WO2018040757A1 - 可穿戴设备及利用其监测运动状态的方法 - Google Patents

可穿戴设备及利用其监测运动状态的方法 Download PDF

Info

Publication number
WO2018040757A1
WO2018040757A1 PCT/CN2017/092739 CN2017092739W WO2018040757A1 WO 2018040757 A1 WO2018040757 A1 WO 2018040757A1 CN 2017092739 W CN2017092739 W CN 2017092739W WO 2018040757 A1 WO2018040757 A1 WO 2018040757A1
Authority
WO
WIPO (PCT)
Prior art keywords
data
state
swimming
user
motion
Prior art date
Application number
PCT/CN2017/092739
Other languages
English (en)
French (fr)
Inventor
苏鹏程
张一凡
Original Assignee
歌尔股份有限公司
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by 歌尔股份有限公司 filed Critical 歌尔股份有限公司
Priority to US16/327,173 priority Critical patent/US20190209050A1/en
Publication of WO2018040757A1 publication Critical patent/WO2018040757A1/zh

Links

Images

Classifications

    • A—HUMAN NECESSITIES
    • A61—MEDICAL OR VETERINARY SCIENCE; HYGIENE
    • A61B—DIAGNOSIS; SURGERY; IDENTIFICATION
    • A61B5/00—Measuring for diagnostic purposes; Identification of persons
    • A61B5/103—Detecting, measuring or recording devices for testing the shape, pattern, colour, size or movement of the body or parts thereof, for diagnostic purposes
    • A61B5/11—Measuring movement of the entire body or parts thereof, e.g. head or hand tremor, mobility of a limb
    • A61B5/1123—Discriminating type of movement, e.g. walking or running
    • A—HUMAN NECESSITIES
    • A63—SPORTS; GAMES; AMUSEMENTS
    • A63B—APPARATUS FOR PHYSICAL TRAINING, GYMNASTICS, SWIMMING, CLIMBING, OR FENCING; BALL GAMES; TRAINING EQUIPMENT
    • A63B71/00—Games or sports accessories not covered in groups A63B1/00 - A63B69/00
    • A63B71/06—Indicating or scoring devices for games or players, or for other sports activities
    • A63B71/0619—Displays, user interfaces and indicating devices, specially adapted for sport equipment, e.g. display mounted on treadmills
    • A—HUMAN NECESSITIES
    • A61—MEDICAL OR VETERINARY SCIENCE; HYGIENE
    • A61B—DIAGNOSIS; SURGERY; IDENTIFICATION
    • A61B5/00—Measuring for diagnostic purposes; Identification of persons
    • A61B5/103—Detecting, measuring or recording devices for testing the shape, pattern, colour, size or movement of the body or parts thereof, for diagnostic purposes
    • A61B5/11—Measuring movement of the entire body or parts thereof, e.g. head or hand tremor, mobility of a limb
    • A61B5/1116—Determining posture transitions
    • A—HUMAN NECESSITIES
    • A61—MEDICAL OR VETERINARY SCIENCE; HYGIENE
    • A61B—DIAGNOSIS; SURGERY; IDENTIFICATION
    • A61B5/00—Measuring for diagnostic purposes; Identification of persons
    • A61B5/103—Detecting, measuring or recording devices for testing the shape, pattern, colour, size or movement of the body or parts thereof, for diagnostic purposes
    • A61B5/11—Measuring movement of the entire body or parts thereof, e.g. head or hand tremor, mobility of a limb
    • A61B5/1121—Determining geometric values, e.g. centre of rotation or angular range of movement
    • A—HUMAN NECESSITIES
    • A61—MEDICAL OR VETERINARY SCIENCE; HYGIENE
    • A61B—DIAGNOSIS; SURGERY; IDENTIFICATION
    • A61B5/00—Measuring for diagnostic purposes; Identification of persons
    • A61B5/68—Arrangements of detecting, measuring or recording means, e.g. sensors, in relation to patient
    • A61B5/6801—Arrangements of detecting, measuring or recording means, e.g. sensors, in relation to patient specially adapted to be attached to or worn on the body surface
    • A61B5/6802—Sensor mounted on worn items
    • A61B5/681—Wristwatch-type devices
    • A—HUMAN NECESSITIES
    • A61—MEDICAL OR VETERINARY SCIENCE; HYGIENE
    • A61B—DIAGNOSIS; SURGERY; IDENTIFICATION
    • A61B5/00—Measuring for diagnostic purposes; Identification of persons
    • A61B5/68—Arrangements of detecting, measuring or recording means, e.g. sensors, in relation to patient
    • A61B5/6801—Arrangements of detecting, measuring or recording means, e.g. sensors, in relation to patient specially adapted to be attached to or worn on the body surface
    • A61B5/6813—Specially adapted to be attached to a specific body part
    • A61B5/6824—Arm or wrist
    • A—HUMAN NECESSITIES
    • A61—MEDICAL OR VETERINARY SCIENCE; HYGIENE
    • A61B—DIAGNOSIS; SURGERY; IDENTIFICATION
    • A61B5/00—Measuring for diagnostic purposes; Identification of persons
    • A61B5/72—Signal processing specially adapted for physiological signals or for diagnostic purposes
    • A61B5/7235—Details of waveform analysis
    • A61B5/7246—Details of waveform analysis using correlation, e.g. template matching or determination of similarity
    • A—HUMAN NECESSITIES
    • A61—MEDICAL OR VETERINARY SCIENCE; HYGIENE
    • A61B—DIAGNOSIS; SURGERY; IDENTIFICATION
    • A61B5/00—Measuring for diagnostic purposes; Identification of persons
    • A61B5/72—Signal processing specially adapted for physiological signals or for diagnostic purposes
    • A61B5/7235—Details of waveform analysis
    • A61B5/7264—Classification of physiological signals or data, e.g. using neural networks, statistical classifiers, expert systems or fuzzy systems
    • A61B5/7267—Classification of physiological signals or data, e.g. using neural networks, statistical classifiers, expert systems or fuzzy systems involving training the classification device
    • A—HUMAN NECESSITIES
    • A63—SPORTS; GAMES; AMUSEMENTS
    • A63B—APPARATUS FOR PHYSICAL TRAINING, GYMNASTICS, SWIMMING, CLIMBING, OR FENCING; BALL GAMES; TRAINING EQUIPMENT
    • A63B24/00—Electric or electronic controls for exercising apparatus of preceding groups; Controlling or monitoring of exercises, sportive games, training or athletic performances
    • A63B24/0062—Monitoring athletic performances, e.g. for determining the work of a user on an exercise apparatus, the completed jogging or cycling distance
    • A—HUMAN NECESSITIES
    • A63—SPORTS; GAMES; AMUSEMENTS
    • A63B—APPARATUS FOR PHYSICAL TRAINING, GYMNASTICS, SWIMMING, CLIMBING, OR FENCING; BALL GAMES; TRAINING EQUIPMENT
    • A63B71/00—Games or sports accessories not covered in groups A63B1/00 - A63B69/00
    • A63B71/06—Indicating or scoring devices for games or players, or for other sports activities
    • A—HUMAN NECESSITIES
    • A61—MEDICAL OR VETERINARY SCIENCE; HYGIENE
    • A61B—DIAGNOSIS; SURGERY; IDENTIFICATION
    • A61B2503/00—Evaluating a particular growth phase or type of persons or animals
    • A61B2503/10—Athletes
    • A—HUMAN NECESSITIES
    • A61—MEDICAL OR VETERINARY SCIENCE; HYGIENE
    • A61B—DIAGNOSIS; SURGERY; IDENTIFICATION
    • A61B2562/00—Details of sensors; Constructional details of sensor housings or probes; Accessories for sensors
    • A61B2562/02—Details of sensors specially adapted for in-vivo measurements
    • A61B2562/0219—Inertial sensors, e.g. accelerometers, gyroscopes, tilt switches
    • A—HUMAN NECESSITIES
    • A63—SPORTS; GAMES; AMUSEMENTS
    • A63B—APPARATUS FOR PHYSICAL TRAINING, GYMNASTICS, SWIMMING, CLIMBING, OR FENCING; BALL GAMES; TRAINING EQUIPMENT
    • A63B24/00—Electric or electronic controls for exercising apparatus of preceding groups; Controlling or monitoring of exercises, sportive games, training or athletic performances
    • A63B24/0062—Monitoring athletic performances, e.g. for determining the work of a user on an exercise apparatus, the completed jogging or cycling distance
    • A63B2024/0068—Comparison to target or threshold, previous performance or not real time comparison to other individuals
    • A—HUMAN NECESSITIES
    • A63—SPORTS; GAMES; AMUSEMENTS
    • A63B—APPARATUS FOR PHYSICAL TRAINING, GYMNASTICS, SWIMMING, CLIMBING, OR FENCING; BALL GAMES; TRAINING EQUIPMENT
    • A63B2220/00—Measuring of physical parameters relating to sporting activity
    • A63B2220/40—Acceleration
    • A—HUMAN NECESSITIES
    • A63—SPORTS; GAMES; AMUSEMENTS
    • A63B—APPARATUS FOR PHYSICAL TRAINING, GYMNASTICS, SWIMMING, CLIMBING, OR FENCING; BALL GAMES; TRAINING EQUIPMENT
    • A63B2220/00—Measuring of physical parameters relating to sporting activity
    • A63B2220/40—Acceleration
    • A63B2220/44—Angular acceleration
    • A—HUMAN NECESSITIES
    • A63—SPORTS; GAMES; AMUSEMENTS
    • A63B—APPARATUS FOR PHYSICAL TRAINING, GYMNASTICS, SWIMMING, CLIMBING, OR FENCING; BALL GAMES; TRAINING EQUIPMENT
    • A63B2220/00—Measuring of physical parameters relating to sporting activity
    • A63B2220/80—Special sensors, transducers or devices therefor
    • A63B2220/83—Special sensors, transducers or devices therefor characterised by the position of the sensor
    • A63B2220/836—Sensors arranged on the body of the user
    • A—HUMAN NECESSITIES
    • A63—SPORTS; GAMES; AMUSEMENTS
    • A63B—APPARATUS FOR PHYSICAL TRAINING, GYMNASTICS, SWIMMING, CLIMBING, OR FENCING; BALL GAMES; TRAINING EQUIPMENT
    • A63B2230/00—Measuring physiological parameters of the user
    • A63B2230/62—Measuring physiological parameters of the user posture
    • A—HUMAN NECESSITIES
    • A63—SPORTS; GAMES; AMUSEMENTS
    • A63B—APPARATUS FOR PHYSICAL TRAINING, GYMNASTICS, SWIMMING, CLIMBING, OR FENCING; BALL GAMES; TRAINING EQUIPMENT
    • A63B2244/00—Sports without balls
    • A63B2244/20—Swimming
    • G—PHYSICS
    • G16—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR SPECIFIC APPLICATION FIELDS
    • G16H—HEALTHCARE INFORMATICS, i.e. INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR THE HANDLING OR PROCESSING OF MEDICAL OR HEALTHCARE DATA
    • G16H50/00—ICT specially adapted for medical diagnosis, medical simulation or medical data mining; ICT specially adapted for detecting, monitoring or modelling epidemics or pandemics
    • G16H50/70—ICT specially adapted for medical diagnosis, medical simulation or medical data mining; ICT specially adapted for detecting, monitoring or modelling epidemics or pandemics for mining of medical data, e.g. analysing previous cases of other patients

