WO2018228218A1

WO2018228218A1 - Identification method, computing device, and storage medium

Info

Publication number: WO2018228218A1
Application number: PCT/CN2018/089499
Authority: WO
Inventors: 王达峰
Original assignee: 腾讯科技（深圳）有限公司
Priority date: 2017-06-16
Filing date: 2018-06-01
Publication date: 2018-12-20
Also published as: CN108304757A

Abstract

Embodiments of the present application relate to the technical field of image analysis. Disclosed are an identification method, a computing device, and a storage medium. The method comprises: obtaining a video recording a target action of an individual to be identified; obtaining motion trajectories of feature points of the target action on the basis of the video; and determining an identity of the individual to be identified according to the trajectory features of the motion trajectories and sample data, wherein the sample data comprises: an identity of at least one sample individual and the trajectory features of a corresponding sample motion trajectory. The embodiments of the present application provide a technical solution for identifying an individual on the basis of an action, and enrich the technical means for identifying an individual.

Description

Identification method, computing device and storage medium

The present application claims the priority of the Chinese Patent Application, which is filed on Jun. 16, 2017, the entire disclosure of which is hereby incorporated by reference. .

Technical field

The embodiments of the present application relate to the field of image analysis technologies, and in particular, to an identity recognition method, a computing device, and a storage medium.

Background technique

Identifying an individual means determining the identity of the individual. Generally speaking, the identity of an individual can be the name of an individual (such as a name).

In a method for identifying an individual based on a face, first obtaining a face image of the individual to be identified and a face image of the target individual, and then calculating the similarity between the two face images by feature matching, when When the similarity is greater than the preset threshold, it is determined that the individual to be identified is the target individual.

Summary of the invention

The embodiments of the present application provide an identity identification method, a computing device, and a storage medium, to improve the accuracy of identity recognition. The technical solution is as follows:

According to an aspect of the present application, an identification method is provided, which is applied to a computing device, the method comprising: acquiring a video recording a target motion of an individual to be identified; and acquiring a motion trajectory of a feature point of the target motion based on the video And determining, according to the trajectory feature of the motion trajectory and the sample data, the identity of the to-be-identified individual, wherein the sample data includes: an identity of the at least one sample individual and a trajectory feature of the corresponding sample motion trajectory.

According to an aspect of the present application, a computing device is provided, comprising: a processor and a memory; the memory storing computer readable instructions that cause the processor to perform an identification method according to the present application.

According to an aspect of the present application, there is provided a non-volatile storage medium storing a data processing program, the data processing program comprising instructions that, when executed by a computing device, cause the computing device to perform according to the present application Instructions for the identification method.

The technical solution provided by the embodiment of the present application can bring the following beneficial effects:

By acquiring a video image recording the motion of the individual to be identified, by processing and analyzing the video image, obtaining an action feature of the individual to be identified, and determining the identity of the individual to be identified based on the action feature, so that identification of the individual is not necessary. Limited to face images, it provides a technical solution for identifying individuals based on actions, enriching the technical means of identifying individuals. In addition, since the face is easy to be made up and easy to accommodate, and the individual's movements are difficult to be imitated by the influence of personal habits, the identification of the individual based on the action is more accurate than the identification based on the face based on the face. high.

DRAWINGS

In order to more clearly illustrate the technical solutions in the examples of the present application, the drawings used in the description of the examples will be briefly described below. Obviously, the drawings in the following description are only some examples of the present application, For ordinary technicians, other drawings can be obtained based on these drawings without paying for creative labor.

FIG. 1A shows a schematic diagram of an application scenario according to some embodiments of the present application; FIG.

FIG. 1B is a flowchart of an identity recognition method provided by some embodiments of the present application; FIG.

2 is a schematic diagram of a sequence of frames in an action cycle provided by some embodiments of the present application;

3 is a schematic diagram of feature points provided by some embodiments of the present application;

4 is a schematic diagram of motion trajectories of feature points provided by some embodiments of the present application;

FIG. 5 is a block diagram of an identity recognition apparatus provided by some embodiments of the present application; FIG.

FIG. 6 is a schematic structural diagram of a computing device provided by some embodiments of the present application.

detailed description

In order to make the objects, technical solutions and advantages of the present application more clear, the embodiments of the present application will be further described in detail below with reference to the accompanying drawings.

