Posture recognition method, device and equipment and computer readable storage medium

Abstract

The invention discloses a posture recognition method, device and equipment and a computer readable storage medium. Point cloud data of a target object can be acquired; determining the state of the target object according to the point cloud data; if the state is a motion state, determining a transition attitude of the target object based on the point cloud data; and determining the target attitude of the target object according to the recent historical attitude and the transition attitude of the target object. Therefore, according to the scheme, state recognition can be carried out on the target object through the point cloud data, then the transition posture of the target object is determined, and finally the target posture of the target object is determined according to the historical posture and the transition posture of the target object. Therefore, the posture of the user is effectively recognized, and the posture recognition accuracy is improved.

Description

Technical Field

[0001] This application relates to the field of computer technology, and in particular, to a method, device, equipment, and computer-readable storage medium for gesture recognition. Background Art

[0002] With the continuous development of intelligent home, smart home services have been widely applied in people's homes. In order to accurately obtain the position of a target in a home space, gesture recognition can be used to assist in tracking the target position. Related technologies can directly detect the gesture of the target using a millimeter-wave radar.

[0003] However, although related technologies can detect the gesture of the target, millimeter-wave radar detection is affected by factors such as multipath effects and wind, generating interference point cloud information such as noise points, and may not detect point cloud information when the target is stationary, resulting in incorrect gesture recognition results. Summary of the Invention

[0004] Embodiments of this application provide a method, device, equipment, and computer-readable storage medium for gesture recognition, which can accurately recognize the gesture of a target object.

[0005] In a first aspect, embodiments of this application provide a method for gesture recognition, including:

[0006] Obtain point cloud data of a target object;

[0007] Determine the state of the target object according to the point cloud data;

[0008] If the state is a motion state, determine a transition gesture of the target object based on the point cloud data;

[0009] Determine a target gesture of the target object according to the most recent historical gesture of the target object and the transition gesture.

[0010] In a second aspect, embodiments of this application provide a gesture recognition device, including:

[0011] An obtaining unit, configured to obtain point cloud data of a target object;

[0012] A first determination unit, configured to determine the state of the target object according to the point cloud data;

[0013] A recognition unit, configured to, if the state is a motion state, determine a transition gesture of the target object based on the point cloud data;

[0014] A second determination unit, configured to determine a target gesture of the target object according to the most recent historical gesture of the target object and the transition gesture.

[0015] In some embodiments, the first determination unit is specifically configured to:

[0016] Determine the number of static points and the number of dynamic points in the point cloud data;

[0017] Determine the state of the target object according to the number of static points and the number of dynamic points.

[0018] In some embodiments, the first determination unit determines the state of the target object according to the number of static points and the number of dynamic points, and is specifically configured to:

[0019] Determine the dynamic weight of the target object according to the number of static points and the number of dynamic points;

[0020] Determine the state of the target object according to the number of dynamic points and the dynamic weight.

[0021] In some embodiments, the first determination unit determines the state of the target object according to the number of dynamic points and the dynamic weight, and is specifically configured to:

[0022] If the product of the number of dynamic points and the dynamic weight is greater than or equal to a preset threshold, determine that the state is a moving state;

[0023] If the product of the number of dynamic points and the dynamic weight is less than the preset threshold, determine that the state is a stationary state.

[0024] In some embodiments, the recognition unit is specifically configured to:

[0025] Determine the pose feature vector of the target object according to the point cloud data;

[0026] Input the pose feature vector into a preset classifier to obtain a classification result;

[0027] Determine the transitional pose of the target object according to the classification result.

[0028] In some embodiments, the pose recognition device further includes a third determination unit for:

[0029] Determine the motion trajectory associated with the target object;

[0030] Determine the historical frame point cloud data associated with the motion trajectory.

[0031] The recognition unit determines the pose feature vector of the target object according to the point cloud data, and is specifically configured to determine the pose feature vector of the target object according to the current frame point cloud data and the historical frame point cloud data.

[0032] In some embodiments, the recognition unit determines the pose feature vector of the target object according to the current frame point cloud data and the historical frame point cloud data, specifically for:

[0033] Extract a preset number of point clouds with the strongest signal intensity from the current frame point cloud data and the historical frame point cloud data respectively;

[0034] According to the extracted point clouds and a preset step, determine the intra-frame feature vectors of the current frame point cloud data and the historical frame point cloud data respectively;

[0035] According to the extracted point clouds and the preset step, determine the inter-frame feature vector between the current frame point cloud data and the historical frame point cloud data;

[0036] Concatenate the intra-frame feature vector and the inter-frame feature vector to obtain the pose feature vector of the target object.

[0037] In some embodiments, there are multiple preset classifiers, and the recognition unit inputs the pose feature vector into the preset classifier to obtain a classification result, specifically for:

[0038] Determine the corresponding target preset classifier according to the most recent historical pose of the target object;

[0039] Input the pose feature vector into the target preset classifier to obtain a classification result.

[0040] In some embodiments, there are multiple target preset classifiers, and the recognition unit determines the transition pose of the target object according to the classification result, specifically for determining the transition state of the target object according to multiple classification results.

[0041] In some embodiments, the pose recognition device further includes:

[0042] A fourth determination unit, configured to, if the state is a stationary state, determine the most recent historical pose of the target object as the target pose of the target object.

[0043] In some embodiments, the target pose includes a motion pose and a stationary pose, and the pose recognition device further includes a fifth determination unit, configured to:

[0044] If the target pose belongs to the stationary pose, determine the most recent historical position information of the target object as the position information of the target object;

[0045] If the target pose belongs to the motion pose, determine the position information of the target object according to the point cloud data.

[0046] In a third aspect, an embodiment of the present application discloses a computer device, including a processor and a memory. The memory stores a computer program, and the processor is configured to run the computer program in the memory to execute the method disclosed in the first aspect or the third aspect above.

[0047] In a fourth aspect, an embodiment of the present application discloses a computer-readable storage medium. The computer-readable storage medium stores multiple instructions, and the instructions are suitable for being loaded by a processor to execute the method disclosed in the first aspect or the third aspect above.

[0048] The embodiment of the present application can obtain the point cloud data of the target object; determine the state of the target object according to the point cloud data; if the state is a motion state, then determine the transition posture of the target object based on the point cloud data; and determine the target posture of the target object according to the most recent historical posture and the transition posture of the target object. Thus, it can be seen that this solution can perform state recognition on the target object through the point cloud data, and then determine the transition posture of the target object. Finally, according to the most recent historical posture and the transition posture of the target object, the target posture of the target object is determined. The present application determines the target posture of the target object by real-time tracking of the most recent historical posture and the transition posture of the target object, and with the assistance of the two posture information, it can avoid the influence of situations such as the inability to detect point cloud information when stationary or the presence of interfering point cloud information in the environment. In this way, the posture of the user can be effectively recognized, and the accuracy of posture recognition is improved. BRIEF DESCRIPTION OF THE DRAWINGS

[0049] In order to more clearly illustrate the technical solutions in the embodiments of the present application, the following will briefly introduce the drawings required for the description of the embodiments. Obviously, the following drawings are only some embodiments of the present application. For those skilled in the art, without creative efforts, other drawings can be obtained based on these drawings.

