Motion monitoring method and device, storage medium and motion camera

Abstract

The motion monitoring method provided by the invention is applied to a motion camera, the motion camera is provided with a sensor, and the method comprises the following steps: obtaining sensor data of the motion camera, the sensor data comprising an angular velocity, a linear acceleration and an anti-shake compensation displacement; resolving the attitude angle of the motion camera, and converting the angular velocity and the linear acceleration into the same coordinate system by using the attitude angle; under the same coordinate system, according to the angular velocity and the linear acceleration, a mapping relation between the angular velocity and the anti-shake compensation displacement is constructed, and the mapping relation represents a corresponding relation between the motion amplitude of the motion camera and the anti-shake compensation displacement; based on the angular velocity and the linear acceleration, after motion data features are extracted, the target motion state of the motion camera is determined in combination with the mapping relation; and based on the target motion state and in combination with sensor data, counting the motion amount corresponding to the motion camera. Therefore, on the premise of not increasing additional sensor or hardware burden, the hardware potential of the motion camera is fully excavated, and the accuracy and applicability of motion monitoring are improved.

Description

Technical Field

[0001] The present application relates to the field of motion monitoring technology, and in particular to a motion monitoring method, device, storage medium, and motion camera. Background Art

[0002] With growing public awareness of fitness, portable devices are widely used in the field of sports monitoring, and users are demanding higher accuracy. Currently, this function is concentrated in smart wearable devices and mobile terminals, but they suffer from hardware redundancy, data limitations, and poor adaptability to different scenarios, making it difficult to strike a balance between portability and monitoring accuracy. Existing solutions to improve monitoring accuracy and applicability often rely on adding dedicated sensors or combining multiple devices. This increases hardware cost and power consumption, limits the flexibility of functional integration, and is difficult to apply to lightweight, integrated devices.

[0003] Action cameras, with their compact design and rich hardware configuration, are frequently used in outdoor and extreme sports settings, and have the potential to become a new form of motion monitoring. However, existing action cameras primarily focus on image acquisition and image stabilization, lacking the ability to identify motion status and measure movement volume, making them difficult to meet users' needs for a comprehensive understanding of their movements. Consequently, existing technology has yet to achieve accurate motion tracking and movement volume measurement. Summary of the Invention

[0004] The purpose of this application is to solve at least one of the above-mentioned technical deficiencies, especially the technical deficiency in the prior art that the action camera has not yet achieved accurate recognition of the user's motion status and accurate statistics of the amount of motion.

[0005] In a first aspect, the present application provides a motion monitoring method, which is applied to a motion camera, wherein the motion camera is equipped with a sensor, and the method comprises:

[0006] Obtain sensor data from the motion camera, including angular velocity, linear acceleration, and anti-shake compensation displacement;

[0007] Calculate the attitude angle of the motion camera and use the attitude angle to convert the angular velocity and linear acceleration to the same coordinate system;

[0008] In the same coordinate system, a mapping relationship between angular velocity and linear acceleration and the anti-shake compensation displacement is established. The mapping relationship is used to represent the corresponding relationship between the motion amplitude of the action camera and the anti-shake compensation displacement.

[0009] Extract motion data features based on angular velocity and linear acceleration, and determine the target motion state of the motion camera based on the motion data features and mapping relationship;

[0010] Based on the target motion state and combined with sensor data, the motion amount corresponding to the motion camera is counted.

[0011] In one embodiment, the step of calculating the attitude angle of the motion camera includes:

[0012] The complementary filtering algorithm is used to fuse the angular velocity and linear acceleration data to update the attitude angle of the motion camera.

[0013] In one embodiment, the step of constructing a mapping relationship between the angular velocity and the linear acceleration and the anti-shake compensation displacement includes:

[0014] Acquire multiple sets of experimental data, each set of experimental data including experimental angular velocity, experimental linear acceleration, and experimental anti-shake compensation displacement;

[0015] Using each set of experimental data, the first calibration parameter of angular velocity and the second calibration parameter of linear acceleration are determined respectively through polynomial fitting;

[0016] A mapping relationship between the angular velocity and the linear acceleration and the anti-shake compensation displacement is established using the first calibration parameter and the second calibration parameter.

[0017] In one embodiment, the step of determining the target motion state of the motion camera according to the motion data characteristics and the mapping relationship includes:

[0018] Determine the initial motion state of the motion camera according to the motion data characteristics;

[0019] When the verification of the initial motion state based on the mapping relationship is passed, the target motion state of the motion camera is obtained.

[0020] In one embodiment, the step of determining the initial motion state of the motion camera according to the motion data characteristics includes:

[0021] Determine a first motion state classification result of the motion camera according to a preset motion data threshold condition for each motion state and motion data characteristics;

[0022] Input the motion data features into a preset machine learning model to obtain a second motion state classification result of the motion camera;

[0023] The first motion state classification result and the second motion state classification result are weightedly fused to obtain the initial motion state of the motion camera.

[0024] In one embodiment, the amount of movement includes the number of steps, and the steps of counting the amount of movement corresponding to the motion camera based on the target motion state and in combination with sensor data include:

[0025] If the target motion state is a state where step counting is possible, then within a preset time window with the current step counting time as the end time, the mean and standard deviation of the linear acceleration are calculated, and the current step counting threshold is determined based on the mean and standard deviation;

[0026] If the current linear acceleration exceeds the current step counting threshold, when the current linear acceleration is at its peak value and the time interval between the current step counting time point and the previous step counting time point is greater than the preset time difference threshold, the step count corresponding to the motion camera is increased by one.

[0027] In one embodiment, the amount of motion includes motion intensity, and the step of calculating the amount of motion corresponding to the motion camera based on the target motion state and in combination with sensor data includes:

[0028] According to the target exercise state, the current metabolic equivalent is determined and normalized to obtain the metabolic intensity;

[0029] Calculate mechanical strength based on sensor data within a preset sliding window;

[0030] After determining the initial cadence and current cadence based on the target motion state and sensor data, the fatigue level is calculated based on the initial cadence and current cadence;

[0031] The metabolic intensity, mechanical strength and fatigue degree are weighted and integrated to obtain the exercise intensity.

[0032] In a second aspect, the present application provides a motion monitoring device, which is applied to a motion camera. The motion camera is equipped with a sensor, and the device includes:

[0033] The sensor data acquisition module is used to obtain sensor data from the motion camera, including angular velocity, linear acceleration, and anti-shake compensation displacement;

[0034] The attitude angle calculation module is used to calculate the attitude angle of the motion camera and use the attitude angle to convert the angular velocity and linear acceleration into the same coordinate system;

[0035] A mapping relationship establishment module is used to establish a mapping relationship between angular velocity and linear acceleration and anti-shake compensation displacement in the same coordinate system. The mapping relationship is used to represent the corresponding relationship between the motion amplitude of the action camera and the anti-shake compensation displacement;

[0036] The target motion state determination module is used to extract motion data features based on angular velocity and linear acceleration, and determine the target motion state of the motion camera according to the motion data features and the mapping relationship;

[0037] The motion statistics module is used to count the motion amount corresponding to the motion camera based on the target motion state and combined with sensor data.

[0038] In a third aspect, the present application provides a storage medium: the storage medium stores computer-readable instructions, and when the computer-readable instructions are executed by one or more processors, the one or more processors execute the steps of any motion monitoring method in the above embodiments.

[0039] In a fourth aspect, the present application provides a motion camera, comprising: one or more processors, and a memory;

[0040] The memory stores computer-readable instructions, and when the computer-readable instructions are executed by one or more processors, the steps of any one of the motion monitoring methods in the above embodiments are performed.