In the embodiment of the present application, a technical solution for identifying an individual based on an action is provided. The action refers to the movement posture of the body part of the individual, such as the walking posture, the running posture, the swing arm posture, and the like. Since there are some differences in the actions of different individuals, individuals can be identified based on actions. For example, during walking, different individuals' step sizes, stride length, knee flexion, swing arm height, elbow curvature, etc. may be different, and they are difficult to change deliberately due to personal habits, so they can be used As a feature of identifying individuals.

In the embodiment of the present application, a video image recording the motion of the individual to be identified is acquired, and the motion image of the individual to be identified is obtained by processing and analyzing the video image, and then determining the identity of the individual to be identified based on the motion feature.

The technical solution provided by the embodiment of the present application can provide the public security department with the auxiliary identification of the criminal suspect. When a criminal suspect commits a crime and runs away, he usually camouflages the face (such as wearing a hat, a mask, or a face mask). Therefore, it is difficult for the surveillance camera to collect a clear and complete face image of the suspect. In this case, the suspect cannot be identified by the face image. However, the surveillance camera will record the suspicion of the suspect in the crime, the gait when the escape, the posture of the swing arm, and so on. Therefore, the suspect can be identified by the action. The technical solution provided by the embodiment of the present application has high practical application value in the field of public security criminal investigation.

Of course, the technical solution provided by the embodiment of the present application is also applicable to other application scenarios that have an identification requirement for an individual identity, which is not limited by the embodiment of the present application. For example, in some key laboratories, archives, or data rooms, individuals are identified. Because faces are easy to be made up and easy to accommodate, and individual actions are difficult to be imitated by personal habits, Identity is more accurate than identifying individuals based on faces.

In addition, in the method provided by the embodiment of the present application, the execution subject of each step is an identity recognition device. For example, the identification device can be a server, or a computer. The server can be a server, a server cluster consisting of multiple servers, or a cloud computing service center. For example, FIG. 1A shows a schematic diagram of an application scenario in accordance with some embodiments of the present application. As shown in FIG. 1A, application scenario 100 can include terminal devices (e.g., 108-a, 108-b, and 108-c, etc.) and identity recognition system 102. Here, the terminal device may be, for example, various smart terminals such as a mobile phone, a tablet computer, a handheld game console, and a video camera. The identity system 102 can include one or more servers. The terminal device can communicate with the identity recognition system 102 over the network 106.

In some embodiments, the terminal device can obtain a video about the individual to be identified. Based on this, the terminal device can perform an identity recognition method based on the video to determine the identity of the individual to be identified.

In some embodiments, the terminal device can upload a video about the individual to be identified to the identity recognition system 102. In this manner, the identification system 102 can perform an identification method on the video to determine the identity of the individual to be identified. For example, the identification system 102 can include an identification application 104. The identity recognition application 104 can perform an identification method. Here, the identity application 104 can be an independent application or a distributed application, which is not limited in this application. Additionally, the identification system 102 can transmit the identification results to the terminal device.

Please refer to FIG. 1B, which shows a flowchart of an identification method provided by some embodiments of the present application. The identification method can be performed in a computing device. Here, the computing device may be, for example, a terminal device or a server in the identity recognition system 102, but is not limited thereto. The method can include the following steps.

Step 101: Acquire a video recording a target action of the individual to be identified. An individual to be identified refers to an individual who needs to identify and determine his or her identity. The action refers to the movement posture of the body part of the individual, such as the walking posture, the running posture, the swing arm posture, and the like. A target action is a specific action, such as a target action being a walking posture. Illustratively, the video of the target action may be a video recording the walking position of the individual to be identified.

Step 102: Acquire a motion trajectory of a feature point of the target motion based on the video. In some embodiments, step 102 can obtain a sequence of frames within one or more target action periods. In some embodiments, step 102 may extract a sequence of frames within any one of the target action periods from the video to be identified. The action cycle is the time taken to perform a complete action. Accordingly, the target action cycle is the time taken to perform a complete target action. Taking the walking position as an example, the individual's walking posture is repetitive. For example, the left foot is the right foot, the left foot is the right foot, the left foot is the right foot, and so on. Then, the action cycle of the walking posture is the complete motion flow from the steps of lifting the left foot, taking the left foot, lowering the left foot, lifting the right foot, taking the right foot, lowering the right foot, and then returning to the left foot. time.

Since the video is composed of multi-frame pictures, the frame sequence within one target action period contains multiple frames of pictures. Taking the target action as a walking posture as an example, a multi-frame picture as shown in FIG. 2 is included in a target action cycle. In the case where the individual walks at a relatively uniform speed, the time of each target action cycle is substantially the same, that is, the number of pictures included in the sequence of frames in each target action cycle is substantially the same.

