Acceleration annotation model generation method, acceleration annotation method, device and medium

Abstract

The invention discloses an acceleration labeling model generation method, an acceleration labeling method, a device and a medium. The acceleration labeling model generation method comprises acquiring original data, wherein the original data comprises a training inertial measurement signal and a training acceleration signal corresponding to the same data identifier; analyzing the training inertial measurement signal to obtain a labeled action category corresponding to the training inertial measurement signal; analyzing the training acceleration signal to obtain a labeled acceleration feature corresponding to the training acceleration signal; forming a labeled training sample based on the labeled action category and the labeled acceleration feature corresponding to the same data identifier; and using the labeled training sample to train a neural network model to generate a target acceleration labeling model. The target acceleration labeling model can directly obtain a target action category according to an acceleration signal to be processed, is easy to use, and has a high recognition accuracy.

Description

Technical Field

[0001] The present invention relates to the field of motion recognition, and in particular to an acceleration labeling model generation method, an acceleration labeling method, a device and a medium. Background Art

[0002] At present, with the development of sensing technology and digital technology, there are more and more methods to determine the action category by detecting the gait information of human motion. The existing technology usually generates a recognition model to identify the motion information and determine the target action category. During the training process of the recognition model, the training samples need to be manually marked in advance, and then the training samples are manually windowed and segmented. This process consumes a lot of manpower and material resources, and is highly dependent on personal experience. The screened acceleration signal and the corresponding action category may be inaccurate, resulting in poor action category recognition effect. Since a model can only recognize one action category, it is not practical. Alternatively, the target action category is identified by using a recognition method. The recognition method requires the placement of acceleration sensors on multiple joints of the human body to ensure a high recognition accuracy. This method can intuitively discover the acceleration characteristics of various action categories, but in actual applications, users are required to carry multiple sensors, which is very inconvenient, has limited usage scenarios, and is costly. Summary of the invention

[0003] The embodiments of the present invention provide an acceleration labeling model generation method, an acceleration labeling method, a device and a medium to solve the problem of inaccurate action recognition or inconvenient recognition operation in the prior art.

[0004] A method for generating an acceleration annotation model, comprising:

[0005] Acquire raw data, where the raw data includes a training inertial measurement signal and a training acceleration signal corresponding to the same data identifier;

[0006] Analyzing the training inertial measurement signal to obtain a marked action category corresponding to the training inertial measurement signal;

[0007] Analyzing the training acceleration signal to obtain a labeled acceleration feature corresponding to the training acceleration signal;

[0008] Based on the annotated action category and the annotated acceleration feature corresponding to the same data identifier, forming an annotated training sample;

[0009] The labeled training samples are used to train a neural network model to generate a target acceleration labeled model.

[0010] An acceleration marking method, comprising:

[0011] Acquiring data to be processed, wherein the data to be processed includes an acceleration signal to be processed;

[0012] Analyzing the acceleration signal to be processed to obtain acceleration features to be processed corresponding to the acceleration signal to be processed;

[0013] The acceleration feature to be processed is input into an acceleration labeling model to obtain a target action category corresponding to the acceleration feature to be processed.

[0014] A computer device comprises a memory, a processor and a computer program stored in the memory and executable on the processor, wherein the processor implements the steps of the acceleration annotation model generation method when executing the computer program.

[0015] A computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the steps of the acceleration annotation model generation method are implemented.

[0016] The above-mentioned acceleration labeling model generation method, acceleration labeling method, device and medium analyze the training inertial measurement signal to obtain the labeled action category corresponding to the training inertial measurement signal, and use the training inertial measurement signal to obtain the labeled action category, so that the labeled action category is objective and avoids manual intervention in the model generation process. The training acceleration signal is analyzed to obtain the labeled acceleration feature corresponding to the training acceleration signal, providing technical support for subsequent neural network model training. Based on the labeled action category and the labeled acceleration feature corresponding to the same data identifier, a labeled training sample is formed, and the training inertial measurement signal is used to obtain the labeled action category, so that the labeled action category is objective and avoids manual intervention in the model generation process. The labeled training sample is used to train the neural network model to generate a target acceleration labeling model, effectively reducing manual intervention and ensuring that the generated target acceleration labeling model is objective. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings required for use in the description of the embodiments of the present invention will be briefly introduced below. Obviously, the accompanying drawings in the following description are only some embodiments of the present invention. For ordinary technicians in this field, other accompanying drawings can be obtained based on these accompanying drawings without paying creative labor.