[0041] It can be seen from the above technical solutions that the embodiments of the present application have the following advantages:

[0042] The motion monitoring method provided in this application addresses the problems of hardware redundancy, data limitations, and poor scene adaptability in existing motion monitoring systems, and proposes a lightweight solution that integrates the built-in sensor data of a motion camera. This method utilizes the angular velocity, linear acceleration, and anti-shake compensation displacement data already available in the motion camera, and achieves fusion processing of multi-source data in a unified coordinate system through attitude angle solution, thereby establishing a mapping relationship between motion amplitude and anti-shake compensation displacement, thereby compensating for the lack of perception of the actual degree of motion. Based on this mapping relationship and the results of motion data feature extraction, the current target motion state of the motion camera can be accurately identified, and further combined with sensor data to achieve accurate statistics of the amount of motion. Without adding additional sensors or hardware burden, this method fully taps the hardware potential of existing motion cameras, improves the accuracy and applicability of motion monitoring, has good integration and low power consumption advantages, and is conducive to promoting the functional expansion and application popularization of motion cameras in the field of motion monitoring. BRIEF DESCRIPTION OF THE DRAWINGS

[0043] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative labor.

[0044] Figure 1 A schematic diagram of a flow chart of a motion monitoring method provided in an embodiment of the present application;

[0045] Figure 2 A schematic diagram of the structure of a motion monitoring device provided in an embodiment of the present application;

[0046] Figure 3 A schematic diagram of the internal structure of a sports camera provided in an embodiment of the present application. DETAILED DESCRIPTION

[0047] The following will be combined with the drawings in the embodiments of this application to clearly and completely describe the technical solutions in the embodiments of this application. Obviously, the embodiments described are only part of the embodiments of this application, not all of the embodiments. Based on the embodiments in this application, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of this application.

[0048] The present application provides a motion monitoring method. The following embodiments are explained by taking the application of this method to a sports camera as an example. It can be understood that sports cameras have a miniaturized design and good wearability, and are widely used in dynamic sports scenes such as outdoor and extreme sports. The sports camera is equipped with sensors, including but not limited to gyroscopes, accelerometers and optical image stabilization modules, which can be used to collect motion data such as angular velocity, linear acceleration and anti-shake compensation displacement. At the same time, the sports camera has a certain data processing capability, and can perform local calculation and processing on the collected sensor data to support further expansion of support for functions such as motion state recognition and motion quantity statistics on the basis of image acquisition functions. As Figure 1 As shown, the method may include the following steps:

[0049] S101: Acquire sensor data of a motion camera, where the sensor data includes angular velocity, linear acceleration, and anti-shake compensation displacement.

[0050] Among them, sensor data refers to the original or pre-processed numerical data collected by the sensors configured in the sports camera, specifically including angular velocity, linear acceleration and anti-shake compensation displacement; among them, angular velocity represents the rate of change of the rotation angle of an object per unit time, usually collected by a three-axis gyroscope, and is used to describe the rotation of the sports camera around each coordinate axis; linear acceleration refers to the acceleration component of an object in the straight line direction, usually obtained by a three-axis accelerometer, and is used to reflect the translational motion characteristics of the camera; anti-shake compensation displacement refers to the displacement caused by the built-in optical image stabilization (OIS) module of the sports camera to adjust the image sensor or lens to stabilize the image. This displacement reflects an indirect expression of the actual motion amplitude of the device.

[0051] To achieve high-precision motion monitoring, the action camera's main control chip initiates sensor data acquisition after entering motion detection mode. The sensors include a three-axis gyroscope, a three-axis accelerometer, and a position compensation unit associated with the optical image stabilization (OIS) module. Each sensor connects to the main control chip via digital communication interfaces such as SPI or I²C. The main control chip issues a unified hardware trigger signal, enabling unified control of the data acquisition process. By configuring trigger interrupts or DMA (Direct Memory Access) mechanisms, efficient and low-power data acquisition is achieved without frequent CPU intervention.

[0052] To ensure the timeline alignment of sensor data, a microsecond-level synchronization mechanism based on hardware triggers can be employed. Specifically, the gyroscope and accelerometer sampling clocks are driven by a unified clock source, and a high-precision timestamp is added when acquiring data. The OIS module is synchronized with the video frame acquisition module via a shared clock reference, ensuring that the acquired data is consistent with the image time. At the software level, a time window can be set to align and match the sampling points, ensuring that multi-source data is time-calibrated within a range of ≤10ms. This time synchronization mechanism ensures the time consistency of data during subsequent attitude solution and displacement mapping.

[0053] Considering that gyroscopes and accelerometers are susceptible to high-frequency interference, bias drift, and mechanical vibration during actual data acquisition, a multi-stage noise suppression strategy can be integrated to enhance data reliability. First, an adaptive Kalman filter algorithm is used to optimize the gyroscope output in real time, dynamically adjusting the process noise covariance matrix to effectively reduce bias drift caused by temperature drift or hardware aging and suppress high-frequency interference noise, thereby improving the measured signal-to-noise ratio by over 30%. Second, a sliding window low-pass filter is applied to the acceleration data, with a cutoff frequency set to 5Hz, to isolate the stable gravity component, approximately 9.8m / s², thereby obtaining a purer dynamic acceleration signal.

[0054] Furthermore, filter parameters can be optionally adjusted based on the motion scenario. For example, the window length can be increased during intense motion to prevent false positives, while filter sensitivity can be increased during static or slow motion to enhance responsiveness to subtle movements. Valid filtered data is stored in a local buffer and chronologically available to the attitude calculation and motion recognition modules, ensuring a stable, continuous, and high-quality data chain across the entire system.

[0055] By acquiring angular velocity, linear acceleration, and anti-shake compensation displacement data in real time, it not only captures the device's rotational and translational motion, but also quantifies the relationship between OIS compensation behavior and the device's actual motion, providing a more accurate estimate of motion amplitude. This acquisition method avoids reliance on external sensors, enabling motion monitoring to be implemented entirely on existing hardware, thereby reducing system complexity and power consumption.

[0056] S102: Calculate the attitude angle of the motion camera, and use the attitude angle to convert the angular velocity and linear acceleration into the same coordinate system.

[0057] The attitude angle refers to the Euler angle parameter used to describe the attitude state of a motion camera in space. A common coordinate system, typically a geographic coordinate system or a camera coordinate system, is used to uniformly describe spatial motion. A unified coordinate system facilitates the integration of multi-source data and consistency analysis.

[0058] To accurately identify motion, a motion camera must first obtain its orientation in three-dimensional space through attitude angle calculation. This step uses the gyroscope's angular velocity data and the accelerometer's linear acceleration data as raw inputs and fuses them using an extended Kalman filter (EKF). Angular velocity is used to estimate high-frequency dynamic attitude changes, while acceleration is primarily used to correct attitude angle drift. This results in an attitude solution output with high short-term accuracy and long-term stability, updating the attitude angle in real time.

[0059] After obtaining the attitude angle, the angular velocity and linear acceleration are converted from the device coordinate system to a unified reference coordinate system. Specifically, the attitude angle is used to construct a rotation matrix or call a rotation quaternion, and the three-axis vectors are mapped to the target coordinate system through matrix transformation. This ensures the consistency of various data in spatial directions, allowing for comprehensive analysis of acceleration and angular velocity in the same reference system.

[0060] Furthermore, for high-dynamic motion or unstable shooting scenes, such as jumping, spinning, and skiing, the system automatically adjusts attitude calculation parameters and conversion frequency to adapt to dramatic environmental changes. For example, during large-angle rotations, the attitude update rate is increased to prevent the accumulation of coordinate system conversion errors; during low-dynamic conditions, the processing frequency is reduced to save energy. Furthermore, the coordinate system conversion results can be filtered once to further eliminate non-physical jumps caused by numerical noise, improving the stability and continuity of motion feature extraction.