In some embodiments, in the case where only the target action of the individual to be identified is recorded in the video, step 102 may include the following sub-steps: dividing the video into a plurality of target action cycles; extracting a sequence of frames within any one of the target action cycles .

In some embodiments, the computing device may identify from the video a target picture of a specified action step in which the target action is recorded, and a time range containing a sequence of frames between adjacent two frames of the target picture as a target action period. Taking the target action as the walking posture, a specified action step of the target action is any action step included in a complete walking posture, such as lifting the left foot, taking the left foot, lowering the left foot, lifting the right foot, and taking out. Any of the action steps in the right foot. In an example, if the second frame, the seventh frame, the twelfth frame, the 17th frame, the 23rd frame, the 28th frame, and the like of the to-be-recognized video are recorded with an action step of lifting the left foot, the second frame is The 7th frame has a total of 6 frames of pictures as a target action cycle, and the 7th frame to the 12th frame have a total of 6 frames. The picture is a target action cycle, and the 12th frame to the 17th frame are a total of 6 frames. The picture is a target action cycle and the 17th frame. A total of 7 frames of pictures in the 23rd frame is a target action cycle, and a total of 6 frames of pictures from the 23rd frame to the 28th frame are a target action cycle, and so on.

In the above manner, the computing device can automatically divide the target action cycle. In other possible implementations, the computing device may also divide the video into multiple target action cycles according to the operation result for labeling the target action cycle.

In some embodiments, the video records other actions of the individual to be identified in addition to the target action of the individual to be identified. The computing device can obtain a video segment selected from the video that only records the target motion of the individual to be identified. Taking the target action as the walking posture as an example, if the walking position and the running posture of the individual to be recognized are recorded in the to-be-identified video, the computing device may acquire a video segment selected from the video and only recording the walking posture of the individual to be identified. The computing device may divide the video segment into multiple target action cycles and extract a sequence of frames within any one of the target action cycles.

In addition, since the content of the picture included in the frame sequence in each target action cycle is usually less different when the individual performs a target action more uniformly, any target action cycle may be selected as the analysis target of the identity recognition.

Step 103: Acquire a motion trajectory of a feature point of the target motion in a sequence of frames.

The feature point of the target action refers to the feature point of the body part involved in performing the target action. Taking the target action as the walking posture, the feature points may include: several feature points of the thigh part (such as the joint position of the thigh and the ankle, the outer side of the thigh, the inner side of the thigh, the connecting position of the thigh and the knee, etc.), and some of the knee parts. Feature points, several feature points of the calf part, and several feature points of the foot. As shown in FIG. 3, taking the target motion as the walking posture, each feature point of the leg is represented by a black small dot. The number and location of feature points can be set according to actual needs. Taking the target motion as the walking posture as an example, the number of feature points may be 30 to 40, and the position of the feature points may be several positions as described above. In the embodiment of the present application, the motion trajectory of the feature point in the frame sequence is used to reflect the action feature.

In some embodiments, the step includes the following substeps: identifying each feature point from each frame of the frame sequence; obtaining the position of each feature point in each frame of the frame sequence; The position of each feature point in the sequence of frames is determined at the position in each frame of the frame sequence.

In some embodiments, the size and resolution of each frame of the frame sequence are the same. Thus, step 103 can employ a uniform coordinate system to represent the location of each feature point in each frame of the picture. For example, taking the lower left corner of each frame as the origin, the bottom edge of the image is the horizontal axis, perpendicular to the bottom edge of the image and intersecting the side of the origin as the vertical axis, establishing a two-dimensional Cartesian coordinate system, each feature point is The position in any one of the frames may be represented by a combination of the abscissa and the ordinate of the feature point in the Cartesian coordinate system. In some embodiments, the computing device may acquire the horizontal and vertical coordinates of each feature point in each frame of the frame sequence, and sequentially connect the horizontal and vertical coordinates according to the order of the pictures in the frame sequence to obtain each The trajectory of the feature point in the sequence of frames.

Referring to Figure 4, a schematic diagram of the motion trajectory of a feature point is shown. A target action cycle includes the second frame to the sixth frame picture, and the coordinates of the feature points located at the ankle position in the above five frames are (x1, y1), (x2, y2), (x3, y3), ( X4, y4) and (x5, y5), the above coordinate points are sequentially connected to obtain a motion trajectory of the feature point located at the position of the ankle in a target action period.

