
本申请实施例还提供了一种计算机可读存储介质，其中，所述计算机可读存储介质存储用于电子数据交换的计算机程序，其中，所述计算机程序使得计算机执行如上述方法实施例中终端设备所描述的部分或全部步骤。
本申请实施例还提供了一种计算机程序产品，其中，所述计算机程序产品包括存储了计算机程序的非瞬时性计算机可读存储介质，所述计算机程序可操作来使计算机执行如上述方法实施例中终端设备所描述的部分或全部步骤。该计算机程序产品可以为一个软件安装包。
本申请实施例所描述的方法或者算法的步骤可以以硬件的方式来实现，也可以是由处理器执行软件指令的方式来实现。软件指令可以由相应的软件模块组成，软件模块可以被存放于随机存取存储器(Random Access Memory，RAM)、闪存、只读存储器(Read Only Memory，ROM)、可擦除可编程只读存储器(Erasable Programmable ROM，EPROM)、电可擦可编程只读存储器(Electrically EPROM，EEPROM)、寄存器、硬盘、移动硬盘、只读光盘(CD-ROM)或者本领域熟知的任何其它形式的存储介质中。一种示例性的存储介质耦合至处理器，从而使处理器能够从该存储介质读取信息，且可向该存储介质写入信息。 当然，存储介质也可以是处理器的组成部分。处理器和存储介质可以位于ASIC中。另外，该ASIC可以位于接入网设备、目标网络设备或核心网设备中。当然，处理器和存储介质也可以作为分立组件存在于接入网设备、目标网络设备或核心网设备中。
本领域技术人员应该可以意识到，在上述一个或多个示例中，本申请实施例所描述的功能可以全部或部分地通过软件、硬件、固件或者其任意组合来实现。当使用软件实现时，可以全部或部分地以计算机程序产品的形式实现。所述计算机程序产品包括一个或多个计算机指令。在计算机上加载和执行所述计算机程序指令时，全部或部分地产生按照本申请实施例所述的流程或功能。所述计算机可以是通用计算机、专用计算机、计算机网络、或者其他可编程装置。所述计算机指令可以存储在计算机可读存储介质中，或者从一个计算机可读存储介质向另一个计算机可读存储介质传输，例如，所述计算机指令可以从一个网站站点、计算机、服务器或数据中心通过有线(例如同轴电缆、光纤、数字用户线(Digital Subscriber Line，DSL))或无线(例如红外、无线、微波等)方式向另一个网站站点、计算机、服务器或数据中心进行传输。所述计算机可读存储介质可以是计算机能够存取的任何可用介质或者是包含一个或多个可用介质集成的服务器、数据中心等数据存储设备。所述可用介质可以是磁性介质(例如，软盘、硬盘、磁带)、光介质(例如，数字视频光盘(Digital Video Disc，DVD))、或者半导体介质(例如，固态硬盘(Solid State Disk，SSD))等。
以上所述的具体实施方式，对本申请实施例的目的、技术方案和有益效果进行了进一步详细说明，所应理解的是，以上所述仅为本申请实施例的具体实施方式而已，并不用于限定本申请实施例的保护范围，凡在本申请实施例的技术方案的基础之上，所做的任何修改、等同替换、改进等，均应包括在本申请实施例的保护范围之内。

Claims (16)