[0061] It should be noted that the reason for first calculating the motion camera's attitude angle and then using the attitude angle to convert the angular velocity and linear acceleration to the same coordinate system is because the data collected by the gyroscope and accelerometer are based on the device's coordinate system, while the analysis of the physical characteristics of the motion state usually needs to be unified into a fixed reference system. By calculating the attitude angle, the device's orientation changes in space can be accurately obtained, and based on this, data coordinate conversion is performed to ensure consistency in the spatial direction of the various sensor data. This approach not only improves the accuracy and reliability of motion feature extraction, state recognition, and displacement estimation, but also effectively suppresses acceleration misjudgments and angular velocity drift caused by attitude changes, which is conducive to achieving highly robust and high-precision motion monitoring functions in complex motion scenarios.

[0062] S103: In the same coordinate system, a mapping relationship between the angular velocity and the linear acceleration and the anti-shake compensation displacement is constructed, where the mapping relationship is used to represent the corresponding relationship between the motion amplitude of the motion camera and the anti-shake compensation displacement.

[0063] The mapping relationship refers to the functional or approximate functional relationship between angular velocity, linear acceleration, and image stabilization compensation displacement established through an algorithmic model within a unified coordinate system. It describes the mathematical correspondence between the actual device motion amplitude and lens displacement. Motion amplitude represents the device's overall spatial or angular displacement per unit time and is used to measure the intensity or severity of motion.

[0064] After completing attitude angle calculation and coordinate system 1, the motion camera's processor converts the angular velocity and linear acceleration data to a georeferenced coordinate system, obtaining the attitude change rate and linear motion trend over time. To improve data quality, the converted signals are further filtered, denoised, and smoothed over time windows. For example, sliding window averaging and bilateral filtering are used to remove sharp noise, while gravity components are separated from the acceleration signal. The processed data is then normalized and aligned with a unified timestamp to ensure a one-to-one correspondence in the time dimension with the OIS-compensated displacement data.

[0065] To establish a quantitative correlation between device motion and compensation displacement, this step constructs a "physics-to-perception" mapping model based on regression learning or parameter fitting. Specifically, the processor selects features such as acceleration modulus, angular velocity modulus, and angular velocity rate of change as input dimensions, and uses the two-dimensional displacement feedback from the OIS system as the output label. Least squares fitting, polynomial regression, or lightweight neural network modeling are performed on the sampled data set to train a function mapping model for the specific device structure and installation parameters.

[0066] During motion monitoring, newly acquired angular velocity and linear acceleration are fed into the trained mapping model in real time to predict theoretical compensation displacement. This is then compared against the actual compensation provided by the OIS module. If the predicted result deviates significantly from the actual result, an online update mechanism is triggered to fine-tune some model parameters to adapt to changes in environmental vibration or user motion state. Furthermore, in extreme sports scenarios, higher-order derivative features or FFT frequency domain features can be added to improve the fitting capability for nonlinear motion, ensuring a good match even in complex dynamic scenes.

[0067] Understandably, the need to establish a mapping relationship between angular velocity and linear acceleration and image stabilization compensation displacement within the same coordinate system arises because the image stabilization compensation displacement is generated by the camera's internal system in real-time response to device motion, but it does not directly reflect the actual amplitude of spatial motion. By establishing this mapping relationship, a quantitative connection can be established between the raw physical quantities collected by the sensor and the image displacement perceived by OIS, enabling a high-precision estimation of the device's motion amplitude. This approach not only indirectly compensates for the lack of a dedicated displacement measurement module in action cameras without the need for additional hardware sensors, but also improves the robustness and accuracy of motion amplitude estimation, making motion state recognition more physically based.

[0068] S104: Extracting motion data features based on the angular velocity and the linear acceleration, and determining the target motion state of the motion camera according to the motion data features and the mapping relationship.

[0069] Motion data features are numerical indicators that characterize the device's motion state, derived through time-domain analysis, frequency-domain analysis, and multi-axis coupling analysis based on angular velocity and linear acceleration signals. The target motion state is the device's current motion type, such as stationary, walking, running, or cycling, determined through a combination of feature analysis and mapping relationships.

[0070] Motion cameras can divide continuously collected data into sliding windows of fixed length, typically 1 to 2 seconds, to ensure that the data within each window is representative and continuous. This then leads to the motion data feature extraction phase. Specifically, a multi-dimensional, cross-domain feature set is constructed based on the angular velocity and linear acceleration data within the sliding window.

[0071] In the time domain, the processor extracts the acceleration variance To assess the vibration intensity and instability of the current motion, for example, when the vertical Significantly greater than When the acceleration modulus is 0.01, it can be preliminarily judged as a high-intensity action such as running or jumping. In addition, the peak interval can be extracted to characterize the rhythmicity of the action, helping to distinguish between continuous action and intermittent movement; at the same time, the posture evolution trajectory can be obtained by integrating the angular velocity signal, which is used to identify large-scale rotation behaviors, such as ski turns or cycling changes of direction. In the frequency domain, the acceleration modulus is converted to the spectrum space using fast Fourier transform to extract the main frequency and its energy weight. In the running scenario, the main frequency mostly falls in the range of 2.5Hz to 4.5Hz, while the typical cadence characteristics of cycling are concentrated in the range of 0.5Hz to 1.5Hz. By setting the frequency band recognition window, the frequency characteristics of different motion states are mapped to the classification boundary to enhance the discrimination of state recognition. In addition, multi-axis coupling features are further extracted to evaluate motion coordination and symmetry. To this end, the covariance matrix between the three-axis acceleration is calculated to measure the joint distribution characteristics of the acceleration in all directions in space; and combined with the cross-correlation function between angular velocity and acceleration, the temporal coordination of inertial and force sensing signals is evaluated to detect whether there are typical motion coupling patterns, such as the "running double peak-arm swing" or "cycling cadence-pitch oscillation" correlation patterns.

[0072] Combining multi-dimensional motion data features, the action camera uses a built-in classification algorithm or model to analyze the feature vectors within the current time window, determine the device's target motion state, and identify different motion modes such as stillness, walking, running, and cycling. To ensure the physical accuracy of the determination, the action camera uses a mapping model to map the extracted motion data features to physical quantities such as anti-shake compensation displacement, achieving a corresponding correction between motion amplitude and features. The continuous determination results from multiple windows are time-series fused and filtered to filter out occasional anomalies and noise, outputting continuous, stable, and accurate motion status results to meet actual motion monitoring needs.

[0073] Extracting motion data features based on angular velocity and linear acceleration enables multi-angle and omnidirectional capture of the action camera's dynamic motion information, improving the accuracy and robustness of motion state recognition. The introduction of a mapping relationship ensures that motion features align with the actual physical motion amplitude, avoiding the bias introduced by simple signal analysis and enhancing the physical reliability of recognition. This helps action cameras accurately identify the user's motion state, meeting the needs of motion monitoring in multiple scenarios and enhancing user experience and application value.

[0074] S105: Based on the target motion state and in combination with the sensor data, the motion amount corresponding to the motion camera is counted.

[0075] Among them, exercise volume refers to the numerical results that reflect exercise intensity and exercise quantitative indicators, which may include number of steps, exercise distance, exercise duration, calorie consumption, etc. It is obtained by statistics and calculation of sensor data combined with the target exercise status, and can quantitatively describe the user's exercise behavior.