In addition, in the embodiment of the present application, the algorithm used for feature point location is not limited, and the related algorithm used for the location of the feature point of the face may be referred to. For example, feature point localization algorithm based on statistical learning, feature point localization algorithm based on principal component analysis, feature point localization algorithm based on Point Distribution Model (PDM), feature point localization algorithm using shape estimation, based on gray scale Feature point location algorithm for information, and so on.

Step 104, extracting trajectory features of the above motion trajectory.

The trajectory feature refers to the characteristics of the motion trajectory. For example, the trajectory feature includes at least one of the following: coordinates of a plurality of feature points on the motion trajectory, curvature of the motion trajectory, length of the motion trajectory, and the like. Exemplarily, the coordinates of the feature point may be the coordinates of the feature point of the target action in each frame picture, and the arc of the motion track may be extracted from each arc position in the motion track, and the length of the motion track may be adopted by the motion track. The number of pixels is represented.

Step 105: Determine the identity of the individual to be identified according to the trajectory feature of the motion trajectory and the sample data.

The sample data includes: an identity of at least one sample individual and a trajectory feature of the corresponding sample motion trajectory. The sample motion trajectory refers to a motion trajectory of a feature point of the target motion in a sequence of frames in which an action period in which the sample individual performs the target motion is recorded. For example, a sample video of a sample individual is obtained in advance, a target motion of the sample individual is recorded in the sample video, a plurality of target motion cycles are extracted from the sample video, and a sample motion trajectory can be extracted from the frame sequence in a target motion cycle. In addition, in order to ensure the accuracy of the identification, for a sample individual, a plurality of sample motion trajectories corresponding to the sample individual are usually acquired.

In some embodiments, step 105 includes the following sub-steps: detecting whether there is a sample motion trajectory matching the motion trajectory according to the trajectory feature of the motion trajectory and the trajectory feature of the sample motion trajectory corresponding to each sample individual; When the motion trajectory matches the sample motion trajectory, the identity of the sample individual corresponding to the sample motion trajectory matching the motion trajectory is determined as the identity of the individual to be identified.

For a sample individual, if the number of sample motion trajectories corresponding to the sample individual is one, the trajectory feature of the sample motion trajectory corresponding to the sample individual is the trajectory feature extracted from the motion trajectory of the sample. For a sample individual, if the number of sample motion trajectories corresponding to the sample individual is multiple, the trajectory features may be extracted from each sample motion trajectory separately, and the extracted trajectory features may be integrated (for example, respectively The average of each trajectory feature is obtained, and the trajectory feature of the sample motion trajectory corresponding to the sample individual is obtained.

In addition, for a sample individual, the computing device can calculate the similarity between the two according to the trajectory feature of the motion trajectory and the trajectory feature of the sample motion trajectory corresponding to the sample individual. When the similarity is greater than the preset threshold, the computing device determines that the motion trajectory matches the sample motion trajectory. When the similarity is less than the preset threshold, the computing device determines that the motion trajectory does not match the sample motion trajectory. The preset threshold is an empirical value set according to requirements, for example, the preset threshold is 95%.

In some embodiments, in a case where the number of sample individuals is plural, the computing device may select a sample motion trajectory with the highest similarity between the motion trajectories and greater than a preset threshold as a sample matching the motion trajectory. Movement track. Illustratively, the preset threshold is 95%. In one example, the number of sample individuals is one, and the identity of the sample individual is Zhang San, assuming that the similarity between the corresponding motion trajectory of the individual to be identified and the sample motion trajectory corresponding to the sample individual is 96%, then Determine the identity of the individual to be identified as Zhang San. In another example, the number of sample individuals is three, and the identity of the three sample individuals is Zhang San, Li Si, and Wang Wu, respectively, and the corresponding motion trajectory of the individual to be identified corresponds to the sample of the above three sample individuals. The similarities between the motion trajectories are 96%, 70%, and 99%, respectively, and the identity of the individual to be identified is determined to be Wang Wu.

In some embodiments, step 105 includes using the trajectory feature of the motion trajectory as an input to the identity recognition model and using the identity recognition model to determine the identity of the individual to be identified.

The identity model is trained based on sample data. See below for an introduction to the training process for the identity model. The computing device inputs the trajectory feature of the motion trajectory into the identity recognition model, and the trajectory feature of the motion trajectory is processed and calculated by the identity recognition model, and the output result of the model is the identity of the individual to be identified.