  1. 一种身份识别方法，其特征在于，包括：
    获取反射信号，其中所述反射信号是信号传输设备在发射无线信号至处于移动状态的目标对象后接收到的反射信号；
    对所述反射信号进行处理，得到所述目标对象的点云数据；
    所述点云数据被输入至经训练的残差网络模型以识别所述目标对象的身份。
  2. 如权利要求1所述的方法，其特征在于，所述信号传输设备发射的无线信号为毫米波，所述信号传输设备接收到的反射信号为毫米波。
  3. 如权利要求1所述的方法，其特征在于，所述对所述反射信号进行处理，得到所述目标对象的点云数据，包括：
    根据多普勒效应或调频连续波原理对所述反射信号进行处理，得到至少一个表面能量点的属性信息。
  4. 如权利要求3所述的方法，其特征在于，所述点云数据被输入至经训练的残差网络模型之前，还包括：
    获取至少一个表面能量点的标识信息，所述标识信息用于指示所述至少一个表面能量点中的第一表面能量点属于同一帧；
    将连续多个帧的表面能量点的属性信息作为一个点云序列，得到至少一个点云序列；
    所述点云数据被输入至经训练的残差网络模型，包括：
    所述至少一个点云序列中的每个点云序列被输入至经训练的残差网络模型。
  5. 如权利要求4所述的方法，其特征在于，所述将连续多个帧的表面能量点的属性信息作为一个点云序列，得到至少一个点云序列之前，还包括：
    当属于同一帧的第一表面能量点的数量小于预设数量阈值时，增加第二表面能量点，所述第二表面能量点和所述第一表面能量点的数量总和达到所述预设数量阈值，所述第二表面能量点的标识信息和所述第一表面能量点的标识信息相同，所述第二表面能量点的属性信息是对所述第一表面能量点的属性信息进行算术平均运算得到的；
    当属于同一帧的第一表面能量点的数量大于预设数量阈值时，删除所述第一表面能量点中的第三表面能量点，所述第一表面能量点中除所述第三表面能量点以外的表面能量点的数量总和达到所述预设数量阈值。
  6. 如权利要求4所述的方法，其特征在于，在所述点云数据被输入至经训练的残差网络模型时，所述经训练的残差网络模型对所述点云序列所包含的表面能量点的目标属性信息进行特征提取，得到所述目标属性信息所指示的目标属性的属性特征，所述目标属性信息为所述表面能量点的任一属性信息；
    将每个属性的属性特征进行处理，得到步态特征。
  7. 如权利要求6所述的方法，其特征在于，所述经训练的残差网络模型对所述点云序列所包含的表面能量点的目标属性信息进行特征提取，得到所述目标属性信息所指示的目标属性的属性特征，包括：
    使用第一预设卷积层对所述点云序列所包含的表面能量点的目标属性信息进行卷积运算，得到第一卷积运算结果；
    对所述第一卷积运算结果进行最大池化操作，得到第一操作结果；
    使用第二预设卷积层对所述第一操作结果进行卷积运算，得到第二卷积运算结果；
    使用所述第二预设卷积层对所述第二卷积运算结果进行卷积运算，得到第三卷积运算结果；
    使用所述第二预设卷积层对所述第三卷积运算结果进行卷积运算，得到第四卷积运算结果；
    使用所述第二预设卷积层对所述第四卷积运算结果进行卷积运算，得到第五卷积运算结果；
    将所述第三卷积运算结果和所述第一操作结果进行残差操作，得到第二操作结果；
    将所述第五卷积运算结果和所述第三卷积运算结果进行残差操作，得到第三操作结果；
    对所述第二操作结果和所述第三操作结果进行平均池化操作，得到所述目标属性的属性特征。
  8. 如权利要求6所述的方法，其特征在于，所述将每个属性的属性特征进行处理，得到步态特征，包括：
    针对任一点云序列，将每个属性信息的属性特征进行连接，得到连接后的属性特征；
    使用预设全连接层对至少一个连接后的属性特征进行分类，得到所述步态特征。
  9. 如权利要求3所述的方法，其特征在于，所述至少一个表面能量点中每个表面能量点的属性信息包括：所述表面能量点向所述信号传输设备的方向移动的移动速度，所述表面能量点与所述信号传输设备之间的距离，所述反射信号与所述信号传输设备之间的角度，信噪比，所述目标对象的位置，所述目标对象的物理尺寸，点密度中的一种或多种。
  10. 如权利要求3所述的方法，其特征在于，所述根据多普勒效应对所述反射信号进行处理，得到至少一个表面能量点的属性信息，包括：
    获取所述反射信号的多普勒红移和多普勒蓝移；
    根据所述多普勒红移和所述多普勒蓝移，计算得到所述至少一个表面能量点向所述信号传输设备的方向移动的移动速度。
  11. 如权利要求3所述的方法，其特征在于，所述根据调频连续波原理对所述反射信号进行处理，得到至少一个表面能量点的属性信息，包括：
    获取所述信号传输设备接收所述反射信号的接收时间和所述信号传输设备发射所述无线信号的发射时间之间的差值；
    获取所述目标对象处于静止状态时的频差，以及所述目标对象处于移动状态时的多普勒频移；
    根据所述目标对象处于静止状态时的频差和所述多普勒频移，得到所述目标对象处于移动状态时的频差；
    将所述目标对象处于移动状态时的频差与所述差值相乘，得到所述至少一个表面能量点与所述信号传输设备之间的距离。
  12. 如权利要求1所述的方法，其特征在于，所述对所述反射信号进行处理，得到所述目标对象的点云数据，包括：
    获取所述信号传输设备的接收天线的相位；
    根据所述接收天线的相位，计算得到所述反射信号与所述信号传输设备之间的角度。
  13. 如权利要求1所述的方法，其特征在于，所述对所述反射信号进行处理，得到所述目标对象的点云数据，包括：
    获取所述反射信号的输出功率，以及所述信号传输设备在接收所述反射信号的过程中接收到的噪声的输出功率；
    根据所述反射信号的输出功率和所述噪声的输出功率，计算得到信噪比。
  14. 一种终端设备，其特征在于，所述终端设备包括用于实现如权1-13任一项所述的身份识别方法的单元。
  15. 一种终端设备，其特征在于，所述终端设备包括处理器和存储器，所述处理器与所述存储器耦合，其特征在于，
    所述存储器，被配置成存储指令；
    所述处理器，被配置成获取反射信号，其中所述反射信号是信号传输设备在发射无线信号至处于移动状态的目标对象后接收到的反射信号；对所述反射信号进行处理，得到所述目标对象的点云数据；所述点云数据被输入至经训练的残差网络模型以识别所述目标对象的身份。
  16. 一种计算机存储介质，其特征在于，所述计算机存储介质存储有计算机程序或指令，当所述程序或指令被处理器执行时，使所述处理器执行如权利要求1-13中任一项所述的身份识别方法。
PCT/CN2019/096053 2019-07-15 2019-07-15 身份识别方法、终端设备及计算机存储介质 WO2021007760A1 (zh)