[0076] The action camera first categorizes and manages the collected sensor data based on the target motion state. Different motion states correspond to distinct motion characteristic patterns, and corresponding sensor data processing strategies and statistical algorithms are selected for each state. For example, in walking and running, analysis focuses on cadence, stride length, and peak acceleration; in cycling, analysis focuses on cadence and rotational motion data. This ensures that subsequent motion statistics are optimized for specific motion states, improving statistical accuracy and applicability.

[0077] Based on motion state classification, the action camera performs detailed calculations on various sensor data. For example, it identifies strides by combining periodic peaks in linear acceleration, estimating cadence and step count by calculating the intervals between peaks. It also uses image stabilization to compensate for displacement and calculates total exercise time by integrating exercise duration. It also combines posture angle changes and angular velocity integral trajectories to assist in calculating dynamic features during exercise, enhancing the richness and accuracy of exercise volume calculations. Furthermore, based on user-preset parameters such as weight and height, combined with exercise intensity, it estimates calorie consumption, achieving multi-dimensional quantification of exercise volume.

[0078] In one example, a sports camera can accurately calculate calorie consumption during exercise by integrating user-provided weight information and real-time metabolic equivalent (MET) values. This calorie calculation is based on a standard metabolic formula, integrating the user's weight and exercise duration. The specific formula is: ,in, It represents the metabolic equivalent at time t, Weight is the user's weight, dt is the time element of the integration, and the unit is seconds. Dividing by 3600 is to convert seconds to hours, so as to be consistent with the hourly energy expenditure in the MET definition. This formula reflects the basic principle that calorie consumption is proportional to the user's weight and has a cumulative relationship with exercise intensity and duration. It is worth mentioning that the MET(t) value is not a static input, but is estimated in real time based on the user's current exercise status and the dynamic characteristics collected by the sensor. Therefore, it can better fit the actual exercise situation and achieve dynamic and accurate tracking of calorie consumption.

[0079] To improve the stability and continuity of activity statistics, the action camera fuses activity parameters across multiple time windows, filtering out outliers and occasional noise to ensure the consistency and reliability of statistical data. Statistics can be exported to the user interface or third-party applications via local storage or cloud synchronization, providing users with intuitive, real-time activity feedback. Furthermore, the device integrates historical activity data to support activity trend analysis and personalized exercise recommendations, further enhancing the practical value of activity monitoring.

[0080] Motion statistics are performed based on the target motion state combined with sensor data, and appropriate calculation methods can be selected for different motion modes, avoiding the recognition bias of general algorithms in diverse motion environments, thereby significantly improving the accuracy and precision of motion statistics.

[0081] In the above-described embodiments, a lightweight solution is proposed that integrates the sensor data built into a motion camera to address the issues of hardware redundancy, data limitations, and poor scene adaptability in existing motion monitoring systems. This method utilizes the angular velocity, linear acceleration, and anti-shake compensation displacement data already available in the motion camera. Through attitude angle calculation, this multi-source data is integrated and processed in a unified coordinate system. This mapping relationship is then established between motion amplitude and anti-shake compensation displacement, addressing the inadequate perception of actual motion levels. Based on this mapping relationship and the results of motion data feature extraction, the current target motion state of the motion camera can be accurately identified. Furthermore, combined with sensor data, precise motion statistics can be achieved. This method fully exploits the hardware potential of existing motion cameras without adding additional sensors or hardware burden, improving the accuracy and applicability of motion monitoring. It offers advantages of good integration and low power consumption, and is conducive to promoting the functional expansion and widespread application of motion cameras in the field of motion monitoring.

[0082] In one embodiment, the step of calculating the attitude angle of the motion camera includes:

[0083] The complementary filtering algorithm is used to fuse the angular velocity and linear acceleration data to update the attitude angle of the motion camera.

[0084] Among them, the complementary filtering algorithm is a filtering technology based on the weighted fusion of gyroscope integration results and accelerometer gravity direction estimation. By adjusting the weights of different data sources to complement their advantages and disadvantages, it can achieve continuous and stable updates of the attitude angle.

[0085] The motion camera first collects the angular velocity data of the gyroscope and the linear acceleration data of the accelerometer in real time. Based on the angular velocity of the gyroscope, the attitude angle at the previous moment is calculated through the quaternion mathematical model. Perform integral update to obtain the initial attitude angle at the current moment Estimate. This integration process takes into account the sampling time interval , ensure the rotation angle Continuity and time synchronization. For example, the formula of quaternion can be .

[0086] Since gyroscope integration is prone to drift errors, sports cameras use accelerometer data to extract the gravity component as a posture reference under static or low-dynamic conditions. By comparing the posture obtained by gyroscope integration with the gravity direction measured by the accelerometer, the posture error vector is calculated. , thereby evaluating the cumulative drift of the gyroscope output. The complementary filter algorithm uses proportional-integral control to feedback correct the attitude error and calculate the corrected angular velocity value. This corrected angular velocity is combined with the original gyroscope data , dynamically adjust the attitude update process, retaining the high-frequency dynamic response of the gyroscope while eliminating the low-frequency drift with the help of the accelerometer. In the specific implementation, set the proportional gain and integral gain To balance the response speed and stability of the system, typical parameters such as =2.0, =0.002. The specific formula can be .

[0087] The quaternion integration of the corrected angular velocity using complementary filtering is then performed again to update the attitude angle in real time. This process is executed continuously, ensuring that the action camera can obtain stable and accurate attitude angles. This approach fully utilizes the gravity direction information measured by the accelerometer to continuously correct attitude estimation errors, ensuring the robustness and reliability of the attitude solution. Through this closed-loop mechanism of data fusion and error feedback, the action camera can accurately perceive its own spatial attitude and maintain stable attitude estimation results even in complex motion environments.

[0088] The complementary filtering algorithm is used to fuse the angular velocity and linear acceleration data, which can effectively compensate for the drift error in the gyroscope integration while retaining the advantages of the gyroscope in dynamic response, thereby achieving high precision and stability of the motion camera's attitude angle.

[0089] In one embodiment, the step of constructing a mapping relationship between the angular velocity and the linear acceleration and the anti-shake compensation displacement includes:

[0090] Acquire multiple sets of experimental data, each set of experimental data including experimental angular velocity, experimental linear acceleration, and experimental anti-shake compensation displacement;

[0091] Using each set of experimental data, the first calibration parameter of angular velocity and the second calibration parameter of linear acceleration are determined respectively through polynomial fitting;

[0092] A mapping relationship between the angular velocity and the linear acceleration and the anti-shake compensation displacement is established using the first calibration parameter and the second calibration parameter.

[0093] The experimental angular velocity refers to the angular velocity data measured by the motion camera's gyroscope sensor during multiple experiments, reflecting the camera's rotation rate around a spatial axis. The experimental linear acceleration is the acceleration measurement of the camera along each axis, collected by the accelerometer, and includes motion and gravitational acceleration components. The experimental anti-shake compensation displacement is the displacement vector calculated during the experiment based on the anti-shake algorithm to offset the jitter caused by camera motion and adjust the position of the image frame to achieve image stability. Polynomial fitting is a mathematical method that uses polynomial functions to curve fit discrete experimental data and is used to describe nonlinear relationships between variables. The first calibration parameter and the second calibration parameter are sets of coefficients determined by the polynomial fitting algorithm for angular velocity and linear acceleration, respectively, and are used to accurately characterize the mapping between sensor data and anti-shake compensation displacement.

[0094] The action camera uses its integrated gyroscope and accelerometer to collect multiple sets of experimental data. Each set includes the corresponding angular velocity, linear acceleration, and compensated displacement calculated by the anti-shake algorithm. During acquisition, the action camera can be placed in different motion states to cover a wide range of motion, thereby obtaining rich and diverse experimental samples. During data acquisition, the sensors must synchronize sampling and timestamp recording to ensure the temporal consistency of the angular velocity, acceleration, and compensated displacement data.