[0018] Figure 1 is a flow chart of a method for generating an acceleration annotation model in one embodiment of the present invention;

[0019] Figure 2 is another flow chart of a method for generating an acceleration annotation model in one embodiment of the present invention;

[0020] Figure 3 is another flow chart of a method for generating an acceleration annotation model in one embodiment of the present invention;

[0021] Figure 4 is another flow chart of a method for generating an acceleration annotation model in one embodiment of the present invention;

[0022] Figure 5 is another flow chart of a method for generating an acceleration annotation model in one embodiment of the present invention;

[0023] Figure 6 is another flow chart of a method for generating an acceleration annotation model in one embodiment of the present invention;

[0024] Figure 7 is a flow chart of an acceleration labeling method in one embodiment of the present invention;

[0025] Figure 8 is a schematic diagram of a computer device in one embodiment of the present invention. DETAILED DESCRIPTION

[0026] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0027] In one embodiment, if Figure 1 As shown, a method for generating an acceleration annotation model is provided, comprising the following steps:

[0028] S101: Acquire original data, where the original data includes a training inertial measurement signal and a training acceleration signal corresponding to the same data identifier.

[0029] Among them, the original data is the data collected for training the neural network model.

[0030] The training inertial measurement signal is a signal collected by an inertial measurement sensor (i.e., an IMU sensor). The training acceleration signal is a signal collected by a mobile device or an acceleration sensor. In this embodiment, the test subject carries an inertial measurement sensor and a mobile device, or carries an inertial measurement sensor and an acceleration sensor to walk, jump, or perform other actions, so as to collect training inertial measurement signals and training acceleration signals corresponding to the same data identifier, so as to subsequently train the neural network model to obtain a target acceleration labeling model, and the target acceleration labeling model can be used to identify the acceleration signal to determine the action category corresponding to the acceleration signal, effectively reducing the number of acceleration sensors and reducing costs. It can be understood that the target acceleration labeling model can be used to directly identify the acceleration signal to be processed collected by the mobile device or the acceleration sensor, so as to determine the corresponding action category without the need for various acceleration sensors. The method is simple and has a wide range of application scenarios.

[0031] It should be noted that the training inertial measurement signal and the training acceleration signal carry a data identifier, which is a unique identifier formed based on the test object information and time information, so as to subsequently determine the training inertial measurement signal and the training acceleration signal of the same data identifier of the test object to train the neural network model. Among them, the time information is the time when the training inertial measurement signal or the training acceleration signal is obtained; the test object information includes the object identifier, and also includes age, height, weight and gender. In this example, during the original data acquisition process, a unique data identifier can be generated based on the object identifier and the time information obtained in real time, so that the training inertial measurement signal collected by the inertial measurement sensor carries the data identifier, and the training acceleration signal collected by the mobile device or the acceleration sensor also carries the data identifier. It can be understood that before the training inertial measurement signal and the training acceleration signal are analyzed, the training inertial measurement signal and the training acceleration signal need to be screened and processed to exclude abnormal signals and ensure the accuracy of the subsequent target acceleration labeling model.

[0032] As an example, the raw data of at least one test object can be collected to train the neural network model and obtain the target acceleration labeling model. As another example, multiple test objects can be classified in advance according to the test object information such as age and weight, so that the test objects with similar test object information are grouped together, and the raw data in the group are collected to obtain the training inertial measurement signal and the training acceleration signal, so as to eliminate the influence of individual differences on the training process and ensure the accuracy of the target acceleration labeling model. In subsequent recognition, the users to be recognized are classified in advance to improve the recognition accuracy.

[0033] S102: Analyze the training inertial measurement signal to obtain a labeled action category corresponding to the training inertial measurement signal.

[0034] The annotated action category is the category of the action behavior of the test object during the movement process, and the annotated action category and the corresponding training acceleration signal are used to train the neural network model. The annotated action category includes but is not limited to slow walking, fast walking, running and jumping, etc., that is, the target acceleration annotation model generated later can determine the corresponding action categories such as slow walking, fast walking, running and jumping according to the acceleration signal.

[0035] Specifically, the server analyzes the training inertial measurement signal to obtain the joint movement posture of the test object during the movement corresponding to the training inertial measurement signal, and then obtains the labeled action category corresponding to the training inertial measurement information according to the joint movement posture. In this embodiment, the labeled action category is obtained by using the training inertial measurement signal, so that the labeled action category is objective and avoids manual intervention in the model generation process. Taking the joint movement posture of the lower limb joint posture when the human body walks as an example, if the initial step is the right heel touching the ground, the right leg hip joint extends backward, and the right leg as a whole is also moving; at the same time, the left leg hip joint bends forward, and the left leg swings as a whole until the left toe leaves the ground. After the left toe leaves the ground, it enters the single-leg support state, the left leg hip joint continues to bend forward, and the left leg follows; at the same time, the right leg hip joint continues to extend backward; then the right hip joint continues to extend backward, the left hip joint continues to bend forward, and the left knee joint begins to extend forward. At this time, the center of gravity continues to move forward until the left heel touches the ground, and the single-leg support state is achieved. The support phase ends; then it enters the second double-leg support state, and the subsequent angle change process is symmetrical with the initial process, except that the left leg supports the right leg and then swings the left leg. At this time, the inertial sensor is used to collect the training inertial measurement signal in real time, and the training inertial measurement signal is analyzed to obtain the training joint angle information of the test process of the hip, ankle and knee joints of the lower limb posture, and the preset mapping rules are queried based on these training joint angle information to obtain the labeled action category corresponding to the training inertial measurement signal.