Priority Applications (2)

Application Number Priority Date Filing Date Title
CN201980092409.1A CN113439274A (zh) 2019-07-15 2019-07-15 身份识别方法、终端设备及计算机存储介质
PCT/CN2019/096053 WO2021007760A1 (zh) 2019-07-15 2019-07-15 身份识别方法、终端设备及计算机存储介质

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
PCT/CN2019/096053 WO2021007760A1 (zh) 2019-07-15 2019-07-15 身份识别方法、终端设备及计算机存储介质

Publications (1)

Publication Number Publication Date
WO2021007760A1 true WO2021007760A1 (zh) 2021-01-21

Family

ID=74210111

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/CN2019/096053 WO2021007760A1 (zh) 2019-07-15 2019-07-15 身份识别方法、终端设备及计算机存储介质

Country Status (2)

Country Link
CN (1) CN113439274A (zh)
WO (1) WO2021007760A1 (zh)

Cited By (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN112836194A (zh) * 2021-01-29 2021-05-25 西安交通大学 一种基于人手内部生物特征的身份认证方法及系统

Citations (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN107507269A (zh) * 2017-07-31 2017-12-22 广东欧珀移动通信有限公司 个性化三维模型生成方法、装置和终端设备
US10121064B2 (en) * 2015-04-16 2018-11-06 California Institute Of Technology Systems and methods for behavior detection using 3D tracking and machine learning