Definitions

  • the present invention relates to the field of wearable devices, and more particularly to a wearable device and a method for monitoring a state of motion thereof.
  • monitoring and tracking the movement state of an athlete is mainly performed by a visual manner, and is analyzed and identified based on video data afterwards.
  • This scheme cannot give statistical recognition results in real time;
  • Some professional sports monitoring equipment can analyze sports data such as swimming posture and exercise volume, but it is expensive and inconvenient to carry, and is not suitable for ordinary swimmers.
  • the accuracy of the motion state monitoring results obtained by the existing motion state monitoring program needs to be improved.
  • the invention provides a wearable device and a method for monitoring the state of motion thereof, so as to solve the problem that the prior art can only monitor and recognize the motion state of an ordinary athlete afterwards, the portability is poor, the accuracy of the monitoring result is low, and the ordinary can not meet the ordinary.
  • a method for monitoring a motion state by using a wearable device comprising: controlling a sensor to collect motion data of the user when a monitoring process starts;
  • test data is matched with the stored template data representing the predetermined motion state, and the template data that is successfully matched with the test data is obtained, and the motion state corresponding to the template data associated with the matched test data is determined to occur.
  • a wearable device in which a sensor is disposed, and the wearable device includes:
  • a data acquisition unit configured to control a sensor to collect motion data of the user when a monitoring process starts
  • a feature extraction unit configured to extract one or more feature quantities for identifying a motion state of the user from the motion data, to obtain test data
  • the state monitoring unit is configured to match the test data with the stored template data representing the predetermined motion state, and obtain template data that is successfully matched with the test data, and determine that the motion state corresponding to the template data associated with the matched test data occurs.
  • the beneficial effects of the present invention are: a method for monitoring motion state using a wearable device according to an embodiment of the present invention, on the one hand, utilizing the programmable capability of the wearable device, and simultaneously embedding a plurality of low-cost micro-electromechanical devices in the wearable device
  • System MEMS sensors such as accelerometers, gyroscopes, etc.
  • most wearable devices are light and compact. Usually, users wear them. They can recognize the movement status at any time during swimming activities, and perform corresponding exercise statistics to give users corresponding feedback and promote better people. Ground movement.
  • the recognition based on the motion state of the motion sensor in the wearable device is flexible and reliable, and is not affected by the environment, the light, etc., the system is simple to implement, the user is convenient to carry, and the use of the ordinary athlete is satisfied. Demand also increases the market competitiveness of wearable devices.
  • the technical solution of the embodiment can identify a plurality of user motion states through a combination of feature quantity extraction and template data matching, and the accuracy of the monitoring result obtained by the solution is high by experiments.
  • FIG. 1 is a flow chart of a method for monitoring a motion state using a wearable device according to an embodiment of the present invention
  • FIG. 2 is a flow chart of a method for monitoring a motion state using a wearable device according to another embodiment of the present invention
  • FIG. 3 is a schematic diagram of data acquisition according to still another embodiment of the present invention.
  • FIG. 4 is a schematic diagram of data windowing processing according to still another embodiment of the present invention.
  • FIG. 5 is a structural block diagram of a wearable device according to an embodiment of the present invention.
  • FIG. 6 is a schematic diagram of a hardware structure of a wearable device according to still another embodiment of the present disclosure.
  • wearable devices such as smart watches have the ability to program multiple low-cost MEMS sensors, such as accelerometers and gyroscopes, to provide hardware and software support for event recognition such as sensor-based swimming.
  • MEMS sensors such as accelerometers and gyroscopes
  • the smart watch is light and compact, and the user can wear it at any time. When performing activities such as swimming, the movement posture can be recognized at any time, and the corresponding amount of exercise statistics can be performed to give feedback to the user and promote better movement.
  • the design concept of the present invention is to realize the detection of the motion state by using the wearable device, and compare the collected test data with the preset template data according to the data collected by the motion sensor in the wearable device, such as the smart watch. Identify the results.
  • the motion state monitoring scheme is flexible and reliable, and is not affected by factors such as environment and light. The system is simple to implement and convenient for users to carry.
  • FIG. 1 is a flow chart of a method for monitoring a motion state using a wearable device according to an embodiment of the present invention.
  • the A sensor is disposed in the wearable device, and the method includes the following steps:
  • Step S101 when the monitoring process starts, the control sensor collects the motion data of the user
  • Step S102 extracting one or more feature quantities for identifying a motion state of the user from the motion data, to obtain test data;
  • step S103 the test data is matched with the stored template data representing the predetermined motion state, and the template data that is successfully matched with the test data is obtained, and the motion state corresponding to the template data associated with the matched test data is determined to occur.
  • the above steps S101 to S103 can be implemented by a function module provided in the wearable device.
  • the method for implementing motion state monitoring by using a wearable device collects motion data of a user by controlling a sensor, and extracts one or the motion data for identifying a motion state of the user.
  • the plurality of feature quantities are used to obtain test data, and the test data is matched with the stored template data representing the predetermined motion state, and the template data that is successfully matched with the test data is obtained, and the motion state corresponding to the template data associated with the matched test data is determined to occur. .
  • wearable characteristics of the wearable device the movement state of the athlete can be monitored and recognized in real time, so that the athlete can understand his or her exercise state and help the exerciser improve the exercise effect.
  • wearable devices are more common and less expensive than professional sports monitoring devices, and can meet the motion monitoring needs of ordinary athletes.
  • the implementation of the method for monitoring the motion state of the wearable device in the embodiment of the present invention is described by taking the swimming motion state monitoring as an example. It should be emphasized that the technical solution of the embodiment of the present invention can also be used for other sports.
  • the monitoring of the status is applied to the identification of other daily activities, such as walking, running, going upstairs, going downstairs, and the like.
  • a wearable device such as a smart watch is taken as an example to illustrate that the smart watch is used to monitor the swimming state of the swimmer.
  • the motion recognition based on the smart watch needs to consider the calculation amount and power consumption.
  • Wearable devices such as smart watches are resource-constrained. In the recognition process, the continuous perception of the smart watch requires a lot of energy. Further effective strategies are needed to control the complexity of the algorithm, reduce the amount of calculation, and improve the perceived efficiency, thereby improving the friendliness of user use.
  • the embodiment of the present invention adopts simple and effective pre-processing measures to remove the influence of noise, and extracts several limited time domain features that are truly distinguishable, and avoids complex feature calculations to reduce the amount of calculation.
  • the support vector machine (Support Vector Machine, SVM for short) is used to identify the swimming posture.
  • SVM is suitable for small sample training sets and does not require too complicated training. Process, and it has excellent generalization ability, which can well recognize the swimming posture of different users.
  • the classifier generated after the SVM is trained is simple. Compared with the recognition algorithm such as KNN (k-Nearest Neighbor), only a small amount of sample information is needed, which saves the storage space of the template data.
  • KNN algorithm is one of the simplest algorithms in data mining classification technology.
  • FIG. 2 is a flow chart of a method for monitoring a motion state by using a wearable device according to another embodiment of the present invention. Referring to FIG. 2, the overall process of monitoring a motion state by using a wearable device is:
  • the four basic swimming postures and the like are recognized, and the occurrence of the folding back motion can also be recognized.
  • motion state monitoring and recognition based on time domain features and support vector machine (SVM) is mainly used.
  • the sliding window processing method is adopted, and there is a certain overlap between adjacent sliding windows.
  • sliding mean filtering is used to remove the influence of noise, and relatively smooth data is obtained.
  • the support vector machine (SVM) method is used for classification and recognition.
  • SVM support vector machine
  • the trained SVM classifier is used for effective identification of the acceleration data generated during the swimming process of the user.
  • the method includes the following steps: the process starts, and step S201 is performed.
  • Step S201 sensor data acquisition
  • the acceleration sensor is used to realize the recognition of the four basic swimming postures, and the occurrence of the folding back movement during swimming can also be recognized.
  • step S201 the control sensor collects motion data of the user in one axial or multiple axial directions; when the first monitoring process starts, the three-axis acceleration sensor is controlled to collect the three-axis acceleration data of the user swimming motion, and the acquired three-axis acceleration is acquired. The data is saved to the cache.
  • FIG. 3 is a schematic diagram of data acquisition according to still another embodiment of the present invention.
  • 31 denotes a three-axis acceleration sensor
  • 32 denotes acceleration data acquired
  • 33 denotes a ring buffer
  • three-axis acceleration sensor 31 collects human body motion.
  • the obtained triaxial acceleration data 32 is used to put the collected triaxial acceleration data 32 into the corresponding ring buffer 33 (shown in FIG. 3 as a ring buffer 33).
  • This embodiment can save the design by using the ring buffer 33.
  • the storage space of the system is also convenient for subsequent sampling of the collected acceleration data and subsequent addition of sliding window processing.
  • the collected acceleration data 32 may not be placed in the ring buffer 33, which is not limited thereto.
  • FIG. 3 is a schematic illustration of taking a three-axis acceleration of a human body motion by an acceleration sensor as an example.
  • the triaxial angular velocity data of the human motion may also be collected by the gyroscope, or both
  • the three-axis acceleration data is acquired by the acceleration sensor, and the three-axis angular velocity data is collected by the gyroscope, and then the acceleration data and the angular velocity data are separately trained, and there is no limitation thereto.
  • the acquired triaxial acceleration data is pre-processed before the one or more feature quantities for identifying the swimming state of the user are extracted from the swimming motion data.
  • step S202 is performed to perform a windowing pre-processing operation on the acceleration data.
  • Step S202 sliding window processing
  • the sliding window processing method is adopted, and the sampling data is simultaneously sampled according to a predetermined frequency, and the sampling data is windowed by a sliding window of a predetermined step to obtain each axis of a predetermined length.
  • the moving step of the sliding window needs to satisfy the condition that the data in the adjacent sliding window partially overlaps; that is, a certain overlap is ensured between the adjacent sliding windows.
  • the reason why the data between adjacent sliding windows is partially overlapped is to prevent inaccurate identification due to data omission.
  • FIG. 4 is a schematic diagram of processing of adding a sliding window according to still another embodiment of the present invention; as shown in FIG. 4, sampling from a ring buffer that respectively stores three-axis acceleration data of the X-axis, the Y-axis, and the Z-axis, according to a predetermined frequency, The sampled data is windowed.