Taking the identification model as an example of a neural network, the neural network includes an input layer, at least one hidden layer, and an output layer. The input layer includes a plurality of input nodes, each of which corresponds to a trajectory feature. The output layer includes at least one output node, each output node corresponding to an identity. The hidden layer is located between the input layer and the output layer and is connected to the input layer and the output layer, respectively. The process of using the neural network for identification is as follows: the trajectory features of the motion trajectory corresponding to the individual to be identified are input to the input layer of the neural network, and the trajectory features are combined and abstracted by the hidden layer to obtain data suitable for classification by the output layer. Finally, the identity of the individual to be identified is output by the output layer. The above is only an example of constructing an identity recognition model using a neural network. In practical applications, other algorithms may be selected to construct an identity recognition model.

In the embodiment of the present application, the identity of the individual may be the name of the individual, for example, the identity of the individual is represented by a name. In some embodiments, the identity of the individual is, for example, the name of the individual. The computing device can identify the individual identity based on the walking posture. The sample data includes the names of multiple sample individuals such as Zhang San, Li Si, Wang Wu, Zhao Liu, and Sun Qi, as well as the trajectory features extracted from the sample motion trajectory corresponding to each sample individual. In this way, the computing device can train the identification data using the sample data described above, and the identification model can be used to determine the name of the individual to be identified.

In some embodiments, the sample data further includes identity association information corresponding to each sample individual, and the identity association information includes personal information such as age, gender, contact information, occupation, address, etc., after determining the identity of the individual to be identified, The identity data of the individual to be identified is obtained in the sample data. The above identity association information may be collected in advance and stored in the sample data.

It should be noted that, in order to ensure the accuracy of the identification, it should be ensured that the body orientation of the individual to be identified in the extracted target action cycle is the same as the body orientation of the sample individual in the corresponding target action cycle, for example, both toward the left side, or All face the right side, or both face forward and so on. In practical applications, when the conditions permit, a sample action cycle of a plurality of different body orientations is recorded in the sample video of each sample individual, and when the identification of the individual to be identified is performed, the target to be identified is first determined in the extracted target. The body orientation within the action cycle, and then the identified individual is identified using sample data (or an identification model) that is consistent with the body orientation.

It should also be noted that in the above example of identifying an individual based on a walking posture, taking the relevant data of the leg as an example. In practical applications, when identifying an individual based on a walking posture, the computing device may collect only relevant data of the leg, and may also collect relevant data of the leg and the upper limb. The feature points of the upper limb may include: several feature points of the upper arm part (such as the joint position of the arm and the shoulder joint, the middle position of the upper arm, the joint position of the upper arm and the elbow joint, etc.), and several characteristic points of the forearm part (such as the forearm and The articulation position of the elbow joint, the middle of the forearm, the position of the forearm and the wrist joint, etc.). In addition, the acquisition of the trajectory of the feature points of the upper limbs and the extraction of the corresponding trajectory features are the same as those of the legs, as described above. When the individual is identified based on the walking posture, the relevant data of the leg and the upper limb is integrated, and the recognition accuracy is improved compared to the relevant data considering only the leg.

In summary, the method provided by the embodiment of the present invention obtains a motion of a to-be-identified individual by acquiring a video recording an action of the individual to be identified, and then determining the to-be-identified function based on the motion feature. The identity of the individual makes the identification of the individual not necessarily limited to the face image, and provides a technical solution for identifying the individual based on the action, enriching the technical means for identifying the individual.

In addition, since the face is easy to be made up and easy to accommodate, and the individual's movements are difficult to be imitated by the influence of personal habits, the identification of the individual based on the action is more accurate than the identification based on the face based on the face. high.

The following describes the training process of the identity model. The training process can include the following steps.

Step 201: Construct a training sample set according to the sample data, where the training sample set includes a plurality of training samples.

Each training sample includes: a trajectory feature extracted from a sample motion trajectory corresponding to a sample individual, and an identity of the sample individual.

The source data acquired by the computing device may be a sample video of the sample individual. For each sample individual, the computing device divides the sample video of the target action recorded by the sample individual into a plurality of target action cycles. The computing device can extract a sequence of frames within one or more target action cycles. For each target action cycle, the computing device can acquire the motion trajectory of each feature point of the target action in the frame sequence of the target action cycle, and extract the trajectory feature, thereby combining the identity of the sample individual to obtain a training sample. For the description of the division of the action cycle, the feature point location, the motion trajectory extraction, and the trajectory feature extraction, refer to the description in the embodiment of FIG. 1B, which is not described in this embodiment. The above source data may be collected in advance and stored in a computing device.