[0036] S103: Analyze the training acceleration signal to obtain the labeled acceleration feature corresponding to the training acceleration signal.

[0037] The annotated acceleration feature is a training acceleration signal converted into a corresponding acceleration value for representing acceleration information, and the annotated acceleration feature includes but is not limited to an acceleration vector. In this embodiment, the training acceleration signal is analyzed to obtain the annotated acceleration feature corresponding to the training acceleration signal, providing technical support for subsequent neural network model training.

[0038] S104: forming a labeled training sample based on the labeled action category and the labeled acceleration feature corresponding to the same data identifier.

[0039] The labeled training samples refer to samples used to train the neural network model. The labeled action categories and labeled acceleration features corresponding to the same data identifier used in this implementation are used as a set of labeled training samples to ensure that the labeled action categories and labeled acceleration features have correspondence, thereby ensuring the accuracy of the subsequent target acceleration labeling model.

[0040] In this embodiment, the labeled action categories are diverse, so that the subsequently generated target acceleration labeling model can determine the corresponding action category according to the acceleration signal collected in real time, so as to solve the problem in the prior art that a model can only identify one action category, and improve the performance of the target acceleration labeling model.

[0041] S105: Using the labeled training samples to train the neural network model to generate a target acceleration labeled model.

[0042] Among them, the target acceleration labeling model is a model determined by training based on the labeled training samples, and can reflect the mapping relationship between the acceleration characteristics and the corresponding action categories. In this embodiment, the training inertial measurement signal is collected by the training inertial measurement sensor carried by the test subject, and the joint movement posture of the test subject can be effectively restored to determine the labeled action category carrying the data identification; the training acceleration signal is collected by the mobile device or acceleration sensor carried by the test subject, and the training acceleration signal is analyzed to determine the labeled acceleration characteristics carrying the data identification; and the labeled action category and the labeled acceleration characteristics of the same data identification are formed into a labeled training sample, so as to train the neural network model using the labeled training sample, so as to generate a target acceleration labeling model for accurately predicting the action category corresponding to the acceleration signal, using a small number of sensors, and effectively reducing human intervention, ensuring that the generated target acceleration labeling model is objective and has a wide range of application scenarios.

[0043] The acceleration labeling model generation method provided in this embodiment analyzes the training inertial measurement signal to obtain the labeled action category corresponding to the training inertial measurement signal, and uses the training inertial measurement signal to obtain the labeled action category, so that the labeled action category is objective and avoids manual intervention in the model generation process. The training acceleration signal is analyzed to obtain the labeled acceleration feature corresponding to the training acceleration signal, providing technical support for subsequent neural network model training. Based on the labeled action category and labeled acceleration feature corresponding to the same data identifier, a labeled training sample is formed, and the labeled action category is obtained using the training inertial measurement signal, so that the labeled action category is objective and avoids manual intervention in the model generation process. The labeled training sample is used to train the neural network model to generate a target acceleration labeling model, which effectively reduces manual intervention and ensures that the generated target acceleration labeling model is objective. In actual use, the target acceleration labeling model directly obtains the target action category according to the acceleration signal to be processed, which is easy to use and has a better recognition effect.

[0044] In one embodiment, if Figure 2 As shown, step S102, i.e., analyzing the training inertial measurement signal to obtain the labeled action category corresponding to the training inertial measurement signal, includes:

[0045] S201: Based on the training inertial measurement signal, obtain training movement information corresponding to the training inertial measurement signal, where the training movement information includes training joint angle information and movement time information.

[0046] The training motion information is information related to joints and time during the movement of the test subject.

[0047] The training joint angle information is the movement angle information of the joints of the test subject during the test, including the maximum angle, minimum angle, angle change range and joint position of the joint. For example, if the joints are hip joints, ankle joints and knee joints, the training joint angle information may be the joint position and the hip flexion range is 0 to 35°, the hip extension range is 0 to 7°, the knee flexion range is 0 to 60°, the ankle dorsiflexion range is 0 to 15° and the flexion range is 0 to 20°.