[0095] For each set of collected experimental data, a polynomial fitting algorithm is used to process the data pairs of angular velocity and image stabilization displacement, as well as the data pairs of linear acceleration and image stabilization displacement. During the fitting process, an appropriate polynomial order is selected, and the fitting curve is optimized using the least squares method to obtain the first calibration parameters corresponding to angular velocity and the second calibration parameters corresponding to linear acceleration. This calibration process can be performed during device initialization or regular calibration to adapt to changes in sensor characteristics or environmental factors.

[0096] Based on the first calibration parameter and the second calibration parameter, the motion camera constructs a mapping relationship model from angular velocity and linear acceleration to anti-shake compensation displacement. The model converts the input real-time sensor angular velocity and linear acceleration into the corresponding anti-shake compensation displacement in the form of a polynomial function. The mapping relationship can be called in real time in the anti-shake control link to assist in the dynamic position adjustment of the image frame and improve the picture stabilization effect. This mapping model not only improves the anti-shake accuracy, but also reduces the computational complexity, enabling the motion camera to efficiently perform anti-shake compensation. For example, the mapping relationship can be expressed as ,in, Indicates the anti-shake compensation displacement in the x-axis direction, represents the linear acceleration in the x-axis direction, Indicates the angular velocity along the z-axis.

[0097] By acquiring multiple sets of experimental data and employing polynomial fitting to determine calibration parameters for angular velocity and linear acceleration, the team was able to scientifically characterize the relationship between sensor data and image stabilization compensation displacement, achieving precise mapping. This mapping relationship enables action cameras to accurately calculate image stabilization compensation displacement based on real-time angular velocity and linear acceleration, improving the accuracy and responsiveness of image stabilization. Furthermore, the continuous function model provided by polynomial fitting simplifies the calculation process, ensuring efficient and stable image stabilization operation.

[0098] In one embodiment, the step of determining the target motion state of the motion camera according to the motion data characteristics and the mapping relationship includes:

[0099] Determine the initial motion state of the motion camera according to the motion data characteristics;

[0100] When the verification of the initial motion state based on the mapping relationship is passed, the target motion state of the motion camera is obtained.

[0101] The initial motion state refers to the current motion category or state of the action camera, such as still, walking, running, etc., preliminarily inferred based on the motion data features before further confirmation or correction.

[0102] Based on the extracted motion data features, the action camera uses a set threshold or machine learning classification model to preliminarily determine the device's current motion type and obtain the initial motion state. For example, if the vertical acceleration variance exceeds a preset threshold and the main frequency is within the running range, it is initially identified as running. This preliminary judgment provides a candidate result for subsequent more accurate state confirmation.

[0103] To improve the reliability of judgment, the motion camera uses a pre-established mapping model to compare and verify the motion data features corresponding to the initial motion state with the feature patterns in the mapping model. This mapping model, based on extensive experimental data and derived through multi-dimensional feature space fitting, effectively verifies and corrects the initial state. The verification process includes calculating feature similarity, determining error ranges, and triggering multiple rounds of correction mechanisms when necessary to ensure that the judgment result is not affected by sensor noise or transient anomalies.

[0104] Once the mapping relationship is verified, the action camera determines the target motion state and uses this state as the output of the motion recognition module to drive subsequent actions such as adjusting the anti-shake compensation strategy, optimizing motion trajectories, and analyzing user motion behavior. This target state is both highly accurate and meets real-time requirements, ensuring stable performance in diverse and complex motion environments.

[0105] Initially determining the action camera's motion state based on motion data features allows for rapid response to current motion changes, providing a foundation for immediate motion recognition. Verifying the initial motion state through mapping relationships effectively eliminates misjudgments caused by sensor errors and environmental interference, improving the accuracy and robustness of state recognition. This dual determination mechanism achieves dynamic adaptability and high reliability in motion state recognition, ensuring the action camera accurately locates its motion state in diverse motion scenarios.

[0106] In one embodiment, the step of determining the initial motion state of the motion camera according to the motion data characteristics includes:

[0107] Determine a first motion state classification result of the motion camera according to a preset motion data threshold condition for each motion state and motion data characteristics;

[0108] Input the motion data features into a preset machine learning model to obtain a second motion state classification result of the motion camera;

[0109] The first motion state classification result and the second motion state classification result are weightedly fused to obtain the initial motion state of the motion camera.

[0110] Among them, the motion data threshold condition refers to the preset quantitative standard or judgment range for different motion states, which is usually based on the statistical characteristics of sensor data such as acceleration and angular velocity, such as acceleration variance, angular velocity standard deviation and frequency range, to roughly divide the rules of motion types. The first motion state classification result refers to the preliminary motion category judgment obtained by rule judgment on real-time motion data features based on the preset motion data threshold condition. The preset machine learning model refers to a pre-trained neural network model based on multi-dimensional motion data features, which is used for deep learning classification tasks and can recognize complex motion patterns. The second motion state classification result refers to the highest probability motion category judgment under the probability distribution output by the model after the motion data features are input into the machine learning model. Weighted fusion refers to the weighted average of the confidence probabilities of the first motion state classification result and the second motion state classification result according to predefined weight parameters, so as to obtain a comprehensive judgment of the motion state.

[0111] The motion camera receives sensor data collected by the gyroscope and accelerometer in real time. After filtering and preprocessing, it calculates multiple motion data features, such as acceleration standard deviation, angular velocity variance, main frequency energy, etc. These features are judged based on the pre-set threshold condition library. For example, the judgment condition for the static state is the standard deviation of acceleration. Less than 0.1g, and the standard deviation of angular velocity Less than 5° per second; when the vertical axis acceleration shows periodic fluctuations, and the peak interval is between 0.4 and 0.7 seconds, and the main frequency is between 1.5 and 2.2 Hz, it can be judged as a walking state; the running state is characterized by a main frequency higher than 2.5 Hz, the amplitude of the three-axis acceleration exceeds 2g, and the standard deviation of the angular velocity is greater than 10° per second; the cycling state is characterized by periodic fluctuations in the horizontal axis acceleration, a cadence range of 0.5 to 1.5 Hz, and a standard deviation of the vertical axis acceleration. The jumping state is characterized by a short period of vertical acceleration peak exceeding 3g, and a sharp fluctuation in angular velocity, with a standard deviation of Greater than 20° per second.

[0112] In parallel with the threshold determination, the action camera feeds the same batch of motion data features into an embedded machine learning model. This model can be a pre-trained one-dimensional convolutional neural network (1D-CNN). Through convolutional and fully connected layers, it deeply extracts and learns comprehensive features from the time and frequency domains, generating a probability distribution encompassing multiple motion states. Trained on large amounts of diverse motion data, the model accurately distinguishes complex and mixed motion scenes, and its output second-motion state classification results demonstrate high generalization and robustness.

[0113] Then, the weight parameters can be dynamically adjusted according to environmental variables such as light intensity, sensor noise level, etc. and , respectively assign threshold classification and machine learning models The results of weighted fusion are different. Weighted fusion calculates the linear combination of the two classification probabilities using the formula The optimal motion state after fusion is obtained. This fusion strategy overcomes the limitations of a single judgment method, taking into account the real-time nature of threshold judgment and the accuracy of the machine learning model, ensuring that a stable and accurate initial motion state can be output in a variety of complex environments.

[0114] Using preset motion data thresholds for initial classification enables fast, low-computational cost, and real-time motion state determination, ideal for real-time response requirements in resource-constrained devices. Inputting motion data features into a machine learning model for classification leverages the strengths of deep learning to capture complex, nonlinear motion patterns, improving classification accuracy and environmental adaptability. By weightedly fusing the two classification results, the advantages of both are leveraged, overcoming potential misjudgments or delays associated with each approach alone, resulting in more stable and reliable initial motion state recognition. This fusion strategy effectively improves the recognition accuracy and robustness of action cameras under various complex motion and environmental conditions.