  • the sampling frequency is 50 Hz (ie, 50 data is sampled in one second)
  • the sliding window has a moving step size of N/2 samples.
  • the size of the sliding window is the length of the original sampled data obtained in T seconds, that is, N sampling data is taken out from the three ring buffers of the X-axis, the Y-axis, and the Z-axis at the same time for test identification.
  • the moving step of the sliding window is half of the size of the sliding window. It can be understood that in other embodiments of the present invention, the moving step of the sliding window may also be 1/ of the size of the sliding window. 3, etc., as long as the data in the adjacent sliding window is partially overlapped.
  • the window function used in the data windowing process in this embodiment is a rectangular window, and the rectangular window belongs to a zero-power window of a time variable.
  • the window function is not limited to a rectangular window, and other window functions may be used, and there is no limitation on the window function.
  • Step S203 filtering processing
  • K-time nearest neighbor mean filtering For each obtained axial acceleration data of a predetermined length, K-time nearest neighbor mean filtering is used for smoothing filtering to remove interference noise. Specifically, for the X, Y, and Z triaxial acceleration data in the sliding window, sliding mean filtering (for example, K-time neighboring mean filtering) is used to remove the influence of noise, and relatively smooth data is obtained.
  • sliding mean filtering for example, K-time neighboring mean filtering
  • filtering the acceleration data of the predetermined length to filter the interference noise includes: filtering the data points of each axial direction of the predetermined length of the original data, and selecting the adjacent one of the left side of the data point The number of data points and a predetermined number of data points adjacent to the right side of the data point are selected, and the average of the selected data points is calculated and the value of the filtered data points is replaced by the mean.
  • the present embodiment uses K-time neighboring mean filtering to perform filtering processing.
  • K time neighbor neighbor mean filtering is set by prior
  • the number K of nearest neighbors is then used as the value of the data point after the filtering process, in the acceleration data of each axis, the average value of the data composed of the K nearest neighbor data points and the right K neighbor data points on the left side of any data point.
  • the K-time neighbor homogenization filtering is:
  • N is the length of the X-axis data, that is, the size of the sliding window (the length of the data in this embodiment is 50)
  • K is the number of neighbors selected in advance, that is, how many nearest neighbors are selected from the left and right of a certain data point.
  • a xj is the component of the acceleration signal a j on the X axis
  • a' xi is the filtered data corresponding to a xj
  • i is the position index of the acceleration data on the X axis
  • j is the position index of the acceleration data on the X axis.
  • j and i are auxiliary relationships.
  • filtering processing methods for example, median filtering, Butterworth filtering, etc., as long as the original The acceleration data can be filtered, and the filtering algorithm is not limited.
  • Step S204 feature extraction
  • a plurality of feature quantities for identifying the swimming state of the user are extracted from the swimming motion data after the filtering process, and test data is obtained.
  • several of the following time domain feature quantities are extracted from each axial motion data: mean, standard deviation, minimum value, maximum value, skewness, kurtosis, and correlation coefficient.
  • time domain features are extracted for the X, Y, and Z axes in the sliding window, including: mean, standard deviation, minimum, maximum, skewness, kurtosis, and correlation coefficient.
  • X, Y are extracted.
  • the seven time domain features on the Z triaxial form a 21-dimensional feature vector.
  • time domain features such as X-axis or Y-axis or Z-axis
  • a plurality of seven time domain features of an axial data for example, extracting four time domain features of mean, minimum, maximum, and skewness of data on the X axis.
  • several of the seven time-domain features of the three axes may be extracted to form a feature vector, for example, when extracting the mean, minimum, maximum, and skewness of the data on the X-axis, the Y-axis, and the Z-axis, respectively. Domain characteristics, no restrictions on this.
  • the method in this embodiment further includes: The statistical analysis is used to calculate the correlation between the test data composed of several feature quantities and the user's motion state, and the test data is filtered according to the correlation between the test data and the user's motion state, and the filtered test data is obtained. Match the filtered test data with the template data.
  • Step S205 template training
  • the template data is generated by the collected standard swimming state data of a plurality of users and stored in the smart watch.
  • the standard swimming state data includes at least the following category data: breaststroke data, freestyle data, butterfly data, backstroke data, and foldback status data;
  • the swimming state corresponding to the template data associated with the test data is: the swimming state of the user is identified as the breaststroke swimming stroke, the freestyle swimming posture, the butterfly swimming posture, the back swimming stroke, or the template data associated with the test data. Foldback status.
  • Step S206 the SVM classification model
  • the template data to train the support vector machine SVM classifier select one of the two types of template data from the template data to train one SVM two classifiers, and get trained to distinguish any two of the N template data.
  • the SVM two types of classifiers respectively match the test data with each of the trained SVM classifiers, and obtain matching results between the test data and each SVM classifier, each matching result corresponding to a template data, and statistics
  • the number of template data that appears, and the template data with the most occurrences is used as the template data that matches the test data successfully.
  • step S205 template training and step S206
  • the SVM classification model can be pre-trained and stored in the smart watch, so that the user does not need to go to the training template during the process of using the smart watch to perform the swimming state. Train the SVM classification model to save time in swimming state recognition. That is to say, in actual application, step S205 and step S206 may be omitted.
  • Step S207 the SVM recognizes that, when the user's swimming posture is recognized, the collected three-axis acceleration sensor data is similarly processed, and the test data of each sliding window is extracted, and the trained SVM two-class classifier is used. , can identify the swimming posture currently used by the user.
  • Step S208 the swimming posture and the folding back movement recognition result
  • step S207 the swimming state currently adopted by the user is determined. And if the position of the user's return point at the time of swimming is recognized, the number of parameters swept by the user can be further counted, and according to the length of the pool, parameters such as the speed of swimming can be calculated.
  • the method in this embodiment further includes: after determining that the current swimming state of the user is a foldback state, determining whether the time interval between the time point when the current foldback state occurs and the time point when the previous foldback state occurs is greater than If the time threshold is set, it is determined that the determined return status is valid, otherwise, it is determined that the determined return status is invalid; and when the return status is determined to be valid, the time point at which the current return status occurs is used and used.
  • the reentry status occurs at the time point when the rewind status occurs, the stored reentry status occurs. Point.
  • the acceleration sensor built in the smart watch is used for processing, convenient to carry, flexible to use, and real-time recognition of the swimming posture, and given User real-time feedback makes it easy for users to keep track of their sports statistics.
  • a limited number of typical and well-discriminating time domain features are extracted, which avoids complexity compared to other frequency domain or time-frequency domain features. The feature calculation reduces the amount of calculation.
  • SVM support vector machine
  • Step 31 the sensor collects data
  • the X-axis acceleration of the X, Y and Z of the swimming action is acquired by using the built-in acceleration sensor of the smart watch.
  • the collected data is stored in a ring buffer of length Len, as shown in Figure 3.
  • Step 32 sliding window processing.
  • the three-axis acceleration data is taken out from the ring buffer to add a sliding window, as shown in FIG.
  • sliding window processing is to take a fixed length of data segment from the sensor data, which can be understood as covering the sensor data with a sliding window, and there is a certain length of overlap between adjacent sliding windows, such as overlapping Half a window long.
  • Step 33 filtering processing
  • the collected raw acceleration data is filtered to filter out interference noise.
  • the K-time nearest neighbor averaging filter is used for processing.
  • the K-time nearest neighbor mean filtering is to set the mean value of the sequence of the nearest neighbors K in each axial acceleration data as the value of the point after preprocessing. .
  • special treatment must be done. Take as many neighbors as possible as the object of the averaging process.
  • the feature extraction of the acceleration data is performed. Due to the complex extraction of frequency domain and time-frequency domain features (such as wavelet features), the computation of these features on a smart watch is costly, which increases the computation time and is not conducive to real-time swimming gesture recognition on smart watches.
  • the specific calculation methods of each time domain feature are as follows:
  • the standard deviation reflects the degree of dispersion of the acceleration data, and it is also an important feature for identifying static motions and dynamic motions.
  • the standard deviation can be used to determine whether the user is currently in a relatively static state. If the standard deviation of the acceleration data in the three axial directions is less than a preset threshold, it is considered that the user is not currently swimming, and no further recognition processing is performed.
  • ⁇ x is the standard deviation of the X-axis acceleration samples (data).
  • Skewness is a statistical feature used to measure the skew direction and skewness of acceleration data distribution.
  • ⁇ x is the standard deviation of the X-axis acceleration samples (data)
  • f i is the sampling interval of the acceleration samples (data).
  • the kurtosis reflects the steepness of the acceleration data at the peak of the data curve and is an important statistical feature.
  • the correlation coefficient is an indicator of the degree of linear correlation between variables.
  • time-domain features of mean, standard deviation, minimum, maximum, skewness, kurtosis and correlation coefficient can also be calculated for the data on the Y-axis and the Z-axis. Then for each sliding window, the time domain features extracted from the X, Y, Z triaxial data can form a 21 dimensional feature vector.
  • the time domain features used in the embodiments of the present invention are important statistical features, and have sufficient distinguishing ability for swimming postures, etc., and compared with these time domain features. When the features such as FFT and wavelet transform are extracted, the recognition performance is not significantly improved, but the amount of calculation is increased.
  • the feature selection method may be further used to further reduce the dimension of the feature vector without degrading the recognition performance.
  • the feature selection method may use Principal Component Analysis (PCA), Linear Discriminant Analysis (LDA) or boosting algorithm.
  • Support vector machine is suitable for small sample training set, and has excellent generalization and promotion ability. It has good non-user dependence and can well recognize the swimming posture of different users.
  • the classifier generated by the SVM after training is simple. Compared with the identification method such as KNN, only a small amount of sample information is needed, which saves the storage space of the template. Therefore, in this embodiment, the SVM classifier is used to identify the swimming posture. Compared with DTW (Dynamic Time Warping), SVM classification recognition can adopt a fixed-length sliding window. The processing is relatively simple, and it is not necessary to calculate the start time and end time of the user action, and the processing speed is improved.
  • K(s, s i ) is a kernel function, which corresponds to the inner product operation in the transformation space.
  • the test data to be identified is substituted into the above formula (9) for calculation, and the category of the test data can be judged according to the symbol output by the symbol function.