[0048] The motion time information refers to the time that the test subject moves during the test. The motion time information includes at least one joint motion cycle, and each joint motion cycle corresponds to an action category, so that a labeled action category for training the neural network can be analyzed based on a joint motion cycle to ensure the validity of the labeled action category. In this embodiment, the training inertial measurement signal is analyzed to quickly determine the training action information and avoid artificial interference.

[0049] S202: Analyze the training joint angle information and the movement time information to obtain the labeled action category corresponding to the training inertial measurement signal.

[0050] Specifically, the training joint angle information is identified to determine the maximum joint angle, minimum joint angle and joint angle change range, etc., to determine all action categories completed in the motion time information, and the motion time information is analyzed to determine all joint motion cycles included in the motion process, to determine the labeled action category of each joint motion cycle of the test subject during the motion process, to provide support for the subsequent determination of the labeled acceleration characteristics and the corresponding labeled action categories.

[0051] The acceleration annotation model generation method provided in this embodiment obtains the training action information corresponding to the training inertial measurement signal based on the training inertial measurement signal, and the training action information includes the training joint angle information and the motion time information. The training inertial measurement signal is analyzed to quickly determine the training action information and avoid human interference, so as to subsequently determine the annotated action category corresponding to the training inertial measurement signal. The training joint angle information and the motion time information are analyzed to obtain the annotated action category corresponding to the training inertial measurement signal, so that the determination process of the annotated action category is more objective, avoids human interference, and provides support for the subsequent determination of the annotated acceleration features and the corresponding annotated action category, so that the model generated by the subsequent training is objective.

[0052] In one embodiment, if Figure 3 As shown, step S202, analyzing the training joint angle information and the motion time information to obtain the labeled action category corresponding to the training inertial measurement signal, includes:

[0053] S301: Analyze the training joint angle information and movement time information to obtain the joint movement cycle and the joint movement posture within the joint movement cycle.

[0054] The joint motion cycle refers to the cycle used by the test subject to complete an action category, for example, the joint motion cycle required for the test subject to complete the action category of walking (i.e., taking a step) is 1 second. It can be understood that the motion time information includes at least one joint motion cycle.

[0055] Taking human walking as an example, if the initial step is that the right heel touches the ground, the right hip joint extends backward, and the right leg as a whole is also moving; at the same time, the left hip joint bends forward, and the left leg swings as a whole until the left toe leaves the ground. After the left toe leaves the ground, it enters the single-leg support state, and the left hip joint continues to bend forward, and the left leg follows; at the same time, the right hip joint continues to extend backward; then the right hip joint continues to extend backward, the left hip joint continues to bend forward, and the left knee joint begins to extend forward. At this time, the center of gravity continues to move forward until the left heel touches the ground, and the single-leg support phase ends; then it enters the second double-leg support state, and the subsequent angle changes and other processes are symmetrical with the initial process, except that the left leg supports the right leg and swings, and then the left leg swings. At this point, a complete right lower limb walking cycle is over, that is, a complete joint movement cycle is completed.

[0056] The joint movement posture refers to the joint movement of the test object during the test process, specifically, the joint movement posture when the test object completes the action category during the test process. For example, the joint movement posture is the maximum joint angle and the range of joint angle change.

[0057] In this implementation, the joint movement cycle and the joint movement posture within the joint movement cycle are determined based on the training joint angle information and movement time information, so that the subsequent labeled action categories within the joint movement cycle are objective, thereby solving the errors caused by the need to manually screen out specific action categories in the prior art.

[0058] S302: Determine the joint movement angle range based on the joint movement posture.

[0059] Specifically, the joint motion posture mainly includes the joint position. In this example, the server can determine the joint motion speed and joint motion acceleration according to the joint position and the joint motion cycle, and substitute the joint motion speed, joint motion acceleration and joint position values ​​into the inverse dynamics equation to obtain the joint motion angle range, so as to objectively judge the annotation action category in the future and reduce manual interference.

[0060] S303: Determine the joint extension time and the joint flexion time based on the joint movement cycle and the joint movement angle range.

[0061] The joint extension time refers to the time of joint extension in the joint movement cycle, which can be understood as the time of joint extension in the process of completing a movement category. The joint bending time refers to the time of joint bending in the joint movement cycle, which can be understood as the time of joint bending in the process of completing a movement category, so as to objectively judge the labeled action category in the future and reduce manual interference.

[0062] S304: Based on the joint motion angle range, joint extension time, and joint bending time, obtain the labeled action category corresponding to the training inertial measurement signal.

[0063] Specifically, the preset mapping rules are used to process the joint motion angle range, joint extension time and joint bending to determine the labeled action category that matches the joint motion angle range, joint extension time and joint bending time, making the determination process of the labeled action category more objective and the labeled action category more accurate. This provides technical support for the subsequent training of the target acceleration labeling model based on the labeled action category and the labeled acceleration features.