[0050] Figure 1 It is a schematic diagram of the scenario of the posture recognition system provided by the embodiment of the present application;

[0051] Figure 2 It is a schematic flowchart of a posture recognition method provided by the embodiment of the present application;

[0052] Figure 3 It is a schematic flowchart of another posture recognition method provided by the embodiment of the present application;

[0053] Figure 4 It is a schematic flowchart of still another posture recognition method provided by the embodiment of the present application;

[0054] Figure 5 It is a schematic flowchart of still another posture recognition method provided by the embodiment of the present application;

[0055] Figure 6 It is a flowchart of a gesture recognition method provided by an embodiment of the present application;

[0056] Figure 7 It is a schematic structural diagram of a gesture recognition device provided by an embodiment of the present application;

[0057] Figure 8 It is a schematic structural diagram of an electronic device provided by an embodiment of the present application. Detailed implementation manners

[0058] Next, the technical solutions in the embodiments of the present application will be clearly and completely described in conjunction with the accompanying drawings in the embodiments of the present application. Obviously, the described embodiments are only a part of the embodiments of the present application, rather than all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative efforts belong to the scope of protection of the present application.

[0059] An embodiment of the present application provides a gesture recognition method, device, equipment and computer-readable storage medium. Specifically, the embodiment of the present application will be described from the perspective of a gesture recognition system. The gesture recognition system may specifically include a data processing end, and the data processing end may be, for example, Figure 1 the server 10 and / or the detection device 20 as shown. Among them, the server 10 may be an independent physical server, or a server cluster or distributed system composed of multiple physical servers, or a cloud server that provides basic cloud computing services such as cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communications, middleware services, domain name services, security services, CDN, and big data and artificial intelligence platforms, etc., but is not limited thereto.

[0060] The gesture recognition method provided by the embodiment of the present application can be applied to various application scenarios, such as, including but not limited to the application scenario of gesture recognition in a home space, and will be specifically described through the following embodiments:

[0061] For example, referring to Figure 1 , it is a schematic diagram of a gesture recognition system provided by an embodiment of the present application. The system may include a data processing end, and information can be transmitted between the data processing ends. The specific scenarios implemented can be as follows:

[0062] The data processing end may be a detection device or a server, or a combination of a detection device and a server. Among them, the detection device 20 may be a device with a function of position detection, for example, specifically may be a human body sensor, a millimeter wave radar, a camera and other devices.

[0063] In one embodiment, the data processing end can be used to: obtain the point cloud data of the target object; determine the state of the target object according to the point cloud data; if the state is a moving state, determine the transition pose of the target object based on the point cloud data; and determine the target pose of the target object according to the most recent historical pose and the transition pose of the target object.

[0064] The following will be described in detail respectively. It should be noted that the order of the following embodiments does not limit the preferred order of the embodiments.

[0065] See Figure 2 , Figure 2 which is a schematic flowchart of a pose recognition method provided by an embodiment of the present application. The embodiment of the present application is applied to Figure 1 the data processing end in, and the specific process is as follows:

[0066] 101. Obtain the point cloud data of the target object.

[0067] The target object can be an object that can move in the target space, such as a human body, an animal, etc.

[0068] The point cloud data can be a set of discrete three-dimensional point coordinates obtained by a lidar or other sensors. Each point contains the position information, speed, signal-to-noise ratio, etc. of the target object in the space coordinate system where it is located, and usually also includes other attributes such as reflection intensity, color, etc. The point cloud data can represent information such as the surface shape, distance, and motion state of the target object.

[0069] In one embodiment, multiple sensors can be used to obtain the point cloud data of the target object. Exemplarily, a millimeter-wave radar can be used to obtain the point cloud data of the target object. A millimeter-wave radar is a sensor that can measure the position and speed of the target object. It can obtain the distance information of the target object by transmitting and receiving microwave signals. By analyzing and processing the echo signal, the point cloud data of the target object can be obtained. For example, in an indoor environment, the millimeter-wave radar can be installed on one side of the room and oriented at a certain angle towards the human body to be measured in the room. The radar emits microwave signals and receives the echo signals emitted by the target human body. By analyzing the received echo signals, the position information and distance information of the target human body can be obtained. Further processing the echo signals can obtain the point cloud data of the target human body, that is, a set of distance and position information of the points around the human body. By obtaining the point cloud data of the target object, the state and pose of the target object can be further recognized. It is worth noting that this method can be applied in various environments and has certain real-time performance and accuracy.

[0070] By using a millimeter-wave radar to obtain the point cloud data of a target object, it is possible to assist in identifying the state and posture of the target object, providing more accurate and reliable information for applications such as target tracking and positioning.

[0071] 102. Determine the state of the target object according to the point cloud data.

[0072] The state refers to the posture or action state of the target object, such as a stationary state, a moving state, etc.

[0073] In one implementation, after using the millimeter-wave radar to obtain the point cloud data of the target object, according to the information in the point cloud data, it can be analyzed whether the state of the target object belongs to a moving state or a stationary state. In some embodiments, a dynamic and static classifier can be used to classify the point cloud data to determine the state of the target object. According to the characteristics of the point cloud data, a dynamic and static classifier can be pre-trained to identify whether the state of the target object belongs to a stationary state or a moving state. In one implementation, a large number of point cloud information of the target object in a stationary state and point cloud information of the target object in a moving state can be collected in advance for classification training to obtain a dynamic and static classifier. In a specific application scenario, when it is necessary to judge the state of the target object, the currently obtained point cloud data can be input into the classifier to identify whether the target object is in a stationary state or a moving state.

[0074] In some embodiments, the state of the target object can be determined according to the number of static points and dynamic points in the point cloud data. For example, step 102 may include:

[0075] (102.1) Determine the number of static points and the number of dynamic points in the point cloud data;

[0076] (102.2) Determine the state of the target object according to the number of static points and the number of dynamic points.

[0077] A point cloud data frame will include one or more points. Static points refer to points with a relatively small movement speed in the point cloud data frame, while dynamic points are points with a relatively large movement speed in the point cloud data frame.

[0078] In one implementation, the static points and dynamic points in the point cloud data can be separated and counted. The state of the target object can be determined by judging the number of static points and dynamic points. By counting the number of static points and dynamic points and comparing them, the state of the target object can be inferred. For example, when the number of static points occupies the vast majority, it can be considered that the target object is in a stationary state; when the number of dynamic points is relatively large, it can be considered that the target object is in a moving state.

[0079] In some embodiments, the state of the target object can be determined based on the number of static points and the dynamic weights between the dynamic points. For example, step (102.2) may include:

[0080] (102.2.1) Determine the dynamic weight of the target object according to the number of static points and the number of dynamic points;

[0081] (102.2.2) Determine the state of the target object according to the number of dynamic points and the dynamic weight.