[0115] In one embodiment, the amount of movement includes the number of steps, and the steps of counting the amount of movement corresponding to the motion camera based on the target motion state and in combination with sensor data include:

[0116] If the target motion state is a state where step counting is possible, then within a preset time window with the current step counting time as the end time, the mean and standard deviation of the linear acceleration are calculated, and the current step counting threshold is determined based on the mean and standard deviation;

[0117] If the current linear acceleration exceeds the current step counting threshold, when the current linear acceleration is at its peak value and the time interval between the current step counting time point and the previous step counting time point is greater than the preset time difference threshold, the step count corresponding to the motion camera is increased by one.

[0118] The preset time window is a fixed-length time interval with the current step counting time as the end point, which is used to count and analyze acceleration data. The mean and standard deviation of linear acceleration refer to the statistical characteristics of the acceleration data collected within this time window. The mean reflects the overall acceleration level, and the standard deviation reflects the amplitude of the acceleration change. The current step counting threshold is a judgment threshold dynamically calculated based on the mean and standard deviation, which is used to determine whether the acceleration meets the standard for valid step counting. The peak value refers to the local maximum value of the current linear acceleration value, that is, the acceleration at this point is greater than the acceleration at its adjacent time points. The preset time difference threshold is to limit the minimum interval between two consecutive step counting times to avoid repeated step counting. The number of steps represents the cumulative number of valid steps detected by the motion camera.

[0119] First, the action camera collects linear acceleration data in real time and, based on the current target motion state, determines whether a step can be counted, such as walking or running. If a step is counted, a preset time window (for example, 2 or 3 seconds) is defined, with the current step count time as the end point, to collect linear acceleration data within that time period. Within this time window, the mean and standard deviation of the linear acceleration data are calculated, and the current step counting threshold is dynamically generated based on this data. This ensures that the threshold can adapt to data fluctuations caused by varying motion intensities and environmental changes, maintaining a balance between sensitivity and accuracy.

[0120] Subsequently, peak detection is performed on the real-time linear acceleration signal. If the acceleration at a particular moment exceeds the current step counting threshold and is a local peak value—that is, significantly higher than the values before and after—the system further determines whether the time interval between the current step count and the previous step count is greater than a preset time difference threshold. This time difference threshold is set to prevent duplicate step counting due to frequent acceleration fluctuations, thereby improving the stability and accuracy of the step counting results.

[0121] If all of the above conditions are met, the action camera increments the step counter by one, recording a valid step. This process continues continuously, ensuring the real-time and accurate counting of steps. Simultaneously, the step counting threshold is continuously updated within the sliding time window to adapt to changes in exercise intensity.

[0122] In one example, the detection of step count is mainly based on the periodic changes of acceleration data collected by the accelerometer, especially the acceleration in the vertical direction. In order to accurately capture the step signal, a dynamic threshold method is used for peak detection. Specifically, set The vertical linear acceleration value at time t, which has removed the effect of gravity. Output by the machine learning model to indicate the current motion category, for example, 0 for stillness, 1 for walking, 2 for running, 3 for cycling, 4 for jumping, etc. In a time interval with a sliding window size of W (for example, 2 seconds) Inside, right Perform statistics and calculate the mean acceleration within the time window and standard deviation Then, according to the current motion state Corresponding adjustment coefficient , dynamically calculate step counting threshold , the formula is ,This threshold can adapt to the changing characteristics of acceleration under different ,motion states, ensuring the sensitivity and robustness of detection.

[0123] When at time t, the vertical acceleration Exceeding dynamic threshold , and the current motion state When it is in the preset step counting state, further judgment is made Whether the peak condition is met, that is, the acceleration at this moment is a local extreme value, which is usually manifested as the acceleration derivative changing from positive to negative. In addition, to prevent miscounting caused by acceleration fluctuations, the time interval between the current step counting time point and the last valid step time point is required to be greater than the minimum step interval threshold corresponding to the state. When all of the above conditions are met, the action camera's step counter increments by one step. Otherwise, the step increment is zero. This method, by combining dynamic thresholding and peak detection, effectively improves the accuracy of step recognition and adapts to the needs of step counting in diverse motion states.

[0124] Dynamically calculating the mean and standard deviation of linear acceleration and adjusting the step counting threshold accordingly allows the system to adapt to fluctuations in acceleration data under varying motion states and environmental conditions, thereby improving the sensitivity and accuracy of step detection. Peak detection combined with threshold determination effectively filters out acceleration interference caused by non-steps, preventing false positives. Setting a minimum threshold for the interval between consecutive step counts further prevents duplicate step counting caused by high-frequency fluctuations in acceleration data, ensuring the stability and continuity of step counting. This effectively improves the accuracy and robustness of step counting in dynamic and complex environments, ensuring the reliability and practicality of step counting results.

[0125] In one embodiment, the amount of motion includes motion intensity. The step of calculating the amount of motion corresponding to the motion camera based on the target motion state and in combination with sensor data includes:

[0126] According to the target exercise state, the current metabolic equivalent is determined and normalized to obtain the metabolic intensity;

[0127] Calculate mechanical strength based on sensor data within a preset sliding window;

[0128] After determining the initial cadence and current cadence based on the target motion state and sensor data, the fatigue level is calculated based on the initial cadence and current cadence;

[0129] The metabolic intensity, mechanical strength and fatigue degree are weighted and integrated to obtain the exercise intensity.

[0130] Among them, the metabolic equivalent of task (MET) is an indicator that measures the intensity of energy consumption during exercise. It is assigned values based on different preset values for different exercise states and reflects the metabolic load of exercise. Mechanical intensity is based on the acceleration and angular velocity data collected by the motion camera sensor, reflecting the mechanical impact and amount of exercise during exercise. Cadence refers to the number of steps completed by the athlete per unit time. It includes two concepts: initial cadence and current cadence, representing the cadence at the beginning of exercise and the current moment, respectively, and is used to assess changes in exercise rhythm. The motion camera forms a comprehensive exercise intensity index by weighted fusion of metabolic intensity, mechanical intensity, and fatigue, reflecting the athlete's overall exercise load and state.

[0131] First, the action camera retrieves the corresponding metabolic equivalent value from a pre-set table of motion states and metabolic equivalents based on the target's real-time motion state. This value is then normalized by dividing the metabolic equivalent by a fixed standard value, such as 10, to ensure that the metabolic intensity remains within the range of 0 to 1, making it easier to unify the scale with other intensity indicators and facilitate integrated calculations.

[0132] Next, the motion camera calculates mechanical strength using data from its built-in inertial measurement unit (IMU) sensor within a preset sliding time window, such as the past three seconds. This method collects triaxial acceleration data within this time period. The effects of gravity are removed from this data, and the absolute acceleration values are integrated or averaged to produce a value representing the mechanical load of the motion. This value is then adjusted using a scaling factor to ensure that the mechanical strength range is suitable for subsequent fusion.

[0133] The action camera then determines the initial and current cadences based on current sensor data and the target's motion state. Initial cadence is typically determined through short-term data analysis at the beginning of exercise and represents the baseline rhythm of the movement. Current cadence is updated in real time through continuous monitoring of the gait cycle. Using these two parameters, fatigue is calculated. Specifically, fatigue is measured by the percentage of cadence decrease, which reflects the athlete's fatigue level and decreased athletic ability.