[0064] The preset mapping rule refers to the mapping relationship between the training inertial measurement signal and the annotated action category specified in advance. For example, when the annotated action category is walking, the preset mapping rule is that the joint motion angle range of the hip flexion activity is 0 to 35°, the joint motion angle range of the extension activity is 0 to 7°, the joint motion angle range of the knee flexion activity is 0 to 60°, the joint motion angle range of the ankle dorsiflexion activity is 0 to 15°, the joint motion angle range of the flexion activity is 0 to 20°, and the normal step frequency is 95 to 125 / min.

[0065] The acceleration labeling model generation method provided in this embodiment analyzes the training joint angle information and movement time information, obtains the joint movement cycle and the joint movement posture within the joint movement cycle, so as to be objective in the subsequent labeling of action categories based on the joint movement cycle, so as to solve the mistakes caused by the need to manually screen out specific action categories in the prior art. Based on the joint movement posture, the joint movement angle range is determined, and based on the joint movement cycle and the joint movement angle range, the joint extension time and joint bending time are determined, so as to objectively judge the labeling action category in the subsequent process and reduce manual interference. Based on the joint movement angle range, the joint extension time and the joint bending time, the labeling action category corresponding to the training inertial measurement signal is obtained, providing technical support for the subsequent training of the target acceleration labeling model according to the labeling action category and the labeling acceleration characteristics.

[0066] In one embodiment, if Figure 4 As shown, step S101, i.e. obtaining original data, includes:

[0067] S401: using an inertial measurement sensor disposed on the test object to collect a training inertial measurement signal carrying a data identifier in real time.

[0068] In this embodiment, an inertial measurement sensor can be used to restore the joint movement posture of the test object, so as to determine the labeled action category of the test object based on the joint movement posture, which is objective and makes the subsequently generated target acceleration labeling model more accurate, so as to solve the problem in the prior art that the labeled action category is manually determined based on the acceleration signal collected by the acceleration sensor, which has strong manual dependence and may cause errors.

[0069] S402: Using a mobile device or an acceleration sensor disposed on the same test object, a training acceleration signal carrying a data identifier is collected in real time.

[0070] In this embodiment, a mobile device or acceleration sensor of the same test object is used to collect training acceleration signals to ensure that the training acceleration signals and training inertial measurement signals collected by the same test object at the same time carry the same data identifier, wherein the mobile device can be a mobile device such as a smart watch and a mobile phone.

[0071] It can be understood that when a mobile device set on the same test object is used to collect training acceleration signals, only inertial measurement sensors need to be set in the solution, thereby reducing the number of sensors. While ensuring the accuracy of model training, costs can be reduced. The subsequent application of the target acceleration labeling model is very convenient and suitable for a variety of scenarios.

[0072] S403: Acquire raw data based on the training inertial measurement signal and the training acceleration signal with the same data identifier.

[0073] The acceleration labeling model generation method provided in this embodiment uses an inertial measurement sensor set on the test object to collect the training inertial measurement signal carrying the data identifier in real time, which is objective, so that the target acceleration labeling model generated subsequently is more accurate, so as to solve the problem that the manual determination of the labeling action category based on the acceleration signal collected by the acceleration sensor in the prior art is highly manual and may cause errors. A mobile device or acceleration sensor set on the same test object is used to collect the training acceleration signal carrying the data identifier in real time to ensure that the training acceleration signal carrying the same data identifier and the training inertial measurement signal have a corresponding relationship, so as to ensure the accuracy of the original data obtained based on the training inertial measurement signal and the training acceleration signal with the same data identifier, and ensure the feasibility of model training using the target acceleration labeling model.

[0074] In one embodiment, the inertial measurement sensors are arranged on the hip joint, knee joint and ankle joint of the test subject. In this embodiment, the inertial measurement sensors are arranged on the hip joint, knee joint and ankle joint of the test subject to respectively collect the inertial measurement signals corresponding to the hip joint, knee joint and ankle joint of the test subject, so as to restore the joint motion posture of the test subject during the motion process according to the inertial measurement signals corresponding to the hip joint, knee joint and ankle joint, and then determine the joint angles of the hip joint, knee joint and ankle joint, so as to provide a basis for the subsequent determination of the labeled action category and the labeled acceleration characteristics of the data identification.

[0075] In one embodiment, if Figure 5As shown, before step S401, that is, before using the inertial measurement sensor set on the test object to collect the training inertial measurement signal carrying the data identifier in real time, the acceleration annotation model generation method also includes:

[0076] S501: using an inertial measurement sensor disposed on the test object to collect and calibrate inertial measurement signals in real time.