  • the SVM classifier in step 35 includes two parts of training and use (for identification). Specifically, the training of the SVM classifier includes:
  • the kernel function may select a Radial basis function (RBF).
  • RBF Radial basis function
  • the parameters to be determined are the kernel function parameter ⁇ and the penalty factor C.
  • the present invention employs a cross-validation based grid search method. That is, different pairs of parameters (C, ⁇ ) are searched, and the pair of parameters with the highest precision is selected as the optimal result by the cross-validation method.
  • Classification using the trained SVM classifier includes:
  • the acquired three-axis acceleration data is processed according to steps S201-S204 in FIG. 2, and the 21-dimensional feature vector of each sliding window is extracted, and the trained SVM multi-class classifier is adopted. , can identify the swimming posture currently used by the user.
  • the embodiment of the invention adopts a "one-to-one" method. That is, from the N classifications, if training samples of any two categories are selected to train a two-class classifier, a total of N*(N-1)/2 two-class classifiers are needed. Although the method has a large number of classifiers, the correct rate is high. The test sample is input into the SVM classifier, and the final recognition result is generated using a voting (max-wins-voting, MWV) strategy.
  • the template data of the standard swimming state is four types: breaststroke, freestyle, backstroke, and butterfly.
  • the folding operation is also used as one type in this embodiment, that is, in this embodiment.
  • SVM training there are two ways to implement SVM training:
  • the first way is to select a two-class classifier for training samples of any two categories, and a total of N*(N-1)/2 two-class classifiers are needed.
  • N 4
  • the two categories of breaststroke and backstroke are selected to form a two-class classifier, namely (frogstroke/backstroke)
  • test data A is brought into the optimal classification of the two types of classifiers.
  • the function formula it can be obtained that the test data A belongs to the breaststroke or the matching result belonging to the backstroke. Then, the test data A and the remaining five two types of classifiers (swim categories) are respectively compared, and five matching results are obtained.
  • the number of occurrences of the template data in the matching result of the test data A will be The most template data is used as the template data of the test data A matching, that is, the category to which the test data A belongs.
  • the test data A was identified as belonging to the breaststroke.
  • the second way is that when training the SVM two types of classifiers, in order to reduce the number of comparisons, one type of stroke can be regarded as the first category, and all categories except the strokes can be regarded as the second category.
  • the recognition result can be obtained.
  • breaststroke is taken as a category
  • other categories other than breaststroke ie, freestyle, backstroke, butterfly, and reentry actions
  • the test data is matched with the two types of classifiers of the SVM. After one comparison, it can be determined that the test data is not a breaststroke, and then the test can be determined by comparing with other trained SVM classifiers. The specific category of the data.
  • Step 36 reentry point identification
  • the acceleration threshold judgment and other methods can not reliably identify the occurrence of the reentry action.
  • the acceleration data may be abrupt when reversing, there is a way to judge the slope change of the acceleration curve. If the slope suddenly increases and exceeds a certain threshold, the reentry action is judged.
  • Another method is to judge the magnitude of the acceleration. When the amplitude suddenly increases and exceeds a certain threshold, the folding back action is judged to occur.
  • these methods are not reliable, because for different swimming postures and different people, the folding back movements vary greatly, and the situation is varied. Sometimes the acceleration changes do not conform to these laws, and simply rely on the acceleration threshold to judge Not reliable.
  • the folding back movement is also recognized as one type, and is put together with the four basic swimming postures, and the above methods are used for training and recognition, and a total of five types of patterns need to be identified.
  • the timing is started after the folding back operation is recognized, and the folding back operation may occur again after the certain time threshold TH_T is exceeded. . Since there is a minimum time interval TH_T for each swim after swimming, the next foldback action will not occur during the minimum time interval after judging the foldback. If a foldback action is detected within this minimum time interval, it is ignored directly.
  • FIG. 5 is a structural block diagram of a wearable device according to an embodiment of the present invention.
  • a sensor is disposed in the wearable device, and the wearable device 50 includes:
  • the data collection unit 501 is configured to control the sensor to collect motion data of the user when the monitoring process starts;
  • a feature extraction unit 502 configured to extract one or more feature quantities for identifying a motion state of the user from the motion data, to obtain test data;
  • the state monitoring unit 503 is configured to match the test data with the stored template data representing the predetermined motion state, and obtain template data that is successfully matched with the test data, and determine that the motion state corresponding to the template data associated with the matched test data occurs.
  • the data collecting unit 501 is specifically configured to control the sensor to collect motion data of the user in one axial or multiple axial directions;
  • the feature extraction unit 502 is specifically configured to extract one or more of the following time domain feature quantities from each axial motion data: mean, standard deviation, minimum, maximum, skewness, kurtosis, and correlation coefficient. .
  • the wearable device is specifically configured to monitor a swimming state of the user, and when the monitoring process starts, the control sensor collects the swimming motion data of the user; and extracts the swimming motion data for identifying the swimming state of the user.
  • a plurality of feature quantities are obtained, and test data is matched, and the test data is matched with each template data representing the state of swimming motion, and template data matching the test data is obtained, and the swimming state of the user is identified as corresponding to the template data associated with the test data.
  • the template data is generated by the collected standard swimming state data of a plurality of users and stored in the wearable device.
  • the standard swimming state data includes at least the following category data: breaststroke data, freestyle data, butterfly data, backstroke data, and foldback.
  • State data the swimming state corresponding to the template data that identifies the user's swimming state as the test data is associated with:
  • the swimming state of the user is identified as a breaststroke stroke, a freestyle stroke, a butterfly stroke, a backstroke stroke, or a foldback state corresponding to the template data associated with the test data.
  • the triaxial acceleration sensor is controlled to collect the triaxial acceleration data of the user swimming motion, the collected triaxial acceleration data is saved into the buffer, and the swimming state is extracted from the swimming motion data for identifying the swimming state of the user.
  • Performing the following preprocessing operations on the acquired triaxial acceleration data simultaneously sampling from the buffer according to a predetermined frequency, and windowing the sampled data by a sliding window of a predetermined step to obtain each axial acceleration data of a predetermined length, wherein The moving step of the sliding window needs to satisfy the condition that the data in the adjacent sliding window partially overlaps; and, for each obtained axial acceleration data of a predetermined length, respectively, K-time neighboring mean filtering is used for smoothing filtering to remove interference noise.
  • the wearable device further includes: a dimensionality reduction processing unit, configured to calculate, by using statistical analysis, a correlation between test data composed of one or more feature quantities and a motion state of the user, and The test data is filtered according to the correlation between the test data and the user's motion state, and the filtered test data is obtained, and the filtered test data is matched with the template data.
  • a dimensionality reduction processing unit configured to calculate, by using statistical analysis, a correlation between test data composed of one or more feature quantities and a motion state of the user, and The test data is filtered according to the correlation between the test data and the user's motion state, and the filtered test data is obtained, and the filtered test data is matched with the template data.
  • the state monitoring unit is specifically configured to train the support vector machine SVM classifier by using the template data, select a template data of any two categories from the template data, and train a two-class classifier to obtain a training.
  • the SVM two types of classifiers that can distinguish any two of the N template data are matched, and the test data is matched with each of the trained SVM classifiers to obtain test data and each SVM classifier.
  • the matching result, each matching result corresponds to a template data, and counts the number of template data that appears, and the template data with the most occurrences is used as template data that matches the test data successfully.
  • the wearable device further includes: a foldback action confirming unit, configured to determine a time point at which the current foldback state occurs and a previous foldback after determining that the current swim state of the user is a foldback state Whether the time interval between the time points when the state occurs is greater than the preset time threshold, and if yes, determining that the determined return status is valid, otherwise, determining that the determined return status is invalid; and, when the return status is determined as When it is valid, the time point at which the rewind status occurs is saved and the stored reentry status occurrence time point is updated at the time point when the rewind status occurs.
  • a foldback action confirming unit configured to determine a time point at which the current foldback state occurs and a previous foldback after determining that the current swim state of the user is a foldback state Whether the time interval between the time points when the state occurs is greater than the preset time threshold, and if yes, determining that the determined return status is valid, otherwise, determining that the determined return status is invalid; and
  • the wearable device further includes: a static determining unit, configured to separately calculate a standard deviation of the collected sensor data on each axis before matching the test data with each template data; The standard deviation of the sensor data on each axis is compared with a preset standard deviation threshold. If the standard deviation of the sensor data on each axis is less than the standard deviation threshold, it is determined that the user is not in motion and no further matching processing is performed.
  • a static determining unit configured to separately calculate a standard deviation of the collected sensor data on each axis before matching the test data with each template data. The standard deviation of the sensor data on each axis is compared with a preset standard deviation threshold. If the standard deviation of the sensor data on each axis is less than the standard deviation threshold, it is determined that the user is not in motion and no further matching processing is performed.
  • the wearable device of the present embodiment can be applied to the foregoing method for using a wearable device to perform a motion state.
  • the wearable device of this embodiment can be applied to the foregoing method for using a wearable device to perform a motion state.
  • the description of the method part of using the wearable device to exercise state will not be repeated here.
  • the technical solution of the embodiment of the present invention is compared with the prior art, such as video analysis or professional detection equipment, by using an acceleration sensor built in a smart watch for convenient carrying, flexible use, and swimming posture.
  • Real-time identification is easy for users to keep track of their movements.
  • the present invention extracts a limited number of typical and well-identified time domain features, compared to other frequency domain or time-frequency domain features. Avoid complex feature calculations and reduce the amount of calculations.
  • using the support vector machine SVM with good generalization ability for recognition non-user-restricted recognition capability can be realized, that is, the swimming postures of different users can be well recognized, and each user is avoided.
  • the technical solution of the embodiment can identify a plurality of user motion states through a combination of feature quantity extraction and template data matching, and the accuracy of the monitoring result obtained by the solution is high by experiments.