[0082] Further, step (102.2.2) may include:

[0083] (A.1) If the product of the number of dynamic points and the dynamic weight is greater than or equal to a preset threshold, determine that the state is a moving state;

[0084] (A.2) If the product of the number of dynamic points and the dynamic weight is less than the preset threshold, determine that the state is a stationary state.

[0085] The dynamic weight can be the number weight of the dynamic points in all point clouds. In a point cloud data frame, the greater the dynamic weight, the more dynamic points there are, and the more the state of the target object tends to be a moving state.

[0086] The preset threshold can be a value set in advance according to specific requirements and application scenarios, and is used to determine whether a certain condition is met or a certain state is established. The preset threshold can be used here to determine the state of the target object. Specifically, if the product of the number of dynamic points and the dynamic weight is greater than or equal to the preset threshold, determine that the state of the target object is a moving state; conversely, if the product of the number of dynamic points and the dynamic weight is less than the preset threshold, determine that the state of the target object is a stationary state.

[0087] Determining the dynamic weight of a target object according to the number of static points and the number of dynamic points means calculating the dynamic weight value of the target object by counting the number of static points and dynamic points in the point cloud data. In one implementation, a model can be trained using machine learning methods for classifying static points and dynamic points. First, the static points and dynamic points of multiple frames of point cloud can be counted to obtain the number of static points and the number of dynamic points in each frame. Then, the dynamic points of each frame are concatenated together in chronological order as positive samples; the static points of each frame are concatenated together in chronological order as negative samples. Next, a model is trained using the Support Vector Machine (SVM) algorithm, and this model can classify points as static points or dynamic points according to the characteristics of the points. When training the model, the number of dynamic points in each frame can be used as a weight to adjust the importance of each frame of dynamic points in classification. If the number of dynamic points in a certain frame is large, it indicates that this frame is likely to be the point cloud data collected when the target is in a moving state, so its dynamic weight will be adjusted to a larger value; on the contrary, if the number of dynamic points in a certain frame is small, it indicates that this frame may be the point cloud data collected when the target is in a stationary state, so its dynamic weight will be adjusted to a smaller value. When training the model, the weight can also be determined according to the difference between the dynamic points and static points in each frame. If the difference between the dynamic points and static points in a certain frame is large, the dynamic weight of this frame will be adjusted to a larger value; on the contrary, if the difference between the dynamic points and static points in a certain frame is small, the dynamic weight of this frame will be adjusted to a smaller value.

[0088] Exemplarily, continuous N frames of point cloud data of a trajectory a can be taken, and then the number of static points and dynamic points in the N frames of point cloud data of a is counted. The number of static points and the number of dynamic points of the N frames of point cloud are concatenated and used to train a model and classify it using the SVM algorithm. For example, the dynamic points of N frames are concatenated as positive samples (for example, there are 5 frames, and after concatenating the dynamic points, it is assumed to be (3, 5, 4, 7, 10)), and the static points of N frames are concatenated as negative samples (for example, there are 5 frames, and after concatenating the static points, it is assumed to be (2, 2, 4, 5, 1)). By comparing the differences in the number of static and dynamic points in each frame between the positive and negative samples, the greater the difference, the greater the weight of the positive sample corresponding to the frame. For example, in the fifth frame, there are 10 dynamic points and only 1 static point, and the difference between them is large, so the weight of the fifth frame is also relatively large; for example, in the first frame, there are 3 dynamic points and 2 static points, and the difference between them is small, so the weight of the fifth frame is also small.

[0089] Based on the number of dynamic points and the dynamic weight, the state of the target object can be further determined. If the product of the number of dynamic points and the dynamic weight is less than the preset threshold, it indicates that the target object has hardly moved, and its state can be determined to be the stationary state. On the contrary, if the product of the number of dynamic points and the dynamic weight is greater than or equal to the preset threshold, it indicates that the target object has moved significantly, and its state can be determined to be the moving state.

[0090] Exemplarily, it is assumed that the posture of a person needs to be recognized and tracked. The point cloud data collected by devices such as radar can reflect the motion state and posture changes of the person. First, the number of static points and dynamic points in the point cloud data can be counted. Then, based on these statistical results, the dynamic weight of each frame can be calculated. Suppose the number of dynamic points in a certain frame is 10. Then, according to the calculation formula of the weight, the dynamic weight of this frame may be relatively large. Next, based on the product of the number of dynamic points and the dynamic weight, the state of the target object can be determined. Suppose the threshold is 20. If the product of the number of dynamic points and the dynamic weight is greater than 20, then it can be determined that the target object is in the moving state. In this way, by calculating the dynamic weight based on the number of static points and dynamic points, and determining the state of the target object based on the product of the number of dynamic points and the dynamic weight, the motion state of the target object within a certain time period can be accurately judged and recognized. And this method can be applied to various scenarios of target tracking and posture recognition, such as human posture recognition, object motion detection, etc.

[0091] By recognizing the state of the target object, the accuracy of target tracking can be improved, and the target can be prevented from being lost or taken away by other interferences. In addition, by recording the posture of the target object, the habits and behavior patterns of the target object can also be analyzed, such as the frequently sitting position, how long it has been sitting, and other information. Generally speaking, determining the state of the target object based on the point cloud data is an effective method and can be widely applied in the fields of target tracking and behavior analysis.

[0092] 103. If the state is the moving state, then based on the point cloud data, determine the transitional posture of the target object.

[0093] The transitional posture can refer to the intermediate state when the target object changes from one behavioral posture to another. The transitional postures can include the following postures: standing to sitting, sitting to standing, sitting to lying, lying to sitting, moving while standing.

[0094] If the state of the target object is a moving state, the transitional posture of the target object is further determined based on the point cloud data. The three-dimensional point cloud information in the point cloud data can be used to determine the transitional posture of the target object. The three-dimensional point cloud information can provide the position and shape information of the target object, thereby helping to identify the posture of the target object. In a specific implementation, the following method can be used to judge the target posture and determine the transitional posture. First, by processing and analyzing the point cloud data, determine whether the target object is in a stationary state or a moving state. Then, if the target object is in a moving state, further determine the transitional posture of the target object according to the point cloud data. For example, in a target tracking scenario, assume the target object is a person. The state of the person can be judged by analyzing the point cloud data. If the point cloud data shows that the target object is moving, then the transitional posture of the target object can be inferred based on this data, such as transitioning from a standing posture to a sitting posture.

[0095] Specifically, by analyzing the point cloud data and matching it with a preset classification model, it can be determined which transitional posture the target object is in.

[0096] In some embodiments, the posture feature vector of the target object can be determined according to the point cloud data, so as to confirm the transitional posture of the target object. For example, step 103 may include:

[0097] (103.1) Determine the posture feature vector of the target object according to the point cloud data;

[0098] (103.2) Input the posture feature vector into a preset classifier to obtain a classification result;

[0099] (103.3) Determine the transitional posture of the target object according to the classification result.