[0077] Specifically, an inertial measurement sensor is installed on the joint of the test subject, the test subject stands still, and the calibration inertial measurement signal is measured to calibrate the calibration inertial measurement signal measured by the inertial measurement sensor. In this example, the inertial measurement sensors are installed on the joints of the test subject, that is, the inertial measurement sensors are installed on the hip joint, the knee joint, and the ankle joint, respectively.

[0078] S502: Convert the calibration inertial measurement signal to obtain calibration joint coordinates corresponding to the calibration inertial measurement signal.

[0079] Typically, the calibration inertial measurement signal is waveform data. A computer is used to convert the standard inertial measurement signal to obtain the joint position and joint motion angle range. A coordinate system is established in the joint coordinates to obtain the calibration joint coordinates so that the calibration mapping relationship can be obtained later.

[0080] S503: Acquire a calibration mapping relationship based on the calibration joint coordinates and the preset joint coordinates.

[0081] The preset joint coordinates are coordinates on the preset joint coordinate system. Since the coordinate systems in each inertial measurement sensor are inconsistent, the training joint angle information subsequently determined by the inertial measurement sensors on each joint is inaccurate. Therefore, the preset joint coordinate system is set to obtain the calibration mapping relationship between the coordinate system in each inertial measurement sensor and the preset joint coordinate system. The preset joint coordinate system is used to ensure that the training joint angle information is formed based on a unified calibration standard, so as to ensure that accurate training joint angle information is obtained subsequently.

[0082] The calibration mapping relationship is the relationship between the coordinate system in the inertial measurement sensor and the preset joint coordinates. It can be understood that one inertial measurement sensor corresponds to one calibration mapping relationship to ensure that the subsequent training joint angle information is based on the same preset joint coordinate system. Accurate training joint angle information is obtained, thereby improving the accuracy of the subsequently generated model. For example, the calibration mapping relationship can be a function, etc.

[0083] The acceleration annotation model generation method provided in this embodiment uses an inertial measurement sensor set on a test object to collect a calibration inertial measurement signal in real time, convert the calibration inertial measurement signal, and obtain the calibration joint coordinates corresponding to the calibration inertial measurement signal, so as to obtain the calibration mapping relationship later. Based on the calibration joint coordinates and the preset joint coordinates, the calibration mapping relationship is obtained to ensure that the subsequent training joint angle information is converted based on the same preset joint coordinate system, and accurate training joint angle information is obtained, thereby improving the accuracy of the subsequently generated model.

[0084] Accordingly, step S201, i.e., based on the training inertial measurement signal, obtaining training action information corresponding to the training inertial measurement signal, the training action information including training joint angle information and motion time information, specifically includes the following steps:

[0085] S2011: Based on the training inertial measurement signal, obtain the corresponding original joint angle information.

[0086] The original joint angle information is the uncalibrated joint angle information obtained after analyzing the training measurement signal. The original joint angle information includes the maximum angle, the minimum angle, the angle variation range and the joint position.

[0087] S2012: Calibrate the original joint angle information using the calibration mapping relationship to obtain training joint angle information.

[0088] In this embodiment, the coordinates in the original joint angle information are transformed according to the calibration mapping relationship to obtain the training joint angle information of each joint. Since the training joint angle information is transformed based on the same preset joint coordinate system, the accuracy of the training joint angle information is improved, thereby improving the accuracy of the subsequently generated model.

[0089] The acceleration annotation model generation method provided in this embodiment obtains the corresponding motion time information and original joint angle information based on the training inertial measurement signal, calibrates the original joint angle information using a calibration mapping relationship, obtains the training joint angle information, and improves the accuracy of the training joint angle information, thereby improving the accuracy of the subsequently generated model.

[0090] In one embodiment, if Figure 6 As shown, step S105, i.e., using the labeled training samples to train the neural network model to generate a target acceleration labeled model, includes:

[0091] S601: Use the labeled training samples to train the neural network model to generate an original acceleration labeled model.

[0092] S602: Obtain the labeling accuracy of the original acceleration labeling model. If the labeling accuracy is greater than a preset accuracy, determine the original acceleration labeling model as the target acceleration labeling model.

[0093] All labeled training samples are divided into a training set and a validation set. The labeled training samples in the training set are used to train the neural network model and generate the original acceleration labeling model. Then, the labeled training samples in the validation set are input into the original acceleration labeling model to obtain the verification action category corresponding to each labeled training sample; the number of accurate samples corresponding to the labeled training samples with the same labeled action category and verification action category is counted, and the total number of samples corresponding to all labeled training samples in the validation set is obtained. The accuracy of the original acceleration labeling model is calculated based on the number of accurate samples and the total number of samples. If the accuracy is greater than the preset accuracy, the original acceleration labeling model is determined as the target acceleration labeling model to ensure that the accuracy of the target acceleration labeling model reaches the preset standard.