Landscapes

  • Health & Medical Sciences (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Engineering & Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • General Health & Medical Sciences (AREA)
  • Molecular Biology (AREA)
  • Animal Behavior & Ethology (AREA)
  • Veterinary Medicine (AREA)
  • Biophysics (AREA)
  • Pathology (AREA)
  • Public Health (AREA)
  • Biomedical Technology (AREA)
  • Heart & Thoracic Surgery (AREA)
  • Medical Informatics (AREA)
  • Surgery (AREA)
  • Physiology (AREA)
  • Artificial Intelligence (AREA)
  • Oral & Maxillofacial Surgery (AREA)
  • Dentistry (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Psychiatry (AREA)
  • Signal Processing (AREA)
  • Geometry (AREA)
  • Physical Education & Sports Medicine (AREA)
  • Evolutionary Computation (AREA)
  • Fuzzy Systems (AREA)
  • Mathematical Physics (AREA)
  • Human Computer Interaction (AREA)
  • Measurement Of The Respiration, Hearing Ability, Form, And Blood Characteristics Of Living Organisms (AREA)

Abstract

一种可穿戴设备（50）及利用其监测运动状态的方法，该可穿戴设备（50）中设置有传感器（31），该方法包括：当一次监测过程开始时控制传感器（31）采集用户的运动数据（S101）；从运动数据中提取用于识别用户运动状态的一个或多个特征量，得到测试数据（S102）；将测试数据与存储的代表预定运动状态的模板数据进行匹配，得到与测试数据匹配成功的模板数据，确定该匹配到的测试数据关联的模板数据对应的运动状态发生（S103）。通过可穿戴设备（50）实现对佩戴者的运动状态进行监测，由于可穿戴设备（50）的可穿戴特性，用户随身佩戴，在运动过程中可以随时对运动状态进行监测识别，方便用户了解自己的运动状态，帮助改善运动效果，提升了可穿戴设备（50）的用户体验。