[0100] The posture feature vector can be a vector composed of the point cloud data information of the target object, or a vector calculated according to the point cloud data information of the target object, used to describe the posture attributes of the target object. The definition of the posture feature vector can vary according to specific posture recognition methods and application scenarios.

[0101] The preset classifier can refer to a set of pre-trained classifier models. These classifier models are designed to classify the transitional postures of the target object according to the input posture feature vector. The training and tuning of the preset classifier are carried out during the development of the posture recognition system to ensure that it can accurately classify the target object. There can be one or multiple preset classifiers.

[0102] In one embodiment, based on the point cloud data, the pose feature vector of the target object can be determined. The point cloud data is the coordinate information of a series of points on the target surface obtained by a 3D lidar or other depth sensors. By analyzing the relative positions and spatial distributions of these points, the pose feature vector of the target object can be extracted, including information such as the position, orientation, and rotation angle of the target object. Next, the pose feature vector is input into a preset classifier, and the pose of the target object is classified through a machine learning algorithm. The preset classifier has been trained according to the training data set, can compare the input pose feature vector with the known classification results, and give the classification result of which pose the target object belongs to. Finally, the transition pose of the target object is determined according to the classification result. According to the classification result, the current transition pose of the target object can be judged.

[0103] Exemplarily, by analyzing the point cloud data, the joint point coordinate information of the human body can be extracted, and the pose feature vector of the human body can be calculated, such as the orientation of the body, the rotation angle of the head, etc. Inputting these pose feature vectors into a preset classifier can obtain classification results, such as "sitting to standing" or "sitting to lying down". Thus, based on the point cloud data, the pose feature vector of the target object can be determined, the classification result can be obtained through the preset classifier, and the transition pose of the target object can be determined according to the classification result.

[0104] In some embodiments, there are multiple preset classifiers, and the target preset classifier can be determined according to the historical pose of the target object, so as to obtain the classification result. For example, step (103.2) includes:

[0105] (103.2.1) Determine the corresponding target preset classifier according to the most recent historical pose of the target object;

[0106] (103.2.2) Input the pose feature vector into the target preset classifier to obtain the classification result.

[0107] The most recent historical pose may refer to the pose of the target object at the previous moment.

[0108] In one embodiment, determining the corresponding target preset classifier according to the most recent historical posture of the target object means selecting the corresponding classifier according to different historical postures of the target object. For example, if the most recent historical posture of the target object is a sitting state, then select the classifier related to sitting. If the most recent historical posture of the target object is a standing state, then select the classifier related to standing. By selecting the corresponding target preset classifier, the classification accuracy can be improved. After inputting the posture feature vector into the target preset classifier, the obtained classification result can be used to determine the current posture of the target. For example, according to the input posture feature vector, if the classification result of the transition posture is from standing to sitting and the most recent historical posture of the target object is standing, it indicates that the target has changed from the standing state to the sitting state, and the target posture of the target object can be determined as sitting; if the classification result of the transition posture is from sitting to standing and the most recent historical posture of the target object is sitting, it indicates that the target has changed from the sitting state to the standing state, and the target posture of the target object can be determined as standing. By determining the transition posture of the target object, it can assist in application fields such as target tracking and behavior analysis.

[0109] Exemplarily, in the case where the most recent historical posture is standing, if the target object is in a moving state, extract the posture feature vector and input it into the classifier of "standing VS standing to sitting" to obtain a classification result and directly obtain the transition posture; in the case where the historical posture is lying, if the target object is in a moving state, extract the posture feature vector and input it into the classifier of "lying VS lying to sitting" to obtain a classification result and directly obtain the transition posture.

[0110] In some embodiments, if there are multiple target preset classifiers determined according to the most recent historical posture, then step (103.2.2) includes: determining the transition state of the target object according to multiple classification results.

[0111] Exemplarily, in the case where the most recent historical posture is sitting: if the target is in a moving state, extract the posture feature vector and input it into the three classifiers of "sitting VS sitting to standing, sitting VS sitting to lying, sitting to standing VS sitting to lying" respectively. The results of the three classifiers determine the final transition posture by voting. Whichever transition posture has more votes is determined as that posture. If the number of votes is the same, the most recent historical posture remains unchanged. In this way, the transition posture of the target object can be determined.

[0112] To more conveniently understand step 103, the present application provides an example to illustrate how to obtain a classifier (binary classifier) that can distinguish the transition posture of the target object:

[0113] Suppose there are 5 categories of the transition postures of the target object: moving while standing, standing to sitting, sitting to standing, sitting to lying, and lying to sitting. When the state of the target object is recognized as the motion state in step 102, the feature vectors can be constructed from the point cloud data and classified. In one implementation: Take N consecutive frames of point cloud data n1, n2,..., n N , from the point cloud data n i (1 ≤ i ≤ N (N is the number of frames)), extract 10 point clouds in the order of decreasing signal strength (if there are not 10 point clouds in n i , then copy them to 10 in the order of decreasing signal strength); According to the step size s, subtract the 10 point cloud data of the point cloud n j (1 ≤ j ≤ N - s) pairwise to obtain the feature vector Subtract the 10 point cloud data of the point cloud n j+s pairwise to obtain the feature vector According to the step size s, subtract the 10 point cloud data of the point cloud n j (1 ≤ j ≤ N - s) pairwise from the 10 point cloud data of the point cloud n j+s to obtain the feature vector Concatenate the feature vectors , and train and classify the model with an iterative algorithm (adaptive boos, adaboost algorithm). In this way, 5 binary classifiers are trained: standing VS standing to sitting, lying VS lying to sitting, sitting VS sitting to lying, sitting VS sitting to standing, sitting to standing VS sitting to lying.

[0114] In one implementation, the present application also provides an adaboost algorithm, and the specific algorithm is as follows:

[0115] Input:

[0116] T = {(x1, y1), (x2, y2),..., (x N , y N )}, where y i ∈ Y = -1, 1: training data set

[0117] M: number of iterations

[0118] Output:

[0119] Category

[0120] Method:

[0121] (1) Initialize the weight distribution of the training samples:

[0122] D1 = (w 1,1 , w 1,2 ,..., w 1,N ),

[0123] (2) For m = 1, 2, ..., M

[0124] Use the training data set with the weight distribution D m to learn and obtain the weak classifier:

[0125] G m (x);

[0126] Calculate the classification error rate of G m (x) on the training data set:

[0127]

[0128] Calculate the weight of G m (x) in the strong classifier:

[0129]

[0130] Update the weight distribution of the training data set:

[0131]

[0132]

[0133] Obtain the final classifier:

[0134]

[0135] It should be understood that the provided adaboost algorithm above is only an exemplary illustration of the iterative algorithm, and the specific iterative algorithm may vary according to specific circumstances and needs to be specifically designed and implemented according to actual requirements.

[0136] Through the point cloud data, the transitional pose of the target object can be well determined. Such pose information helps to more accurately track and identify the target object, improving the accuracy and stability of target tracking. At the same time, through the analysis of the target pose, the living habits and behavior patterns of the target object can also be understood and recorded, providing more information for subsequent applications.