In some embodiments, the identity of the individual is the name of the individual, and the identification of the individual's identity based on the walking posture is taken as an example. The sample data includes the names of multiple sample individuals such as Zhang San, Li Si, Wang Wu, Zhao Liu, and Sun Qi, as well as the trajectory features extracted from the sample motion trajectory corresponding to each sample individual. Taking Zhang San as an example, the computing device divides the sample video recorded with the three-three walking posture into a plurality of action cycles, and extracts a sequence of frames in one or more action cycles. For each action cycle, the computing device can acquire the motion trajectory of each feature point of the walking posture in the frame sequence in the action period, and extract the trajectory feature, thereby combining the name of the sample individual of "Zhang San" to obtain a training. sample. The computing device can acquire a plurality of training samples related to Zhang San. Similarly, training samples of other sample individuals such as Li Si, Wang Wu, Zhao Liu, and Sun Qi are also obtained in the above manner.

In some embodiments, when the name of the individual is identified, if the computing device only acquires the training samples associated with one sample individual (eg, Zhang San), the identification model obtained by the subsequent training can be used to determine the name of the individual to be identified. Whether it is Zhang San.

In step 202, the training sample is trained by using a machine learning algorithm to obtain an identity recognition model.

In the embodiment of the present application, the machine learning algorithm may adopt a Bayesian algorithm, a support vector machine (SVM) algorithm, a decision tree algorithm, a neural network algorithm, a deep learning algorithm, and the like. Make a limit.

The computing device can input the trajectory feature of the sample motion trajectory corresponding to the sample individual and the identity of the sample individual into the identity recognition model, and train the model by using a machine learning algorithm, and finally obtain an identity recognition model whose accuracy meets the requirement.

In some embodiments, to ensure the accuracy of the identity model, step 202 can verify the identity model in the following manner.

Step 202 can construct a verification sample set based on the verification data. The verification data includes: at least one identity of the verification individual and a corresponding trajectory feature of the verification motion trajectory. The verification motion trajectory refers to a motion trajectory of each feature point of the target motion in a sequence of frames in which an action period in which the verification individual performs the target motion is recorded. For example, the verification video of the verification individual is obtained in advance, the target motion of the verification individual is recorded in the verification video, and a plurality of target action cycles are extracted from the verification video, and a verification motion track may be extracted from the frame sequence in one target action cycle.

The validation sample set includes multiple validation samples that are used to validate the model. The verification sample is also called a test sample. Each verification sample includes: a trajectory feature extracted from a verification motion trajectory corresponding to a verification individual, and the identity of the verification individual.

For each verification sample, step 202 may use the trajectory feature of the verification motion trajectory corresponding to the verification sample as an input of the identity recognition model, and use the identity recognition model to determine the identity of the verification individual.

Step 202 may determine the accuracy of the identity recognition model according to the identity of each verification individual output by the identity recognition model and the identity of each verification individual recorded in the verification sample.

For example, if the number of verified individuals is 100, wherein the identity of the 95 verified individuals output by the identity model is accurate, and the identity of the other 5 verified individuals is wrong, the accuracy of the identity model is 95%. In the event that the accuracy of the identity model reaches a predetermined requirement, step 202 stops training. In the event that the accuracy of the identity model does not meet the preset requirements, step 202 may continue to train the identity model with more training samples.

In summary, the method provided by the embodiment of the present application can obtain the identity recognition model according to the sample data, and adopt the modeling method for identity recognition, which helps to improve the accuracy of the identity recognition.

The following is an embodiment of the apparatus of the present application, which may be used to implement the method embodiments of the present application. For details not disclosed in the device embodiments of the present application, please refer to the method embodiments of the present application.

Please refer to FIG. 5, which shows a block diagram of an identification device provided by some embodiments of the present application. The device has the function of implementing the above method examples. The functions may be implemented by hardware, or may be implemented by hardware by executing corresponding software. The identification device can reside, for example, in a computing device. The apparatus may include: a video acquisition module 501, a frame sequence extraction module 502, a trajectory acquisition module 503, a feature extraction module 504, and an identity determination module 505.

The video obtaining module 501 is configured to acquire a video to be identified that records a target action of the individual to be identified.

The frame sequence extraction module 502 is configured to extract a sequence of frames in any one of the target action periods from the to-be-identified video, where the target action period refers to a time taken to perform a complete target action.

The trajectory obtaining module 503 is configured to acquire a motion trajectory of each feature point of the target motion in the sequence of frames.

The feature extraction module 504 is configured to extract a trajectory feature of the motion trajectory.