Description

可穿戴设备及利用其监测运动状态的方法 技术领域
本发明涉及可穿戴设备领域，尤其涉及一种可穿戴设备及利用其监测运动状态的方法。
背景技术
传统地，对运动者的运动状态(例如，游泳状态)进行监控和跟踪主要通过视觉的方式进行，事后根据视频数据进行分析和识别，这种方案不能实时地给出统计识别结果；另外，也有一些专业的运动监测设备可以对游泳的姿态和运动量等运动数据进行分析，但是价格昂贵，而且携带不方便，不适用于普通的游泳爱好者。并且，现有的运动状态监测方案得到的运动状态监测结果的准确性也有待提高。
发明内容
本发明提供了一种可穿戴设备及利用其监测运动状态的方法，以解决现有技术只能事后对普通运动者的运动状态进行监测识别，便携性差，监测结果的准确度低，不能满足普通用户游泳运动时对游泳状态的监测需求的问题。
根据本发明的一个方面，提供了一种利用可穿戴设备监测运动状态的方法，该可穿戴设备中设置有传感器，方法包括：当一次监测过程开始时控制传感器采集用户的运动数据；
从运动数据中提取用于识别用户运动状态的一个或多个特征量，得到测试数据；
将测试数据与存储的代表预定运动状态的模板数据进行匹配，得到与测试数据匹配成功的模板数据，确定该匹配到的测试数据关联的模板数据对应的运动状态发生。
根据本发明的另一个方面，提供了一种可穿戴设备，该可穿戴设备中设置有传感器，可穿戴设备包括：
数据采集单元，用于当一次监测过程开始时控制传感器采集用户的运动数据；
特征提取单元，用于从运动数据中提取用于识别用户运动状态的一个或多个特征量，得到测试数据；
状态监测单元，用于将测试数据与存储的代表预定运动状态的模板数据进行匹配，得到与测试数据匹配成功的模板数据，确定该匹配到的测试数据关联的模板数据对应的运动状态发生。
本发明的有益效果是：本发明实施例的利用可穿戴设备进行运动状态监测的方法，一方面，利用可穿戴设备具有的可编程能力，以及可穿戴设备中可同时嵌入多种低成本微机电系统MEMS传感器(如加速度计、陀螺仪等)为基于传感器的游泳等运动状态识别提供了软硬件支持。另一方面，可穿戴设备大多都轻便小巧，通常用户会随身佩戴，在进行游泳等活动时可以随时对运动状态进行识别，并进行相应的运动量统计，给用户相应的反馈，并促进人们更好地运动。相比现有技术方案，基于可穿戴设备中运动传感器的运动状态(如游泳姿势)识别灵活可靠，不受环境、光线等的影响，系统实现简单，用户携带方便，满足了普通运动者的使用需求，也提高了可穿戴设备的市场竞争力。并且，本实施例的技术方案通过特征量提取和模板数据匹配相结合的技术手段，能够识别多种用户运动状态，经实验验证，本方案得到的监测结果的准确度较高。
附图说明
图1是本发明一个实施例的一种利用可穿戴设备监测运动状态的方法的流程图；
图2是本发明另一个实施例的一种利用可穿戴设备监测运动状态的方法的流程图；
图3是本发明又一个实施例的数据采集的示意图；
图4是本发明又一个实施例的数据加窗处理的示意图；
图5是本发明一个实施例的可穿戴设备的结构框图。
图6是本公开又一个实施例的可穿戴设备的硬件结构示意图。
具体实施方式
目前，智能手表等可穿戴设备的兴起给运动识别的广泛应用提供了可能。首先，智能手表等可穿戴设备具有可编程能力，能够同时嵌入多种低成本MEMS传感器，如加速度计、陀螺仪等，对基于传感器的游泳等活动识别提供了软硬件支持。其次，智能手表轻便小巧，用户随身佩戴，在进行游泳等活动时可以随时对运动姿势等进行识别，并进行相应的运动量统计，给用户反馈，促进人们更好地运动。
基于此，本发明的设计构思在于：利用可穿戴设备实现运动状态的检测，具体根据可穿戴设备如智能手表中运动传感器采集的数据，将这些采集的测试数据和预设的模板数据比较，得到识别结果。相比传统方式，这种运动状态监测的方案灵活可靠，不受环境、光线等因素的影响，系统实现简单，用户携带方便。
实施例一
图1是本发明一个实施例的一种利用可穿戴设备监测运动状态的方法的流程图，参见图1，该可 穿戴设备中设置有传感器，该方法包括如下步骤：
步骤S101，当一次监测过程开始时控制传感器采集用户的运动数据；
步骤S102，从运动数据中提取用于识别用户运动状态的一个或多个特征量，得到测试数据；
步骤S103，将测试数据与存储的代表预定运动状态的模板数据进行匹配，得到与测试数据匹配成功的模板数据，确定该匹配到的测试数据关联的模板数据对应的运动状态发生。
上述步骤S101至S103可以由设置在可穿戴设备中的功能模块实现。
由图1所示的方法可知，本发明实施例的这种利用可穿戴设备实现运动状态监测的方法，通过控制传感器采集用户的运动数据，从运动数据中提取用于识别用户运动状态的一个或多个特征量，得到测试数据，将测试数据与存储的代表预定运动状态的模板数据进行匹配，得到与测试数据匹配成功的模板数据后确定匹配到的测试数据关联的模板数据对应的运动状态发生。如此，利用可穿戴设备的可穿戴特性，可以实时对运动者的运动状态进行监测和识别，方便运动者了解自己的运动状态，帮助运动者改善运动效果。另外，与专业的运动监测设备相比，可穿戴设备更常用并且价格低，能够满足普通运动者的运动状态监测需求。
以下实施例中，以游泳运动状态监测为例对本发明实施例的利用可穿戴设备监测运动状态的方法的实现过程进行说明，需要强调的是，本发明实施例的技术方案还可以用于其他运动状态的监测，例如应用于其它日常活动的识别，如走、跑、上楼、下楼等动作。其他运动状态监测的具体实现可以参见下面实施例中对游泳运动状态监测的说明。另外，下面实施例中，以智能手表这种可穿戴设备为例说明，利用智能手表来实现对游泳者的游泳运动状态的监测。
游泳等运动状态监测识别在智能手表上的实际应用仍然面临着一些有待解决的问题，具体有：
(1)在游泳时，不同用户的姿势和习惯都不同，甚至相差比较大，而且在泳池里水的阻力和波动等对传感器读数的影响也比较大，需要采取合理的技术手段，提取有效的特征，并采用合适的识别方法，以对不同用户都能有较高的识别能力。
(2)基于智能手表的运动姿势识别需要考虑计算量和功耗的问题。智能手表等可穿戴设备是资源受限的。在识别过程中，智能手表的持续感知需要消耗不少的能量。需要进一步采取有效的策略，控制算法的复杂性，减少计算量，提高感知效率，从而提高用户使用的友好性。
针对这些问题，本发明实施例采用简单有效的预处理措施以去除噪声的影响，并提取真正有区别能力的、有限的几种时域特征，避免复杂的特征计算，以降低计算量。另外，采用支持向量机(Support Vector Machine，简称SVM)对游泳姿势进行识别。SVM适合小样本训练集，不需要太复杂的训练 过程，而且它具有优秀的泛化能力，能够很好地识别不同用户的游泳姿态。同时，SVM经过训练后产生的分类器简洁，相比KNN(k-Nearest Neighbor)等识别算法，只需要用到很少的样本信息，节约了模板数据的存储空间。KNN算法是数据挖掘分类技术中最简单的算法之一。
实施例二
图2是本发明另一个实施例的一种利用可穿戴设备监测运动状态的方法的流程图，参见图2，利用可穿戴设备监测运动状态的方法整体的流程是：
首先，对四种基本的游泳姿势等进行识别，同时也可以识别出折返动作的发生。这里主要采用基于时域特征和支持向量机(SVM)的运动状态监测识别。其次，对于游泳运动时采集的三轴加速度数据，采用滑动窗的处理方法，相邻滑动窗之间要有一定的重叠。对滑动窗中X、Y、Z三轴加速度数据，分别采用滑动均值滤波以去除噪声的影响，得到比较平滑的数据。然后，在每个滑动窗中，分别提取时域特征。最后，采用支持向量机(SVM)方法进行分类识别。开始识别前，要采集几种标准的不同姿势的游泳数据进行训练，得到SVM多类分类器。实际使用过程中，针对用户游泳过程中所产生的加速度数据，利用所训练的SVM分类器进行有效的识别。
具体的，参见图2包括如下步骤：流程开始，执行步骤S201，
步骤S201，传感器数据采集；
本发明实施例利用加速度传感器，实现对四种基本的游泳姿势进行识别，同时也可以识别出游泳时的折返动作的发生。
在步骤S201中，控制传感器采集一个轴向或多个轴向上用户的运动数据；当一次监测过程开始时控制三轴加速度传感器采集用户游泳运动时的三轴加速度数据，将采集的三轴加速度数据保存到缓存中。
图3是本发明又一个实施例的数据采集示意图，参见图3，其中，31表示三轴加速度传感器，32表示采集得到的加速度数据，33表示环形缓冲区；三轴加速度传感器31采集人体动作时得到的三轴加速度数据32，将采集的三轴加速度数据32放入对应的环形缓冲区33(图3示出了一个环形缓冲区33)中，本实施例采用环形缓冲区33的设计可以节省系统的存储空间，也方便对采集的加速度数据后续进行采样以及后续添加滑动窗处理。
本领域技术人员能够理解，在本发明的其他实施例中，也可以不采用环形缓冲区33放置采集的加速度数据32，对此不作限制。
此外，需要强调的是，图3是以通过加速度传感器采集人体动作的三轴加速度为例进行的示意性说明。但是在本发明的其他实施例中，也可以通过陀螺仪采集人体动作的三轴角速度数据，或者既 通过加速度传感器采集三轴加速度数据又通过陀螺仪采集三轴角速度数据，然后分别对加速度数据和角速度数据进行训练，对此不做限制。
本实施例的方法采集到原始的加速度数据后，在从游泳动作数据中提取用于识别用户游泳状态的一个或多个特征量之前，对采集的三轴加速度数据进行预处理操作
具体来说，是在执行步骤S201之后，执行步骤S202以对加速度数据进行加窗预处理操作。
步骤S202，滑动窗处理；
对于游泳运动时采集的三轴加速度数据，采用滑动窗的处理方法，按照预定的频率同时从缓存中采样，并以预定步长的滑动窗对采样数据进行加窗处理，得到预定长度的各轴向加速度数据，其中，滑动窗的移动步长需满足相邻滑动窗中的数据部分重叠的条件；即，相邻滑动窗之间保证有一定的重叠。而之所以设置相邻滑动窗之间的数据有部分重叠是为了防止数据遗漏造成识别不准确。