[0137] 104. Determine the target pose of the target object according to the recent historical pose and the transitional pose of the target object.

[0138] The recent historical pose may refer to the historical pose with the smallest time span from the current frame among all the historical poses of the target object.

[0139] The target pose may refer to the pose finally recognized by the target object.

[0140] In one embodiment, after obtaining the data of the historical posture and transitional posture of the target object, these data can be processed and analyzed. By performing time series analysis on the historical posture, the trend and pattern of the posture change of the target object in different time periods can be observed. At the same time, by analyzing the transitional posture, the transitional manner and process of the target object between different postures can be understood. Based on the analysis results of the historical posture and transitional posture, calculations and predictions can be made using mathematical models or algorithms, and finally the target posture of the target object can be obtained.

[0141] In one embodiment, the present application also provides a relationship table between the target posture, the nearest historical posture, and the transitional posture. The specific table is as follows:

[0142]

[0143]

[0144] It can be seen that when both the nearest historical posture and the transitional posture are standing, the target posture of the target object is standing; when the nearest historical posture is standing and the transitional posture is from standing to sitting, the target posture of the target object is sitting; when the nearest historical posture is sitting and the transitional posture is from sitting to standing, the target posture of the target object is standing; when the nearest historical posture is sitting and the transitional posture is from sitting to lying, the target posture of the target object is lying; when the nearest historical posture is lying and the transitional posture is from lying to sitting, the target posture of the target object is sitting.

[0145] It should be understood that the provided relationship table between the target posture, the nearest historical posture, and the transitional posture is only an exemplary illustration of the target posture recognition. The specific recognition method may vary according to specific circumstances and needs to be specifically designed and implemented according to actual requirements.

[0146] Through the above method, by analyzing the nearest historical posture and the transitional posture of the target object, determining the target posture can provide more accurate results. The accuracy and real-time performance of the posture recognition of the target object are improved.

[0147] By implementing any one embodiment or combination of embodiments in the embodiments of the present application, the application scenario of the posture recognition process can be realized.

[0148] As described above, the embodiments of the present application can obtain the point cloud data of the target object; determine the state of the target object according to the point cloud data; if the state is a motion state, determine the transition posture of the target object based on the point cloud data; and determine the target posture of the target object according to the nearest historical posture and the transition posture of the target object. It can be seen that this solution can identify the state of the target object through the point cloud data, and then determine the transition posture of the target object. Finally, according to the nearest historical posture and the transition posture of the target object, the target posture of the target object is determined. By real-time tracking the nearest historical posture and the transition posture of the target object, and assisted by the two posture information, the target posture of the target object is judged, which can avoid the influence of situations such as the inability to detect point cloud information when stationary or the appearance of interfering point cloud information in the environment. In this way, the posture of the user can be effectively identified, and the accuracy of posture recognition is improved.

[0149] Please refer to Figure 3 , Figure 3 which is a schematic flowchart of another posture recognition method provided by the embodiments of the present application. The specific process is as follows:

[0150] 201. Obtain the point cloud data of the target object.

[0151] Among them, the specific content of step 201 can refer to the content of step 101.

[0152] 202. Determine the state of the target object according to the point cloud data.

[0153] Among them, the specific content of step 202 can refer to the content of step 102.

[0154] 203. If the state is a motion state, determine the motion trajectory associated with the target object.

[0155] 204. Determine the historical frame point cloud data associated with the motion trajectory.

[0156] 205. Determine the posture feature vector of the target object according to the current frame point cloud data and the historical frame point cloud data.

[0157] Combined with steps 203-205, a detailed description is as follows:

[0158] The historical frame point cloud information can refer to the point cloud data obtained in the past period of time, including the point cloud data of multiple historical frames. In posture recognition, the historical frame point cloud information is used to construct the posture feature vector of the target object to help determine the current posture of the target.

[0159] The current frame point cloud information can refer to the point cloud data obtained from the sensor at the current time point. The current frame point cloud information contains the three-dimensional spatial distribution information of the object or scene at the current time point and can be used for tasks such as target recognition and posture prediction.

[0160] In one embodiment, determining the motion trajectory associated with a target object refers to tracking and recording the motion state of the target object over a period of time through certain methods and algorithms, and representing it as a continuous trajectory line, so as to better understand the movement pattern and behavioral characteristics of the target object. When determining the motion trajectory associated with a target object, a technique called object detection and tracking is usually utilized. First, an image processing algorithm is used to detect the position of the target object in each frame of the image, including information such as the bounding box and center point of the target. Then, through a tracking algorithm, the positions of the target in different frames are associated to establish the motion trajectory of the target. Commonly used tracking algorithms include the Kalman filter, particle filter, etc. It is worth noting that each point on the motion trajectory is obtained based on the historical point cloud data of each frame. The points in the point cloud data of each frame are aggregated, calculated, etc. to obtain a position coordinate, and the motion trajectory is obtained by stringing together the position coordinates corresponding to the historical point cloud data of each frame. The point cloud data is composed of a large number of discrete three-dimensional points, which records the geometric shape and spatial position information of the target object at different time points. By determining the historical frame point cloud data associated with the motion trajectory, the motion characteristics and environmental interaction of the target object can be further analyzed. For example, motion parameters such as the motion speed, acceleration, and curvature of the trajectory of the target object can be calculated through the point cloud data to study the motion pattern and behavioral characteristics of the target object. At the same time, the point cloud data can also be used to model and analyze the surrounding environment of the target object, such as identifying obstacles and locating the ground plane.

[0161] After determining the historical frame point cloud data associated with the motion trajectory of the target object, the point clouds in the current frame point cloud information and the historical frame point cloud information can be extracted and processed to obtain relevant information about the target object. For example, a deep learning algorithm can be used to process the point cloud data to extract features such as the coordinates and speed of the point cloud. Then, based on these extracted features, a pose feature vector of the target object is constructed.

[0162] In some embodiments, the pose feature vector of the target object can be determined according to the intra-frame feature vector and inter-frame feature vector between the current frame point cloud information and the historical frame point cloud information. For example, step 205 may include:

[0163] (205.1) Extract a preset number of point clouds with the strongest signal intensity from the current frame point cloud data and the historical frame point cloud data respectively;

[0164] (205.2) Determine the intra-frame feature vectors of the current frame point cloud data and the historical frame point cloud data respectively according to the extracted point clouds and the preset step size;

[0165] (205.3) Determine the inter-frame feature vector between the current frame point cloud data and the historical frame point cloud data according to the extracted point cloud and the preset step size;

[0166] (205.4) Concatenate the intra-frame feature vector and the inter-frame feature vector to obtain the pose feature vector of the target object.