[0134] Finally, the action camera performs a weighted fusion of metabolic intensity, mechanical intensity, and fatigue. The weighting can be dynamically adjusted based on different exercise environments, user needs, or sensor accuracy to generate a comprehensive exercise intensity index. This index comprehensively reflects the athlete's metabolic load, mechanical load, and fatigue status.

[0135] In one example, a sports camera can construct a comprehensive exercise intensity index through real-time analysis of multi-dimensional physiological and exercise data to accurately assess the user's current exercise load level. The exercise intensity index is scored on a scale of 0 to 10 and integrates three core components, including metabolic intensity, mechanical strength, and fatigue. First, the metabolic intensity index is calculated based on metabolic equivalents (METs). The real-time estimated MET value is divided by 10 for normalization, i.e. This approach is based on the fact that the upper limit of METs is typically no more than 20. Normalized values typically fall between 0 and 2, and can be further adjusted using weighting factors. If your sports camera has heart rate monitoring, you can also use percentage of heart rate reserve (%HRR) as an alternative to metabolic intensity to improve individualized accuracy.

[0136] Secondly, the mechanical strength is calculated by integrating the acceleration signal. The motion camera collects the total acceleration after removing gravity in a sliding time window. , and calculate the time integral of its absolute value. The specific formula is ,in is the length of the integration time window, which can be 3 seconds, A scaling factor, such as 0.1, is used to ensure that the resulting value remains within a reasonable range, such as 0 to 1. This indicator reflects the impact and physical output during the current movement.

[0137] Fatigue is estimated based on changes in cadence. Comparing the current cadence with the initial cadence can indicate whether the user is experiencing fatigue due to continued exercise. Calculation methods include (current cadence / initial cadence) or a more attenuated form [1 - (current cadence / maximum cadence)]. Furthermore, action cameras can be combined with machine learning models to further improve fatigue detection accuracy by analyzing gait characteristics over time, such as periodic variation and reduced acceleration amplitude.

[0138] Finally, the motion camera normalizes these three components according to the preset weights, such as metabolic intensity. Proportion = 0.5, mechanical strength Proportion = 0.3, fatigue Proportion = 0.2, and then multiplied by a scaling factor, such as 5, to form the final exercise intensity index. :

[0139]

[0140] This index can reflect an individual's comprehensive exercise status in real time, provide quantitative support for exercise monitoring, health warning, training rhythm control, etc., and enhance the use value of exercise data.

[0141] In this embodiment, by first determining the metabolic equivalent according to the target motion state and normalizing it to obtain the metabolic intensity, the energy consumption level of the athlete can be accurately reflected, ensuring that the quantification of the metabolic load is scientific and reasonable; calculating the mechanical intensity within the preset sliding window effectively captures the mechanical impact and amount of exercise during the exercise process, supplementing the deficiencies of metabolic indicators; calculating fatigue based on the initial cadence and current cadence helps to dynamically monitor the athlete's fatigue level and promptly reflect changes in exercise capacity. Finally, a weighted fusion of metabolic intensity, mechanical intensity, and fatigue is performed to comprehensively reflect the multi-dimensional characteristics of the motion state, thereby improving the accuracy and robustness of the exercise intensity assessment. This method not only realizes the coordinated use of multi-source information, enhances the adaptability of the motion camera to complex motion environments, but also improves the accuracy of motion intensity assessment.

[0142] The following describes the motion monitoring device provided by the embodiment of the present application. The motion monitoring device described below and the motion monitoring method described above can be referred to in correspondence with each other. Figure 2 As shown, the present application provides a motion monitoring device, which is applied to a motion camera. The motion camera is equipped with a sensor, and the device includes:

[0143] The sensor data acquisition module 201 is used to acquire sensor data of the motion camera, where the sensor data includes angular velocity, linear acceleration, and anti-shake compensation displacement;

[0144] An attitude angle calculation module 202 is used to calculate the attitude angle of the motion camera and use the attitude angle to convert the angular velocity and linear acceleration into the same coordinate system;

[0145] A mapping relationship establishment module 203 is used to establish a mapping relationship between the angular velocity and the linear acceleration and the anti-shake compensation displacement in the same coordinate system, wherein the mapping relationship is used to represent the corresponding relationship between the motion amplitude of the motion camera and the anti-shake compensation displacement;

[0146] The target motion state determination module 204 is used to extract motion data features based on angular velocity and linear acceleration, and determine the target motion state of the motion camera according to the motion data features and the mapping relationship;

[0147] The motion amount statistics module 205 is used to count the motion amount corresponding to the motion camera based on the target motion state and in combination with the sensor data.

[0148] In one embodiment, the attitude angle calculation module 202 includes:

[0149] The attitude angle calculation unit is used to fuse the angular velocity and linear acceleration data using a complementary filtering algorithm to update the attitude angle of the motion camera.

[0150] In one embodiment, the mapping relationship establishing module 203 includes:

[0151] An experimental data acquisition unit, used to acquire multiple sets of experimental data, each set of experimental data including experimental angular velocity, experimental linear acceleration and experimental anti-shake compensation displacement;

[0152] a calibration parameter determination unit, configured to determine a first calibration parameter of angular velocity and a second calibration parameter of linear acceleration by polynomial fitting using each set of experimental data;

[0153] The mapping relationship establishing unit is used to establish a mapping relationship between angular velocity and linear acceleration and anti-shake compensation displacement by using the first calibration parameter and the second calibration parameter.

[0154] In one embodiment, the target motion state determination module 204 includes:

[0155] an initial motion state determining unit, configured to determine an initial motion state of the motion camera according to motion data characteristics;

[0156] The target motion state determining unit is configured to obtain the target motion state of the motion camera when the verification of the initial motion state based on the mapping relationship is passed.

[0157] In one embodiment, the initial motion state determining unit includes:

[0158] A first motion state classification result determining subunit, configured to determine a first motion state classification result of the motion camera according to a preset motion data threshold condition and motion data features for each motion state;

[0159] A second motion state classification result determination subunit is configured to input motion data features into a preset machine learning model to obtain a second motion state classification result of the motion camera;

[0160] The initial motion state determination subunit is used to perform weighted fusion on the first motion state classification result and the second motion state classification result to obtain the initial motion state of the motion camera.

[0161] In one embodiment, the amount of exercise includes the number of steps, and the exercise amount statistics module 205 includes:

[0162] a current step counting threshold determination unit, configured to, if the target motion state is a step counting state, calculate the mean and standard deviation of the linear acceleration within a preset time window ending at the current step counting time, and determine the current step counting threshold based on the mean and standard deviation;

[0163] The first motion amount statistics unit is used to increase the number of steps corresponding to the motion camera by one if the current linear acceleration exceeds the current step counting threshold, when the current linear acceleration is at a peak value and the time interval between the current step counting time point and the previous step counting time point is greater than a preset time difference threshold.

[0164] In one embodiment, the amount of exercise includes exercise intensity, and the exercise amount statistics module 205 includes:

[0165] A metabolic intensity determination unit is used to determine the current metabolic equivalent according to the target exercise state, and normalize the current metabolic equivalent to obtain the metabolic intensity;

[0166] a mechanical strength determination unit, configured to calculate the mechanical strength based on sensor data within a preset sliding window;

[0167] a fatigue determination unit, configured to determine an initial cadence and a current cadence according to a target motion state and sensor data, and then calculate fatigue according to the initial cadence and the current cadence;

[0168] The second exercise quantity statistical unit is used to perform weighted fusion of metabolic intensity, mechanical intensity and fatigue to obtain exercise intensity.

[0169] In one embodiment, the present application also provides a storage medium storing computer-readable instructions. When the computer-readable instructions are executed by one or more processors, the one or more processors execute the steps of the motion monitoring method as described in any of the above embodiments.