图4是本发明又一个实施例的添加滑动窗处理示意图；如图4所示，从分别保存X轴、Y轴、Z轴三轴加速度数据的环形缓冲区中，按照预定的频率采样，对采样数据进行加窗处理。本实施例中，采样频率为50Hz(即，一秒钟采样得到50个数据)，每个滑动窗的大小N＝50*T个采样数据，其中T为滑动窗覆盖的时间长度(即，秒数)，滑动窗的移动步长为N/2个采样数据。滑动窗的大小即为T秒内得到的原始采样数据的长度，也就是说，同时从X轴、Y轴、Z轴三个环形缓冲区中每次分别取出N个采样数据进行测试识别。
需要说明的是，本实施例中设置滑动窗的移动步长为滑动窗大小的一半，可以理解，在本发明的其他实施例中，滑动窗的移动步长也可以是滑动窗大小的1/3等，只要保证相邻滑动窗中的数据部分重叠即可。另外，本实施例中数据加窗处理采用的窗函数为矩形窗，矩形窗属于时间变量的零次幂窗。但是在本发明实施例中窗函数不限于矩形窗，也可以使用其它窗函数，对窗函数没有限制。
步骤S203，滤波处理；
对得到的预定长度的各轴向加速度数据，分别采用K时间近邻均值滤波进行平滑滤波处理，以去除干扰噪声。具体是对滑动窗中X、Y、Z三轴加速度数据，分别采用滑动均值滤波(例如，K时间近邻均值滤波)以去除噪声的影响，得到比较平滑的数据。
本实施例中，对预定长度的加速度数据进行滤波处理以滤除干扰噪声包括：对预定长度的原始数据的每个轴向的进行滤波处理的数据点，选取该数据点左侧相邻的预定数目的数据点以及选取该数据点右侧相邻的预定数目的数据点，计算选取出的数据点的均值并由该均值替换滤波处理的数据点的数值。
即，本实施例采用K时间近邻均值滤波进行滤波处理。K时间近邻均值滤波是通过事先设定时 间最近邻的个数K，然后在各轴加速度数据中，把任意一数据点左边K个近邻数据点和右边K个近邻数据点所组成的数据的均值作为滤波处理后该数据点的值。
对于时间序列中前K个数据点和最后K个数据点，须做特殊处理，取尽可能多的邻居数据点作为均值化处理的对象。
以三轴加速度数据中的X轴的数据为例，K时间近邻均值化滤波为：
Figure PCTCN2017092739-appb-000001
其中，N是X轴数据的长度，即滑动窗的大小(本实施例中数据长度为50)，K是预先选取的邻居个数，即选取某一个数据点左、右各多少个最近邻的邻居，axj为加速度信号aj在X轴上的分量，a'xi是axj对应的滤波后的数据，i表示X轴上加速度数据的位置索引，j表示X轴上加速度数据的位置索引，j和i之间为辅助关系。
需要说明的是，本发明其他实施例中，除了K时间近邻均值化滤波之外，还可以采用其它滤波处理方法，例如，中值滤波，巴特沃斯(Butterworth)滤波等，只要能够实现对原始加速度数据进行滤波处理即可，对滤波算法不作限制。
步骤S204，特征提取；
本实施例从滤波处理后的游泳动作数据中提取用于识别用户游泳状态的若干个特征量，得到测试数据。具体的，本实施例中从每个轴向的运动数据中提取下述时域特征量中的若干个：均值、标准差、最小值、最大值、偏度、峰度和相关系数。
例如，对滑动窗口中的X、Y、Z轴均提取七种时域特征，包括：均值、标准差、最小值、最大值、偏度、峰度和相关系数，优选地，提取X、Y、Z三轴上的七种时域特征共组成21维的特征向量。
需要说明的是，在进行特征提取时，在满足所需识别性能的前提下，可以只提取一轴向(如X轴或Y轴向或Z轴向)的七种时域特征，或者只提取一轴向数据的七种时域特征中的若干个，例如，提取X轴上数据的均值、最小值、最大值、偏度这四种时域特征。或者，可以提取三轴向的七种时域特征中的若干个组成特征向量，例如，分别提取X轴、Y轴、Z轴上数据的均值、最小值、最大值、偏度这四种时域特征，对此不作限制。
进一步的，在提取特征后，为提高数据运算速度，降低计算复杂度，本实施例的方法还包括： 利用统计分析计算由若干个特征量组成的测试数据与用户运动状态之间的相关性，并根据测试数据与用户运动状态之间的相关性对测试数据进行筛选，得到筛选后的测试数据，后续利用筛选后的测试数据与模板数据进行匹配。
步骤S205，模板训练；
模板数据是由采集到的多个用户的标准游泳状态数据生成，并存储在智能手表中，标准游泳状态数据至少包括如下类别数据：蛙泳数据、自由泳数据，蝶泳数据、仰泳数据以及折返状态数据；识别用户的游泳状态为该测试数据关联的模板数据对应的游泳状态包括：识别用户的游泳状态为该测试数据关联的模板数据对应的蛙泳泳姿、自由泳泳姿、蝶泳泳姿、仰泳泳姿或折返状态。
步骤S206，SVM分类模型；
利用模板数据训练支持向量机SVM分类器，从模板数据中选择任意两种类别的模板数据训练一个SVM两类分类器，得到训练好的能够将N种模板数据中任意两种模板数据区分开的SVM两类分类器，分别将测试数据与训练好的每个SVM两类分类器进行匹配，获取测试数据与每个SVM两类分类器的匹配结果，每一匹配结果对应一模板数据，并统计出现的模板数据的数目，将出现次数最多的模板数据作为与测试数据匹配成功的模板数据。
需要说明的是，步骤S205，模板训练和步骤S206，SVM分类模型；可以预先训练好后存储在智能手表中，这样在用户使用智能手表进行游泳状态的识别过程中，不需要再去训练模板和训练SVM分类模型，以节省游泳状态识别的时间。也就是说，在实际应用时，步骤S205和步骤S206可以省略。
步骤S207，SVM识别，在对用户的游泳姿势进行识别时，对采集到的三轴加速度传感器数据，按照类似的处理，提取出各个滑动窗的测试数据后，采用训练出的SVM两类分类器，可以识别出用户当前所采用的游泳姿势。
步骤S208，游泳姿势和折返动作识别结果；
根据步骤S207的识别结果，确定用户当前采用的游泳状态。并且如果识别出用户在游泳时的折返点位置，则可以进一步统计出用户所游的趟数，再根据泳池长度，可以计算出游泳的速度等参数。
另外，本实施例中的方法还包括：在确定出用户当前的游泳状态为折返状态后，判断本次折返状态发生的时间点与上一次折返状态发生的时间点之间的时间间隔是否大于预设的时间阈值，是则，确定本次判断出的折返状态有效，否则，确定本次判断出的折返状态无效；以及，当折返状态判断为有效时，保存本次折返状态发生的时间点并用本次折返状态发生的时间点更新存储的折返状态发生时 间点。
至此，流程结束。
由上可知，相比以往的传统方法，如视频分析或专业的检测设备，采用内置在智能手表中的加速度传感器进行处理，携带方便，使用灵活，而且对游泳姿势可以进行实时的识别，并给予用户实时的反馈，方便用户随时掌握自己的运动统计状况。而且，为了适应智能手表等可穿戴设备的资源受限状况，提取了有限的几种典型而且有很好区别能力的时域特征，相比其它的频域或时-频域特征，避免了复杂的特征计算，减少了计算量。另一方面，采用具有很好泛化能力的支持向量机(SVM)进行识别，可以提供非用户受限的识别能力，即，可以对不同用户的游泳姿势都可以很好地识别，避免了每个用户使用前都需要单独训练的情况，方便用户的使用，提高了用户体验。
以下对利用智能手表监测用户的游泳运动状态进行更详细的说明。
实施例三
本实施例的利用智能手表监测游泳运动状态的方法具体包括如下步骤：
步骤31，传感器采集数据
采用智能手表内置的加速度传感器，采集游泳动作的X、Y、Z三轴加速度。采集的数据分别保存到长度为Len的环形缓冲区(ring buffer)中，如图3所示。
步骤32，滑动窗处理。
从环形缓冲区中取出三轴加速度数据分别添加滑动窗，如图4所示。每个窗大小为N个采样，覆盖的时间长度为T秒，滑动窗的移动步长为step个采样(可以取步长＝N/2，即相邻窗的数据重叠半个窗长)。后续对每个窗的数据进行滤波处理。
滑动窗处理的目的：滑动窗处理是每次从传感器数据中取一个固定长度的数据段，可以理解为用滑动窗来覆盖传感器数据，而相邻滑动窗之间有一定长度的重叠，如重叠半个窗长。每次对当前的滑动窗内的数据进行处理，包括提取特征后进行识别，可以判断出当前时间段所对应的游泳姿势。这样采用一个固定长度的滑动窗，处理相对简单，而且可以每次都对相同长度的数据提取特征，方便下一步识别。
步骤33，滤波处理
对采集的原始加速度数据进行滤波处理，以滤除干扰噪声。
本发明一个实施例中，采用K时间近邻均值滤波器进行处理。K时间近邻均值滤波是通过事先设定时间最近邻的个数K，在各轴向加速度数据中，把任意一点左、右各K个元素所组成的序列的均值作为预处理后该点的值。对于时间序列中前K个数据点和最后K个数据点，必须做特殊处理， 取尽可能多的邻居作为均值化处理的对象。
步骤34，特征提取
在对加速度数据进行预处理的基础上，进行加速度数据的特征提取。由于频域和时-频域特征(如小波特征)提取复杂，在智能手表上计算这些特征开销较大，会增加运算时间，不利于在智能手表上进行实时的游泳姿势识别。
对此，本发明一个实施例中只提取真正有区别能力的、有限的几种时域特征，避免复杂的特征计算，以降低计算量。具体的，本发明一个实施例中是对每个滑动窗中X、Y、Z三轴加速度数据，分别提取七种时域特征，包括：均值、标准差、最小值、最大值、偏度、峰度和相关系数，组成了共21维的特征向量。以X轴加速度数据xi,i＝1,2,…,N为例，下面公式(2)至公式(8)中出现的i表示X轴上加速度数据的位置索引，例如，i＝5表示该数据是X轴上加速度数据序列中的第五个数据。各个时域特征具体计算方式如下：
(1)均值：
Figure PCTCN2017092739-appb-000002
(2)标准差：
Figure PCTCN2017092739-appb-000003
其中
Figure PCTCN2017092739-appb-000004
为X轴加速度样本(数据)的均值。标准差反映了加速度数据的离散程度，它也是识别静态动作与动态动作的重要特征。可以利用标准差来判断用户目前是否处于相对静止状态，如果三个轴向上的加速度数据的标准差都小于一个预设的阈值，则认为用户目前没有在游泳，不做进一步的识别处理。
(3)最小值和最大值：
min_x＝min({xi,i＝1,2,…N})      公式(4)
max_x＝max({xi,i＝1,2,…N})       公式(5)
(4)偏度：
Figure PCTCN2017092739-appb-000005
其中
Figure PCTCN2017092739-appb-000006
为X轴加速度样本(数据)的均值，σx为X轴加速度样本(数据)的标准差。偏度是用来度量加速度数据分布偏斜方向和偏斜程度的统计特征。