[0167] The intra-frame feature vector can refer to a vector obtained by processing and calculating the point cloud data in the same frame of point cloud information, which describes the internal features of the point cloud. The intra-frame feature vector can be used to represent the pose state of the target object in the current frame, including information such as the number of point clouds, coordinates, and speed. By comparing the changes in different intra-frame feature vectors, the pose of the target object can be identified and tracked.

[0168] The inter-frame feature vector refers to a feature vector obtained by calculating the point cloud data between different frames. In pose recognition, the frame point cloud can be compared with the historical frame point cloud, and the difference between them can be calculated to obtain the inter-frame feature vector. The inter-frame feature vector can reflect the change of the pose of the target object and is used to judge the state and position of the target object.

[0169] The preset step size can refer to the value of an increment or decrement variable, and can execute a loop or iteration process according to a certain rule. The preset step size can be used to obtain the inter-frame feature vector and the intra-frame feature vector.

[0170] In one implementation, the current frame point cloud data can be clustered first to obtain the clustered point cloud blocks. Then, the features of each point cloud can be extracted from each point cloud block, such as metrics like distance, shape, and intensity. According to the features of the point cloud, the extracted features can be combined into the intra-frame feature vector of the current frame point cloud information. For example, the point clouds in the point cloud block can be sorted according to the signal intensity from strong to weak, and the preset number of point clouds can be subtracted pairwise according to the preset step size to obtain the intra-frame feature vector. Similarly, the point clouds in the historical frame point cloud information can be processed in the same way to obtain the intra-frame feature vector of the historical frame point cloud information.

[0171] Finally, the intra-frame feature vector and the inter-frame feature vector are concatenated to obtain the pose feature vector of the target object. In one implementation, the intra-frame feature vector and the inter-frame feature vector can be arranged in order. Then, the arranged feature vectors can be concatenated together to form the pose feature vector of the target object.

[0172] Exemplarily, extract 10 point clouds from n i (1 ≤ i ≤ N) according to the signal intensity from strong to weak (if there are not 10 point clouds in n i then copy them to 10 according to the signal intensity from strong to weak), n i is any current point cloud data or historical point cloud data;

[0173] According to the preset step s, point cloud n j The feature vectors are obtained by subtracting pairwise 10 pieces of point cloud data where (1≤j≤N - s). Point cloud n j+s The feature vectors are obtained by subtracting pairwise 10 pieces of point cloud data. n j n is any current point cloud data or historical point cloud data;

[0174] According to the preset step s, point cloud n j The feature vectors are obtained by subtracting pairwise 10 pieces of point cloud data where (1≤j≤N - s) from 10 pieces of point cloud data of point cloud n j+s and 10 pieces of point cloud data of point cloud n.

[0175] The feature vectors are concatenated to obtain the pose feature vector of the target object.

[0176] Through the above method, the motion trajectory associated with the target object and the historical frame point cloud data associated with the motion trajectory can be determined, thereby obtaining the pose feature vector of the target object and realizing the pose recognition of the target object subsequently.

[0177] 206. Input the pose feature vector into a preset classifier to obtain a classification result.

[0178] 207. Determine the transitional pose of the target object according to the classification result.

[0179] Among them, the specific content of steps 206 - 207 can refer to the content of step 103.

[0180] 208. Determine the target pose of the target object according to the nearest historical pose and the transitional pose of the target object.

[0181] Among them, the specific content of step 208 can refer to the content of step 104.

[0182] Please refer to Figure 4 , Figure 4 which is a schematic flowchart of another pose recognition method provided by an embodiment of the present application, and the specific process is as follows:

[0183] 301. Obtain the point cloud data of the target object.

[0184] Among them, the specific content of step 301 can refer to the content of step 101.

[0185] 302. Determine the state of the target object according to the point cloud data.

[0186] Among them, the specific content of step 302 can refer to the content of step 102.

[0187] 303. If the state is the stationary state, determine the most recent historical pose of the target object as the target pose of the target object.

[0188] When the target object is in the stationary state, its target pose can be determined through its past pose history. For a target object in the stationary state, its historical pose is often relatively stable, which is easy to analyze and predict, improving the robustness and stability of the system.

[0189] In summary, when the target object is in the stationary state, by analyzing the historical pose of the target object, the target pose of the target object can be determined, thereby improving the accuracy and efficiency of the recognition and tracking of the object pose.

[0190] Please refer to Figure 5 , Figure 5 which is a schematic flowchart of another pose recognition method provided by an embodiment of the present application. The specific process is as follows:

[0191] 401. Obtain the point cloud data of the target object.

[0192] Among them, the specific content of step 401 can refer to the content of step 101.

[0193] 402. Determine the state of the target object according to the point cloud data.

[0194] Among them, the specific content of step 402 can refer to the content of step 102.

[0195] 403. If the state is the motion state, determine the transitional pose of the target object based on the point cloud data.

[0196] Among them, the specific content of step 403 can refer to the content of step 103.

[0197] 404. Determine the target pose of the target object according to the most recent historical pose and the transitional pose of the target object.

[0198] Among them, the specific content of step 404 can refer to the content of step 104.

[0199] 405. Determine the position information of the target object according to the target pose.

[0200] In one embodiment, determining the position information of the target object according to the target pose means determining the position of the target through the pose state of the target. For example, when the target is in a sitting pose, the most recent historical position information of the target can be used as the position information of the target. When the target is in a walking pose, point cloud data can be used to determine the position information of the target. Exemplarily, suppose there is a person in a room. According to the pose of the target, it can be determined whether the person is sitting or standing. If it is detected that the person's pose is a sitting pose and there is no action of getting up, then the position information of the person can be kept in place. If it is detected that the person's pose has changed, such as from a sitting pose to a standing pose, then the position information of the person can be updated to the new position.

[0201] In some embodiments, the target pose can be specifically divided into a motion pose and a static pose, and step 405 can include:

[0202] (405.1) If the target pose belongs to the static pose, then determine the most recent historical position information of the target object as the position information of the target object;

[0203] (405.2) If the target pose belongs to the motion pose, then determine the position information of the target object according to the point cloud data.

[0204] In one embodiment, when the pose of the target object belongs to the static pose, its position information will be determined as the most recent historical position information. This means that even without point cloud data or interference nearby, the position of the target object remains at the previous position. For example, if a person is sitting and watching TV, then his position information will be determined as the position when he was sitting and watching TV before. When the pose of the target object belongs to the motion pose, its position information will be determined according to the point cloud data. For example, if a person stands up from a sitting pose and walks a few steps, then by analyzing the point cloud data, it can be determined that his current position information is at the walking position. In this way, according to the target pose of the target object, the position information of the target object can be determined more accurately, improving the accuracy of target tracking and positioning.

[0205] In this way, by determining the position information of the target object according to the target pose, the position of the target can be accurately tracked and corresponding adjustments can be made according to the pose state of the target.