Patent Citations (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US10121064B2 (en) * 2015-04-16 2018-11-06 California Institute Of Technology Systems and methods for behavior detection using 3D tracking and machine learning
CN107507269A (zh) * 2017-07-31 2017-12-22 广东欧珀移动通信有限公司 个性化三维模型生成方法、装置和终端设备

Non-Patent Citations (2)

* Cited by examiner, † Cited by third party
Title
XU, TIANSHUI: "Information & Technology, China Master's Theses Full-Text Database", GAIT RECOGNITION BASED ON DEPTH INFORMATION, no. 03, 15 March 2015 (2015-03-15), DOI: 20200225103139Y *
YANG, WENHUI: "Information & Technology, China Master's Theses Full-Text Database", STUDY ON OBJECT RECOGNITION METHOD BASED ON LIDAR POINT CLOUD IMAGE, no. 01, 15 January 2019 (2019-01-15), DOI: 20200225102417Y *

Cited By (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN112836194A (zh) * 2021-01-29 2021-05-25 西安交通大学 一种基于人手内部生物特征的身份认证方法及系统
CN112836194B (zh) * 2021-01-29 2023-03-21 西安交通大学 一种基于人手内部生物特征的身份认证方法及系统

Also Published As

Publication number Publication date
CN113439274A (zh) 2021-09-24

Similar Documents

Publication Publication Date Title
US11550031B2 (en) Method and apparatus for biometric authentication using face radar signal
JP7062878B2 (ja) 情報処理方法および情報処理装置
US11670110B2 (en) User authentication using mm-wave sensor for automotive radar systems
Li et al. Deep AI enabled ubiquitous wireless sensing: A survey
US20200050871A1 (en) Method and apparatus for integration of detected object identifiers and semantic scene graph networks for captured visual scene behavior estimation
CN114862837A (zh) 基于改进YOLOv5s的人体安检图像检测方法及系统
Ni et al. Human identification based on natural gait micro‐Doppler signatures using deep transfer learning
WO2020243969A1 (zh) 人脸识别的装置、方法和电子设备
Lim et al. Detection and localization of people inside vehicle using impulse radio ultra-wideband radar sensor
CN116106855B (zh) 摔倒检测方法及摔倒检测装置
CN111095297A (zh) 人脸识别的装置、方法和电子设备
Zhang et al. A novel ship target detection algorithm based on error self-adjustment extreme learning machine and cascade classifier
CN112348777A (zh) 人体目标的检测方法、装置及终端设备
Mohsin Riaz et al. Through‐wall image enhancement based on singular value decomposition
Zhou et al. Multi-modal face authentication using deep visual and acoustic features
Wang et al. Stpointgcn: Spatial temporal graph convolutional network for multiple people recognition using millimeter-wave radar
CN210166794U (zh) 人脸识别的装置和电子设备
Hof et al. Face verification using mmWave radar sensor
WO2021007760A1 (zh) 身份识别方法、终端设备及计算机存储介质
Zhou et al. Machine learning-based multipath components clustering and cluster characteristics analysis in high-speed railway scenarios
Gao et al. Learning to detect open carry and concealed object with 77 GHz radar
CN114417908A (zh) 一种基于多模态融合的无人机探测系统和方法
Yang et al. A lightweight multi-scale neural network for indoor human activity recognition based on macro and micro-doppler features
Singh et al. Wi-vi and li-fi based framework for human identification and vital signs detection through walls
CN114047503B (zh) 活动对象检测方法、装置、电子设备和存储介质

Legal Events

Date Code Title Description
121 Ep: the epo has been informed by wipo that ep was designated in this application

Ref document number: 19937525

Country of ref document: EP

Kind code of ref document: A1

NENP Non-entry into the national phase

Ref country code: DE

122 Ep: pct application non-entry in european phase

Ref document number: 19937525

Country of ref document: EP

Kind code of ref document: A1