The identity determining module 505 is configured to determine an identity of the to-be-identified individual according to the trajectory feature and the sample data of the motion trajectory, where the sample data includes: an identity of the at least one sample individual and a trajectory of the corresponding sample motion trajectory feature.

In some embodiments, the identity determining module 505 is configured to detect whether there is a sample matching the motion trajectory according to a trajectory feature of the motion trajectory and a trajectory feature of a sample motion trajectory corresponding to each sample individual. a motion trajectory; if there is a sample motion trajectory matching the motion trajectory, determining an identity of the sample individual corresponding to the sample motion trajectory matching the motion trajectory as the identity of the to-be-identified individual.

In some embodiments, the identity determining module 505 is configured to: use the trajectory feature of the motion trajectory as an input of an identity recognition model, and determine the identity of the to-be-identified entity by using the identity recognition model; wherein the identity recognition The model is trained based on the sample data.

In some embodiments, the apparatus further includes: a sample building module and a model training module.

a sample construction module, configured to construct a training sample set according to the sample data, where the training sample set includes a plurality of training samples, each training sample includes: a trajectory feature extracted from a sample motion trajectory corresponding to a sample individual, and a The identity of the sample individual.

And a model training module, configured to train the training sample by using a machine learning algorithm to obtain the identity recognition model.

In some embodiments, the trajectory acquisition module 503 includes: a feature recognition unit, a location acquisition unit, and a trajectory acquisition unit.

And a feature recognition unit, configured to identify each of the feature points from each frame of the frame sequence.

And a location acquiring unit, configured to acquire a location of each feature point in each frame of the frame sequence.

And a trajectory acquiring unit, configured to determine a motion trajectory of each feature point in the frame sequence according to a position of each feature point in each frame of the frame sequence.

In some embodiments, the frame sequence extraction module 502 includes: a period dividing unit and a frame sequence extracting unit.

And a period dividing unit, configured to divide the to-be-identified video into multiple target action periods.

The frame sequence extracting unit is configured to extract a sequence of frames in any one of the target action periods.

It should be noted that, when the device provided by the foregoing embodiment implements its function, only the division of the foregoing functional modules is illustrated. In actual applications, the function distribution may be completed by different functional modules according to requirements, that is, the device is required. The internal structure is divided into different functional modules to perform all or part of the functions described above. In addition, the apparatus and method embodiments provided in the foregoing embodiments are in the same concept, and the specific implementation process is described in detail in the method embodiment, and details are not described herein again.

Please refer to FIG. 6 , which is a schematic structural diagram of a computing device provided by some embodiments of the present application. The computing device is for implementing the identity recognition method provided in the above embodiments. Specifically:

The computing device 600 includes a central processing unit (CPU) 601, a system memory 604 including a random access memory (RAM) 602 and a read only memory (ROM) 603, and a system bus 605 that connects the system memory 604 and the central processing unit 601. . The computing device 600 also includes a basic input/output system (I/O system) 606 that facilitates transfer of information between various devices within the computer, and a large capacity for storing the operating system 613, applications 614, and other program modules 616. Storage device 607.

The basic input/output system 606 includes a display 608 for displaying information and an input device 609 such as a mouse or keyboard for user input of information. The display 608 and input device 609 are both connected to the central processing unit 601 via an input and output controller 610 that is coupled to the system bus 605. The basic input/output system 606 can also include an input output controller 610 for receiving and processing input from a plurality of other devices, such as a keyboard, mouse, or electronic stylus. Similarly, input and output controller 610 also provides output to a display screen, printer, or other type of output device.

The mass storage device 607 is connected to the central processing unit 601 by a mass storage controller (not shown) connected to the system bus 605. The mass storage device 607 and its associated computer readable medium provide non-volatile storage for the computing device 600. That is, the mass storage device 607 can include a computer readable medium (not shown) such as a hard disk or a CD-ROM drive.

Without loss of generality, the computer readable medium can include computer storage media and communication media. Computer storage media includes volatile and nonvolatile, removable and non-removable media implemented in any method or technology for storage of information such as computer readable instructions, data structures, program modules or other data. Computer storage media include RAM, ROM, EPROM, EEPROM, flash memory or other solid state storage technologies, CD-ROM, DVD or other optical storage, tape cartridges, magnetic tape, magnetic disk storage or other magnetic storage devices. Of course, those skilled in the art will appreciate that the computer storage medium is not limited to the above. The system memory 604 and mass storage device 607 described above may be collectively referred to as a memory.