[0206] For the convenience of understanding, the present application also provides an application scenario and an application flowchart of the above-mentioned pose recognition method in this application scenario, as Figure 6 shown:

[0207] Collect 10 frames of point cloud data frames in the target space, and then use the 10 frames of point cloud data frames to judge the state of the target object. If the target object is in a stationary state, its posture is the same as the previous moment and remains unchanged. If the target object is in a moving state, detect whether the target object is "standing to sitting", "sitting to standing", "sitting to lying", or "lying to sitting", and then judge whether the target posture of the target object belongs to "standing", "sitting", or "lying". In this way, the posture of the target object can be accurately recognized.

[0208] To better implement the above method, an embodiment of the present application further provides a posture recognition device. For example, as Figure 7 shown, the posture recognition device may include an acquisition unit 501, a first determination unit 502, a recognition unit 503, and a second determination unit 504.

[0209] The acquisition unit 501 is configured to acquire point cloud data of the target object;

[0210] The first determination unit 502 is configured to determine the state of the target object according to the point cloud data;

[0211] The recognition unit 503 is configured to, if the state is a moving state, determine the transition posture of the target object based on the point cloud data;

[0212] The second determination unit 504 is configured to determine the target posture of the target object according to the nearest historical posture and the transition posture of the target object.

[0213] In some embodiments, the first determination unit 502 is specifically configured to:

[0214] Determine the number of static points and the number of dynamic points in the point cloud data;

[0215] Determine the state of the target object according to the number of static points and the number of dynamic points.

[0216] In some embodiments, the first determination unit 502 determines the state of the target object according to the number of static points and the number of dynamic points, and is specifically configured to:

[0217] Determine the dynamic weight of the target object according to the number of static points and the number of dynamic points;

[0218] Determine the state of the target object according to the number of dynamic points and the dynamic weight.

[0219] In some embodiments, the first determination unit 502 determines the state of the target object according to the number of dynamic points and the dynamic weight, and is specifically configured to:

[0220] If the product of the number of dynamic points and the dynamic weight is greater than or equal to a preset threshold, determine that the state is a moving state;

[0221] If the product of the number of dynamic points and the dynamic weight is less than a preset threshold, determine the state as the stationary state.

[0222] In some embodiments, the recognition unit 503 is specifically configured to:

[0223] Determine the pose feature vector of the target object according to the point cloud data;

[0224] Input the pose feature vector into a preset classifier to obtain a classification result;

[0225] Determine the transition pose of the target object according to the classification result.

[0226] In some embodiments, the pose recognition device further includes a third determination unit 505, configured to:

[0227] Determine the motion trajectory associated with the target object;

[0228] Determine the historical frame point cloud data associated with the motion trajectory.

[0229] The recognition unit 503 determines the pose feature vector of the target object according to the point cloud data, specifically by determining the pose feature vector of the target object according to the current frame point cloud data and the historical frame point cloud data.

[0230] In some embodiments, the recognition unit 503 determines the pose feature vector of the target object according to the current frame point cloud data and the historical frame point cloud data, specifically by:

[0231] Extract a preset number of point clouds with the strongest signal intensity from the current frame point cloud data and the historical frame point cloud data respectively;

[0232] Determine the intra-frame feature vectors of the current frame point cloud data and the historical frame point cloud data respectively according to the extracted point clouds and a preset step;

[0233] Determine the inter-frame feature vector between the current frame point cloud data and the historical frame point cloud data according to the extracted point clouds and a preset step;

[0234] Concatenate the intra-frame feature vector and the inter-frame feature vector to obtain the pose feature vector of the target object.

[0235] In some embodiments, there are multiple preset classifiers. The recognition unit 503 inputs the pose feature vector into the preset classifier to obtain a classification result, specifically by:

[0236] Determine the corresponding target preset classifier according to the historical pose of the target object;

[0237] Input the pose feature vector into the target preset classifier to obtain a classification result.

[0238] In some embodiments, there are multiple target preset classifiers, and the recognition unit 503 determines the transition posture of the target object according to the classification results, specifically for determining the transition state of the target object according to multiple classification results.

[0239] In some embodiments, the posture recognition device further includes:

[0240] A fourth determination unit 506, configured to, if the state is a stationary state, determine the most recent historical posture of the target object as the target posture of the target object.

[0241] In some embodiments, the target posture includes a motion posture and a stationary posture, and the posture recognition device further includes a fifth determination unit 507, configured to:

[0242] If the target posture belongs to the stationary posture, determine the most recent historical position information of the target object as the position information of the target object;

[0243] If the target posture belongs to the motion posture, determine the position information of the target object according to the point cloud data.

[0244] As can be seen from the above, the embodiments of the present application can perform state recognition on the target object through the point cloud data, and then determine the transition posture of the target object. Finally, according to the historical posture and the transition posture of the target object, the target posture of the target object is determined. In this way, the user's posture can be effectively recognized, and the accuracy of posture recognition is improved.

[0245] The embodiments of the present application further provide a computer device, as Figure 8 shown, which shows a schematic structural diagram of the computer device involved in the embodiments of the present application. Specifically:

[0246] The computer device may include a processor 601 with one or more processing cores, a memory 602 with one or more computer-readable storage media, a power supply 603, an input unit 604, and other components. Those skilled in the art can understand that Figure 8 the structural diagram of the computer device shown in

[0247] The processor 601 is the control center of the computer device, connecting various parts of the entire computer device through various interfaces and circuits. By running or executing software programs and / or modules stored in the memory 602, and by invoking the data stored in the memory 602, it executes various functions of the computer device and processes data. Optionally, the processor 601 may include one or more processing cores; preferably, the processor 601 may integrate an application processor and a modem processor. Among them, the application processor mainly processes the operating system, user interface, application programs, etc., and the modem processor mainly processes wireless communications. It can be understood that the above-mentioned modem processor may not be integrated into the processor 601 either.

[0248] The memory 602 can be used to store software programs and modules. The processor 601 executes various functional applications and gesture recognition by running the software programs and modules stored in the memory 602. The memory 602 may mainly include a program storage area and a data storage area. Among them, the program storage area can store the operating system, application programs required for at least one function (such as the sound playback function, image playback function, etc.); the data storage area can store data created according to the use of the computer device. In addition, the memory 602 may include high-speed random access memory, and may also include non-volatile memory, such as at least one magnetic disk storage device, flash memory device, or other non-volatile solid-state storage devices. Correspondingly, the memory 602 may also include a memory controller to provide the processor 601 with access to the memory 602.

[0249] The computer device also includes a power supply 603 that powers each component. Preferably, the power supply 603 can be logically connected to the processor 601 through a power management system, so as to implement functions such as management of charging, discharging, and power consumption management through the power management system. The power supply 603 may also include any components such as one or more DC or AC power supplies, a recharge system, a power failure detection circuit, a power converter or inverter, and a power status indicator.

[0250] The computer device may also include an input unit 604, which can be used to receive input digital or character information, and generate keyboard, mouse, joystick, optical or trackball signal inputs related to user settings and function controls.