[0170] In one embodiment, the present application further provides a motion camera having computer-readable instructions stored therein. When the computer-readable instructions are executed by one or more processors, the one or more processors execute the steps of the motion monitoring method as described in any one of the above embodiments.

[0171] Schematically, as Figure 3 As shown, Figure 3 This is a schematic diagram of the internal structure of a sports camera provided by an embodiment of the present application. Figure 3 The motion camera 300 includes a processing component 302, which further includes one or more processors, and a memory resource represented by a memory 301 for storing instructions executable by the processing component 302, such as an application. The application stored in the memory 301 may include one or more modules, each corresponding to a set of instructions. In addition, the processing component 302 is configured to execute the instructions to perform the motion monitoring method of any of the above-described embodiments.

[0172] The action camera 300 may further include a power supply component 303 configured to perform power management of the action camera 300, a wired or wireless network interface 304 configured to connect the action camera 300 to a network, and an input / output (I / O) interface 305. The action camera 300 may operate based on an operating system stored in the memory 301, such as Windows Server™, Mac OS X™, Unix™, Linux™, Free BSD™, or the like.

[0173] Those skilled in the art will understand that Figure 3 The structure shown in the figure is only a block diagram of a part of the structure related to the solution of the present application, and does not constitute a limitation on the sports camera to which the solution of the present application is applied. A specific sports camera may include more or fewer components than shown in the figure, or combine certain components, or have a different component arrangement.

[0174] Finally, it should be noted that, in this article, relational terms such as first and second are merely used to distinguish one entity or operation from another, and do not necessarily require or imply any actual relationship or order between these entities or operations. Moreover, the terms "comprise," "include," or any other variations thereof are intended to encompass non-exclusive inclusion, such that a process, method, article, or device comprising a series of elements includes not only those elements, but also other elements not explicitly listed, or elements inherent to such process, method, article, or device. Without further restriction, an element defined by the phrase "comprising a..." does not exclude the presence of other identical elements in the process, method, article, or device comprising the element. Herein, "one," "said," "the," and "its" may also include plural forms unless the context clearly indicates otherwise. A plurality refers to at least two, such as 2, 3, 5, or 8. "And / or" includes any and all combinations of the relevant listed items.

[0175] The various embodiments in this specification are described in a progressive manner, and each embodiment focuses on the differences from other embodiments. The various embodiments can be combined as needed, and the same or similar parts can be referenced to each other.

[0176] The above description of the disclosed embodiments is intended to enable one skilled in the art to implement or use the present application. Various modifications to these embodiments will be readily apparent to one skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the present application. Therefore, the present application is not limited to the embodiments shown herein, but is intended to conform to the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A motion monitoring method, characterized in that: Applied to a motion camera, the motion camera being equipped with a sensor, the method comprising: Acquiring sensor data of the motion camera, wherein the sensor data includes angular velocity, linear acceleration, and anti-shake compensation displacement; Calculating the attitude angle of the motion camera, and using the attitude angle to convert the angular velocity and the linear acceleration into the same coordinate system; In the same coordinate system, constructing a mapping relationship between the angular velocity and the linear acceleration and the anti-shake compensation displacement, wherein the mapping relationship is used to represent a corresponding relationship between the motion amplitude of the motion camera and the anti-shake compensation displacement; extracting motion data features based on the angular velocity and the linear acceleration, and determining a target motion state of the motion camera according to the motion data features and the mapping relationship; Based on the target motion state and in combination with the sensor data, the motion amount corresponding to the motion camera is counted.

2. The motion monitoring method according to claim 1, characterized in that: The step of calculating the attitude angle of the motion camera includes: A complementary filtering algorithm is used to perform data fusion on the angular velocity and the linear acceleration to update the attitude angle of the motion camera.

3. The motion monitoring method according to claim 1, wherein: The step of constructing a mapping relationship between the angular velocity and the linear acceleration and the anti-shake compensation displacement includes: Acquire multiple sets of experimental data, each set of experimental data including experimental angular velocity, experimental linear acceleration, and experimental anti-shake compensation displacement; Using each set of the experimental data, and through polynomial fitting, respectively determining a first calibration parameter of the angular velocity and a second calibration parameter of the linear acceleration; A mapping relationship between the angular velocity and the linear acceleration and the anti-shake compensation displacement is established by using the first calibration parameter and the second calibration parameter.

4. The motion monitoring method according to claim 1, wherein: The step of determining the target motion state of the motion camera according to the motion data feature and the mapping relationship includes: determining an initial motion state of the motion camera according to the motion data characteristics; When the initial motion state is verified based on the mapping relationship, the target motion state of the motion camera is obtained.

5. The motion monitoring method according to claim 4, characterized in that: The step of determining the initial motion state of the motion camera according to the motion data characteristics includes: Determining a first motion state classification result of the motion camera according to a preset motion data threshold condition for each motion state and the motion data feature; Inputting the motion data features into a preset machine learning model to obtain a second motion state classification result of the motion camera; The first motion state classification result and the second motion state classification result are weightedly fused to obtain an initial motion state of the motion camera.

6. The motion monitoring method according to claim 1, characterized in that: The amount of movement includes the number of steps, and the step of counting the amount of movement corresponding to the motion camera based on the target motion state and in combination with the sensor data includes: If the target motion state is a state where step counting is possible, then within a preset time window with the current step counting time as the end time, the mean and standard deviation of the linear acceleration are counted, and a current step counting threshold is determined based on the mean and standard deviation; If the current linear acceleration exceeds the current step counting threshold, when the current linear acceleration is at a peak value and the time interval between the current step counting time point and the previous step counting time point is greater than a preset time difference threshold, the number of steps corresponding to the motion camera is increased by one.

7. The motion monitoring method according to claim 1, characterized in that: The amount of motion includes motion intensity. The step of calculating the amount of motion corresponding to the motion camera based on the target motion state and in combination with the sensor data includes: Determining a current metabolic equivalent according to the target exercise state, and normalizing the current metabolic equivalent to obtain a metabolic intensity; Calculating mechanical strength based on the sensor data within a preset sliding window; After determining an initial cadence and a current cadence according to the target motion state and the sensor data, calculating fatigue according to the initial cadence and the current cadence; The metabolic intensity, the mechanical intensity and the fatigue degree are weighted and integrated to obtain the exercise intensity.

8. A motion monitoring device, characterized in that: Applied to a motion camera, the motion camera is equipped with a sensor, and the device includes: A sensor data acquisition module, configured to acquire sensor data of the motion camera, wherein the sensor data includes angular velocity, linear acceleration, and anti-shake compensation displacement; An attitude angle calculation module is used to calculate the attitude angle of the motion camera and convert the angular velocity and the linear acceleration into the same coordinate system using the attitude angle; a mapping relationship establishing module, configured to establish, in the same coordinate system, a mapping relationship between the angular velocity and the linear acceleration and the anti-shake compensation displacement, wherein the mapping relationship is used to represent a corresponding relationship between the motion amplitude of the motion camera and the anti-shake compensation displacement; a target motion state determination module, configured to extract motion data features based on the angular velocity and the linear acceleration, and determine the target motion state of the motion camera according to the motion data features and the mapping relationship; The motion amount statistics module is used to count the motion amount corresponding to the motion camera based on the target motion state and in combination with the sensor data.

9. A storage medium, characterized in that: The storage medium stores computer-readable instructions, and when the computer-readable instructions are executed by one or more processors, the one or more processors execute the steps of the motion monitoring method according to any one of claims 1 to 7.

10. A sports camera, characterized in that: include: one or more processors, and memory; The memory stores computer-readable instructions, which, when executed by the one or more processors, perform the steps of the motion monitoring method according to any one of claims 1 to 7.

Images