[0094] The acceleration labeling model generation method provided in this embodiment uses the labeling training sample to train the neural network model to generate the original acceleration labeling model. The labeling accuracy of the original acceleration labeling model is obtained. If the labeling accuracy is greater than the preset accuracy, the original acceleration labeling model is determined as the target acceleration labeling model to ensure the accuracy of the target acceleration labeling model.

[0095] The present invention provides an acceleration labeling method, comprising the following steps:

[0096] S701: Acquire data to be processed, where the data to be processed includes acceleration signals to be processed.

[0097] Among them, the data to be processed refers to the motion data collected in real time, and the data to be processed is processed to determine the corresponding target action category. In this embodiment, the data to be processed is collected by a mobile device or an acceleration sensor, and an inertial measurement sensor is not required. It is easy to use and can be applied to more scenarios, and the cost is effectively reduced. For example, for the care of the elderly, the mobile device carried by the elderly is used to collect and obtain the data to be processed to determine the corresponding target action category, realize the monitoring of the elderly's movements, and quickly alarm for abnormal behaviors such as falls, so as to avoid causing greater harm.

[0098] S702: Analyze the acceleration signal to be processed to obtain acceleration features to be processed corresponding to the acceleration signal to be processed.

[0099] In this embodiment, the acceleration signal to be processed is analyzed to obtain the corresponding acceleration feature to be processed, so that the acceleration feature to be processed can be used for subsequent target acceleration labeling model calculation.

[0100] S703: Input the acceleration feature to be processed into the target acceleration labeling model of steps S101 to S105 to obtain the target action category corresponding to the acceleration feature to be processed.

[0101] The target action category is the action category corresponding to the acceleration feature to be processed. In this embodiment, the acceleration feature to be processed collected by the mobile device is input into the target acceleration annotation model, so that the corresponding target action category can be quickly determined.

[0102] The acceleration labeling method provided in this embodiment obtains data to be processed, and the data to be processed includes an acceleration signal to be processed, so as to ensure the accuracy of the target acceleration labeling model. The acceleration signal to be processed is analyzed to obtain the acceleration feature to be processed corresponding to the acceleration signal to be processed, so that the acceleration feature to be processed can be used for subsequent target acceleration labeling model calculation. The acceleration feature to be processed is input into the target acceleration labeling model to obtain the target action category corresponding to the acceleration feature to be processed, so that the corresponding target action category can be quickly determined.

[0103] In one embodiment, a computer device is provided. The computer device may be a server, and its internal structure diagram may be as follows: Figure 8 As shown. The computer device includes a processor, a memory, a network interface and a database connected through a system bus. Among them, the processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system, a computer program and a database. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The database of the computer device is used to store the labeled action category and the labeled acceleration feature corresponding to the same data identifier. The network interface of the computer device is used to communicate with an external terminal through a network connection. When the computer program is executed by the processor, a method for generating an acceleration labeling model is implemented.

[0104] In one embodiment, a computer device is provided, including a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, the acceleration annotation model generation method in the above embodiment is implemented, for example Figure 1 Steps S101-S105 shown, or Figures 3 to 6 The steps shown in , or the acceleration labeling method in the above embodiment when the processor executes the computer program, for example Figure 7 To avoid repetition, steps S701-S703 are not described again here.

[0105] In one embodiment, a computer-readable storage medium is provided, on which a computer program is stored. When the computer program is executed by a processor, the acceleration annotation model generation method in the above embodiment is implemented, for example Figure 1 Steps S101-S105 shown, or Figures 3 to 6 The steps shown in , or the acceleration labeling method in the above embodiment when the processor executes the computer program, for example Figure 7 To avoid repetition, steps S701-S703 are not described again here.

[0106] Those skilled in the art can understand that all or part of the processes in the above-mentioned embodiment methods can be completed by instructing the relevant hardware through a computer program, and the computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the embodiments of the above-mentioned methods. Among them, any reference to memory, storage, database or other media used in the embodiments provided in this application can include non-volatile and / or volatile memory. Non-volatile memory can include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM) or flash memory. Volatile memory can include random access memory (RAM) or external cache memory. As an illustration and not limitation, RAM is available in many forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), double data rate SDRAM (DDRSDRAM), enhanced SDRAM (ESDRAM), synchronous link (Synchlink) DRAM (SLDRAM), memory bus (Rambus) direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM (RDRAM).

[0107] Those skilled in the art can clearly understand that for the convenience and simplicity of description, only the division of the above-mentioned functional units and modules is used as an example. In actual applications, the above-mentioned functions can be distributed and completed by different functional units and modules as needed, that is, the internal structure of the device can be divided into different functional units or modules to complete all or part of the functions described above.