According to various embodiments of the present application, computing device 600 may also be operated by a remote computer connected to the network via a network such as the Internet. That is, the computing device 600 can be connected to the network 612 through a network interface unit 611 connected to the system bus 605, or can be connected to other types of networks or remote computer systems using the network interface unit 611 (not shown) ).

Storing at least one instruction, at least one program, code set or instruction set in the memory, the at least one instruction, the at least one program, the code set or the instruction set being loaded and executed by the processor to implement the above Identification method.

In an exemplary embodiment, there is also provided a computer readable storage medium having stored therein at least one instruction, at least one program, a code set or a set of instructions, the at least one instruction, the at least one program The code set or instruction set is loaded and executed by a processor of the server to implement the various steps in the above method embodiments. In some embodiments, the computer readable storage medium described above can be a ROM, a random access memory (RAM), a CD-ROM, a magnetic tape, a floppy disk, and an optical data storage device.

In an exemplary embodiment, there is also provided a computer program product for performing the functions of the various steps in the above method embodiments when the computer program product is executed.

It should be understood that "a plurality" as referred to herein means two or more. "and/or", describing the association relationship of the associated objects, indicating that there may be three relationships, for example, A and/or B, which may indicate that there are three cases where A exists separately, A and B exist at the same time, and B exists separately. The character "/" generally indicates that the contextual object is an "or" relationship. The serial numbers of the embodiments of the present application are merely for the description, and do not represent the advantages and disadvantages of the embodiments.

The above description is only exemplary embodiments of the present application, and is not intended to limit the present application. Any modifications, equivalents, improvements, etc. made within the spirit and principles of the present application are included in the protection of the present application. Within the scope.

Claims

An identification method is performed by a computing device, the method comprising:

Obtaining a video recording the target action of the individual to be identified;

Obtaining a motion trajectory of a feature point of the target motion based on the video;

And determining, according to the trajectory feature of the motion trajectory and the sample data, the identity of the to-be-identified individual, wherein the sample data includes: an identity of the at least one sample individual and a trajectory feature of the corresponding sample motion trajectory.
The method according to claim 1, wherein the acquiring a motion trajectory of the feature point of the target motion in the sequence of frames based on the video comprises:

Extracting, from the video, a sequence of frames in any one of the target action periods, where the target action period refers to a time taken to perform a complete target action;

Obtaining a motion trajectory of the feature point of the target motion in the sequence of frames.
The method according to claim 1, wherein determining the identity of the individual to be identified according to the trajectory feature of the motion trajectory and sample data comprises:

Detecting whether there is a sample motion trajectory matching the motion trajectory according to the trajectory feature of the motion trajectory and the trajectory feature of the sample motion trajectory corresponding to each sample individual;

When it is determined that there is a sample motion trajectory matching the motion trajectory, the identity of the sample individual corresponding to the sample motion trajectory matching the motion trajectory is determined as the identity of the to-be-identified individual.
The method according to claim 1, wherein determining the identity of the individual to be identified according to the trajectory feature of the motion trajectory and sample data comprises:

Using the trajectory feature of the motion trajectory as an input of an identity recognition model, using the identity recognition model to determine the identity of the to-be-identified individual;

The identity recognition model is trained according to the sample data.
The method according to claim 4, wherein the trajectory feature of the motion trajectory is used as an input of an identity recognition model, and before the identity of the to-be-identified individual is determined by using the identity recognition model, the method further includes:

Constructing a training sample set according to the sample data, the training sample set includes a plurality of training samples, each training sample comprising: a trajectory feature extracted from a sample motion trajectory corresponding to one sample individual, and an identity of the sample individual;

The training sample is trained by using a machine learning algorithm to obtain the identity recognition model.
The method according to claim 1, wherein the obtaining a motion trajectory of each feature point of the target action in the sequence of frames comprises:

Identifying each of the feature points from each frame of the frame sequence;

Obtaining a position of each feature point in each frame picture of the sequence of frames;

A motion trajectory of each feature point in the frame sequence is determined according to a position of each feature point in each frame picture of the frame sequence, respectively.
The method according to claim 1, wherein extracting a sequence of frames in any one of the target action periods from the to-be-identified video comprises:

Dividing the to-be-identified video into a plurality of target action cycles;

Extract the sequence of frames in any one of the target action cycles.
A computing device comprising: a processor and a memory; the memory having computer readable instructions stored thereon, the processor being operative to perform the method of any of claims 1-7.
A non-volatile storage medium storing a data processing program, the data processing program including instructions that, when executed by a computing device, cause the computing device to perform any of claims 1-7 Method of instruction.