[0251] Although not shown, the computer device may also include a display unit, etc., which will not be elaborated here. Specifically, in the embodiments of this application, the processor 601 in the computer device will load the executable files corresponding to the processes of one or more application programs into the memory 602 according to the following instructions, and the processor 601 will run the application programs stored in the memory 602 to achieve various functions as follows:

[0252] Obtain the point cloud data of the target object; determine the state of the target object according to the point cloud data; if the state is a moving state, then determine the transition posture of the target object based on the point cloud data; determine the target posture of the target object according to the most recent historical posture and the transition posture of the target object.

[0253] For the specific implementation of each of the above operations, reference may be made to the previous embodiments and will not be elaborated herein.

[0254] Thus, it can be seen that this solution can identify the state of the target object through the point cloud data, and then determine the transition posture of the target object. Finally, according to the historical posture and the transition posture of the target object, the target posture of the target object is determined. In this way, the posture of the user can be effectively identified, and the accuracy of posture recognition is improved.

[0255] Those of ordinary skill in the art can understand that all or part of the steps in the various methods of the above embodiments can be completed by instructions, or by controlling relevant hardware through instructions. The instructions can be stored in a computer-readable storage medium and loaded and executed by a processor.

[0256] For this reason, an embodiment of the present application provides a computer-readable storage medium, in which multiple instructions are stored. The instructions can be loaded by a processor to execute the steps in any one of the posture recognition methods provided by the embodiments of the present application. For example, the instructions can execute the following steps:

[0257] Obtain the point cloud data of the target object; determine the state of the target object according to the point cloud data; if the state is a moving state, then determine the transition posture of the target object based on the point cloud data; determine the target posture of the target object according to the most recent historical posture and the transition posture of the target object.

[0258] For the specific implementation of each of the above operations, reference may be made to the previous embodiments and will not be elaborated herein.

[0259] Among them, the computer-readable storage medium may include: read-only memory (ROM, Read Only Memory), random access memory (RAM, Random Access Memory), magnetic disk or optical disk, etc.

[0260] Since the instructions stored in the computer-readable storage medium can execute the steps in any one of the data processing methods provided by the embodiments of the present application, the beneficial effects that can be achieved by any one of the data processing methods provided by the embodiments of the present application can be realized. For details, reference may be made to the previous embodiments and will not be elaborated herein.

[0261] According to one aspect of the present application, there is provided a computer program product or a computer program, which includes computer instructions stored in a computer-readable storage medium. A processor of a computer device reads the computer instructions from the computer-readable storage medium, and the processor executes the computer instructions, so that the computer device executes the methods provided in the various alternative implementations provided in the above embodiments.

[0262] The above has introduced in detail a gesture recognition method, device and computer-readable storage medium provided by the embodiments of the present application. Specific examples are used in this article to elaborate on the principle and implementation manner of the present application. The description of the above embodiments is only used to help understand the method and its core idea of the present application; at the same time, for those skilled in the art, according to the idea of the present application, there will be changes in the specific implementation manner and application scope. In summary, the content of this specification should not be construed as a limitation to the present application.

Claims

1. A posture recognition method, characterized in that, Including: Obtain the point cloud data of the target object; Determine the state of the target object according to the point cloud data; If the state is a moving state, determine the transition pose of the target object based on the point cloud data; Determine the target pose of the target object according to the nearest historical pose and the transition pose of the target object.

2. The method according to claim 1, characterized in that, The determining the state of the target object according to the point cloud data includes: Determine the number of static points and the number of dynamic points in the point cloud data; Determine the state of the target object according to the number of static points and the number of dynamic points.

3. The method according to claim 2, characterized in that, The determining the state of the target object according to the number of static points and the number of dynamic points includes: Determine the dynamic weight of the target object according to the number of static points and the number of dynamic points; Determine the state of the target object according to the number of dynamic points and the dynamic weight.

4. The method according to claim 3, characterized in that, The determining the state of the target object according to the number of dynamic points and the dynamic weight includes: If the product of the number of dynamic points and the dynamic weight is greater than or equal to a preset threshold, determine that the state is a moving state; If the product of the number of dynamic points and the dynamic weight is less than the preset threshold, determine that the state is a stationary state.

5. The method according to claim 1, characterized in that, The determining the transition pose of the target object based on the point cloud data includes: Determine the pose feature vector of the target object according to the point cloud data; Input the pose feature vector into a preset classifier to obtain a classification result; Determine the transition pose of the target object according to the classification result.

6. The method according to claim 5, characterized in that, Before the determining the pose feature vector of the target object according to the point cloud data, it further includes: Determine the motion trajectory associated with the target object; Determine the historical frame point cloud data associated with the motion trajectory; The determining the pose feature vector of the target object according to the point cloud data includes: Determine the pose feature vector of the target object according to the current frame point cloud data and the historical frame point cloud data.

7. The method according to claim 6, characterized in that, The determining the pose feature vector of the target object according to the current frame point cloud data and the historical frame point cloud data includes: Extract a preset number of point clouds with the strongest signal intensity from the current frame point cloud data and the historical frame point cloud data respectively; Determine the intra-frame feature vectors of the current frame point cloud data and the historical frame point cloud data respectively according to the extracted point clouds and a preset step; Determine the inter-frame feature vector between the current frame point cloud data and the historical frame point cloud data according to the extracted point clouds and the preset step; Concatenate the intra-frame feature vector and the inter-frame feature vector to obtain the pose feature vector of the target object.

8. The method according to claim 5, characterized in that, There are multiple preset classifiers, and the inputting the pose feature vector into a preset classifier to obtain a classification result includes: Determine the corresponding target preset classifier according to the nearest historical pose of the target object; Input the pose feature vector into the target preset classifier to obtain a classification result.

9. The method according to claim 8, characterized in that,If there are multiple target preset classifiers, then the determining the transition pose of the target object according to the classification result includes: Determine the transition state of the target object according to the multiple classification results.

10. The method according to any one of claims 1-9, characterized in that, The method further includes: If the state is a stationary state, determine the most recent historical pose of the target object as the target pose of the target object.

11. The method according to any one of claims 1-9, characterized in that, The target pose includes a motion pose and a stationary pose. After determining the target pose of the target object according to the most recent historical pose and the transition pose of the target object, it further includes: If the target pose belongs to the stationary pose, determine the most recent historical position information of the target object as the position information of the target object; If the target pose belongs to the motion pose, determine the position information of the target object according to the point cloud data.

12. An object pose detection device, characterized in that, It includes: An acquisition unit for acquiring point cloud data of a target object; A first determination unit for determining the state of the target object according to the point cloud data; An identification unit for, if the state is a motion state, determining the transition pose of the target object based on the point cloud data; A second determination unit for determining the target pose of the target object according to the most recent historical pose and the transition pose of the target object.

13. A computer device, characterized in that, It includes a processor and a memory. The memory stores a computer program, and the processor is used to run the computer program in the memory to implement the method according to any one of claims 1 to 11.

14. A computer-readable storage medium, characterized in that, The computer-readable storage medium is computer-readable and stores multiple instructions, and the instructions are suitable for being loaded by a processor to execute the method according to any one of claims 1 to 11.

Images