[0108] The embodiments described above are only used to illustrate the technical solutions of the present invention, rather than to limit the same. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that the technical solutions described in the aforementioned embodiments may still be modified, or some of the technical features may be replaced by equivalents. Such modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present invention, and should all be included in the protection scope of the present invention.

Claims

1. A method for generating an acceleration annotation model, characterized in that: include: Acquire original data, wherein the original data includes a training inertial measurement signal and a training acceleration signal corresponding to the same data identifier; the data identifier includes test object information and time information, and the time information is the time corresponding to acquiring the training inertial measurement signal or acquiring the training acceleration signal; The test object information is used to group the test objects and obtain the original data of the group; Analyzing the training inertial measurement signal to obtain a marked action category corresponding to the training inertial measurement signal; Analyzing the training acceleration signal to obtain a labeled acceleration feature corresponding to the training acceleration signal; Based on the annotated action category and the annotated acceleration feature corresponding to the same data identifier, forming an annotated training sample; The labeled training samples are used to train a neural network model to generate a target acceleration labeled model.

2. The acceleration annotation model generation method according to claim 1, characterized in that: The analyzing the training inertial measurement signal to obtain the labeled action category corresponding to the training inertial measurement signal includes: Based on the training inertial measurement signal, obtaining training movement information corresponding to the training inertial measurement signal, wherein the training movement information includes training joint angle information and movement time information; The training joint angle information and the movement time information are analyzed to obtain a labeled action category corresponding to the training inertial measurement signal.

3. The acceleration annotation model generation method according to claim 2, characterized in that: The analyzing the training joint angle information and the movement time information to obtain the labeled action category corresponding to the training inertial measurement signal includes: Analyzing the training joint angle information and the movement time information to obtain a joint movement cycle and a joint movement posture within the joint movement cycle; Based on the joint movement posture, determining the joint movement angle range; Determine the joint extension time and the joint flexion time based on the joint movement cycle and the joint movement angle range; Based on the joint motion angle range, joint extension time and joint bending time, a labeled action category corresponding to the training inertial measurement signal is obtained.

4. The acceleration annotation model generation method according to claim 1, characterized in that: The obtaining of original data comprises: Using an inertial measurement sensor installed on the test object, the training inertial measurement signal carrying data identification is collected in real time; Using a mobile device or an acceleration sensor disposed on the same test object to collect a training acceleration signal carrying a data identifier in real time; Original data is acquired based on the training inertial measurement signal and the training acceleration signal with the same data identifier.

5. The acceleration annotation model generation method according to claim 4, characterized in that: The inertial measurement sensors are disposed on the hip joint, the knee joint, and the ankle joint of the test subject.

6. The acceleration annotation model generation method according to claim 4, characterized in that: Before the inertial measurement sensor disposed on the test object is used to collect the training inertial measurement signal carrying the data identifier in real time, the acceleration annotation model generation method further includes: Using an inertial measurement sensor installed on the test object to collect and calibrate inertial measurement signals in real time; Converting the calibration inertial measurement signal to obtain calibration joint coordinates corresponding to the calibration inertial measurement signal; Based on the calibration joint coordinates and the preset joint coordinates, obtaining a calibration mapping relationship; The method of acquiring training motion information corresponding to the training inertial measurement signal based on the training inertial measurement signal, wherein the training motion information includes training joint angle information and movement time information, comprises: Based on the training inertial measurement signal, obtaining corresponding original joint angle information; The calibration mapping relationship is used to calibrate the original joint angle information to obtain training joint angle information.

7. The acceleration annotation model generation method according to claim 1, characterized in that: The labeled training samples are used to train a neural network model to generate a target acceleration labeled model, including: Using the labeled training samples to train a neural network model to generate an original acceleration labeled model; The labeling accuracy of the original acceleration labeling model is obtained, and if the labeling accuracy is greater than a preset accuracy, the original acceleration labeling model is determined as a target acceleration labeling model.

8. A method for marking acceleration, characterized in that: include: Acquiring data to be processed, wherein the data to be processed includes an acceleration signal to be processed; Analyzing the acceleration signal to be processed to obtain acceleration features to be processed corresponding to the acceleration signal to be processed; The acceleration feature to be processed is input into the acceleration labeling model described in any one of claims 1-7 to obtain the target action category corresponding to the acceleration feature to be processed.

9. A computer device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that: When the processor executes the computer program, the acceleration labeling model generating method according to any one of claims 1 to 7 is implemented, or when the processor executes the computer program, the acceleration labeling method according to claim 8 is implemented.

10. A computer-readable storage medium storing a computer program, characterized in that: When the computer program is executed by a processor, the acceleration labeling model generation method according to any one of claims 1 to 7 is implemented, or when the computer program is executed by a processor, the acceleration labeling method according to claim 8 is implemented.

Images