(5)峰度：
Figure PCTCN2017092739-appb-000007
其中
Figure PCTCN2017092739-appb-000008
为X轴加速度样本(数据)的均值，σx为X轴加速度样本(数据)的标准差，fi为加速度样本(数据)的采样间隔。峰度反映了加速度数据在数据曲线顶峰处的陡峭程度,是一个重要的统计特征。
(6)相关系数：
Figure PCTCN2017092739-appb-000009
其中
Figure PCTCN2017092739-appb-000010
和
Figure PCTCN2017092739-appb-000011
分别为X和Y轴上的加速度样本(数据)的均值。相关系数是衡量变量之间线性相关程度的指标。
同理，对Y轴和Z轴上的数据也可计算出均值、标准差、最小值、最大值、偏度、峰度和相关系数这七种时域特征。那么对于每个滑动窗，从X、Y、Z三轴数据提取的时域特征可组成21维的特征向量。
这几种特征是直接在时域进行计算的，所以不需要复杂的变换系数处理。而对于其它较复杂的傅里叶变换FFT、小波变换等需要变换系数的算法，如果采用这些需要变换系数的算法，对每个滑动窗的数据再进行处理和提取，需要更大的计算量。根据针对游泳场景的试验结果，本发明实施例所采用的这几种时域特征都是重要的统计特征，对于游泳姿势等已具有足够的区分能力，而相比这几种时域特征，采用FFT、小波变换等提取特征的方式进行识别时的识别性能并没有明显提高，而计算量却有所增加。
为了进一步减少计算量，模板训练时，在不明显降低识别性能的前提下可以尝试直接选择这7 种特征中的若干种组合进行识别，根据试验结果选择其中性能最好的组合。或者，在各提取三个轴向的七种时域特征得到共21维特征向量后，在不降低识别性能的前提下，也可以采用特征选择方法进一步降低特征向量的维数。特征选择方法可以采用主成分分析法(Principal Component Analysis，PCA)、线性判决分析法(Linear Discriminant Analysis，LDA)或boosting算法等。
步骤35，SVM分类器
支持向量机(SVM)适合小样本训练集，而且具有优秀的泛化和推广能力，具有较好的非用户依赖性，能够很好地识别不同用户的游泳姿态。同时，SVM经过训练后产生的分类器简洁，相比KNN等识别方法，只需要用到很少的样本信息，节约了模板的存储空间。因此本实施例选用SVM分类器对游泳姿势进行识别。并且相比DTW(Dynamic Time Warping，动态时间归整)，SVM分类识别可采用固定长度的滑动窗，处理相对简单，不需要计算用户动作的开始时间和结束时间，提高了处理速度。
设样本集合为(si,ti),i＝1,2,…,n，其中si为样本点，ti∈{-1,+1}为相应的类别标签。则最优分类函数表示为：
Figure PCTCN2017092739-appb-000012
其中，sgn()为符号函数，
Figure PCTCN2017092739-appb-000013
和b*是在保证正确分类的情况下，求解分类间隔最大时获得的最优解，K(s,si)为核函数，它对应变换空间中的内积运算。
则进行游泳状态识别时，将待识别的测试数据代入上式(9)进行计算，则测试数据的类别可根据符号函数输出的符号来判断。
步骤35中SVM分类器，包括训练和使用(进行识别)两部分内容，具体的，SVM分类器的训练包括：
在SVM分类模型中(参见上述公式9)，核函数可以选择径向基函数(Radial basis function，RBF)，对于RBF，需要确定的参数为核函数参数γ和惩罚因子C。为了确定最优的参数C和γ，以提高SVM分类器的识别精度，本发明采用基于交叉验证(cross-validation)的网格搜索方法。即搜索不同的参数对(C,γ)，通过交叉验证方法选择其中具有最高精度的参数对作为最优的结果。
在训练分类器时，首先对P个不同用户的四种基本游泳姿势的加速度数据进行采样，分别按图2中所示的步骤S201-S204进行处理。然后对每个人截取M个(如采用三轴加速度传感器则M＝3)滑动窗的特征向量，最后得到4*P*M行21列的特征矩阵，用来训练并构造SVM多类分类器。
使用训练完成的SVM分类器进行分类识别包括：
在对用户的游泳姿势进行识别时，对采集到的三轴加速度数据，按照图2中的步骤S201-S204处理，提取出各个滑动窗的21维特征向量，采用训练出的SVM多类分类器，可以识别出用户当前所采用的游泳姿势。
由于SVM处理的是二分类问题，在利用SVM对多个游泳姿势进行识别时，需要构建分类器。本发明实施例采用“一对一”的方法。即从N个分类中，选择任意两种类别的训练样本训练一个两类分类器，则共需要N*(N-1)/2个两类分类器。该方法虽然分类器的数量较多，但正确率较高。测试样本输入SVM分类器中，利用投票(max-wins-voting，MWV)策略产生最终识别结果。
举例而言，标准游泳状态的模板数据为四种：蛙泳、自由泳、仰泳和蝶泳，为了识别用户游泳时的折返动作，本实施例中将折返动作也作为一类，即，本实施例中共有5种类别，则在SVM训练时，有两种实现方式：
第一种方式是，选择任意两种类别的训练样本训练一个两类分类器，则共需要N*(N-1)/2个两类分类器。一个实施例中，假设N＝4，并选择蛙泳和仰泳这两种类别组成一个两类分类器，即(蛙泳/仰泳)，然后，将测试数据A带入该两类分类器的最优分类函数公式中，可得到，测试数据A属于蛙泳或属于仰泳的匹配结果。然后，将测试数据A和剩下的五个两类分类器(游泳的类别)分别比较，可得到五个匹配结果，最后，根据测试数据A的匹配结果中模板数据出现的数目，将出现次数最多的模板数据作为测试数据A匹配的模板数据，即作为测试数据A所属的类别。接上例，测试数据A与六个SVM两类分类器比较后，发现蛙泳出现了三次，而仰泳、自由泳、蝶泳各出现一次，则识别出测试数据A属于蛙泳。
第二种方式是，在训练SVM两类分类器时，为了减少比较的次数，可以将一种泳姿作为第一类别，而将除该泳姿之外的所有类别作为第二类别。这样，本实施例中为了识别泳姿，只需要同SVM两类分类器比较四次，即可得到的识别结果。举例而言，在一个SVM两类分类器中，将蛙泳作为一种类别，而将蛙泳之外的其他类别(即，自由泳、仰泳、蝶泳和折返动作)作为第二种类别。实际应用时，将测试数据和该SVM两类分类器匹配，通过一次比较，即可确定出该测试数据不属于蛙泳，然后再通过与训练好的其他SVM两类分类器比较，可确定出测试数据的具体类别。
步骤36，折返点识别
如果识别出用户在游泳时的折返点位置，则可以进一步统计出用户所游的趟数，再根据泳池长度，可以计算出游泳的速度等参数。但由于不同人的折返习惯和姿势都有很大的不同，简单地依靠加速度阈值判断等方法不能很可靠地识别出折返动作的发生。因为在折返时加速度数据可能会有突变，有一种方法是判断加速度曲线的斜率变化，如果斜率突然增加并超出一定的阈值则判断折返动作发 生；有另一种方法是判断加速度的幅值，当幅值突然增加并超过一定的阈值则判断折返动作发生。但这些方法并不可靠，因为对于不同的游泳姿势和不同的人来说，折返动作变化很大，情况多种多样，有时加速度的变化并不符合这些规律，简单地依靠这种加速度阈值来判断并不可靠。
为了有效地识别出折返动作，本发明实施例中将折返动作也作为一类进行识别，与四种基本泳姿放在一起，采用前述方法进行训练和识别，共需要识别五类模式。而且，为了进一步降低折返动作误识别的发生(即本来没有折返而识别为折返动作的情况)，在识别出折返动作之后启动计时，只有在超出一定的时间阈值TH_T后，才可能再次发生折返动作。因为游泳时游完每一趟都会有一个最小时间间隔TH_T，在判断出折返后，下一个折返动作在该最小时间间隔内是不会发生的。如果在该最小时间间隔内检测到折返动作，则直接忽略掉。
实施例四
本实施例中提供了一种可穿戴设备，图5是本发明一个实施例的可穿戴设备的结构框图，参见图5，该可穿戴设备中设置有传感器，可穿戴设备50包括：
数据采集单元501，用于当一次监测过程开始时控制传感器采集用户的运动数据；
特征提取单元502，用于从运动数据中提取用于识别用户运动状态的一个或多个特征量，得到测试数据；
状态监测单元503，用于将测试数据与存储的代表预定运动状态的模板数据进行匹配，得到与测试数据匹配成功的模板数据，确定该匹配到的测试数据关联的模板数据对应的运动状态发生。
在本发明的一个实施例中，数据采集单元501，具体用于控制传感器采集一个轴向或多个轴向上用户的运动数据；
特征提取单元502，具体用于从每个轴向的运动数据中提取下述时域特征量中的一个或多个：均值、标准差、最小值、最大值、偏度、峰度和相关系数。
在本发明的一个实施例中，可穿戴设备具体用于监测用户的游泳运动状态，当一次监测过程开始时控制传感器采集用户的游泳动作数据；从游泳动作数据中提取用于识别用户游泳状态的若干个特征量，得到测试数据，将测试数据与每个代表游泳运动状态的模板数据进行匹配，得到与测试数据匹配成功的模板数据，识别用户的游泳状态为该测试数据关联的模板数据对应的游泳状态；
其中，模板数据是由采集到的多个用户的标准游泳状态数据生成，并存储在可穿戴设备中，标准游泳状态数据至少包括如下类别数据：蛙泳数据、自由泳数据，蝶泳数据、仰泳数据以及折返状态数据；识别用户的游泳状态为该测试数据关联的模板数据对应的游泳状态包括：
识别用户的游泳状态为该测试数据关联的模板数据对应的蛙泳泳姿、自由泳泳姿、蝶泳泳姿、仰泳泳姿或折返状态。
在本发明的一个实施例中，控制三轴加速度传感器采集用户游泳运动时的三轴加速度数据，将采集的三轴加速度数据保存到缓存中，在从游泳动作数据中提取用于识别用户游泳状态的特征量之前，
对采集的三轴加速度数据进行如下预处理操作：按照预定的频率同时从缓存中采样，并以预定步长的滑动窗对采样数据进行加窗处理，得到预定长度的各轴向加速度数据，其中，滑动窗的移动步长需满足相邻滑动窗中的数据部分重叠的条件；以及，对得到的预定长度的各轴向加速度数据，分别采用K时间近邻均值滤波进行平滑滤波处理，以去除干扰噪声。
在本发明的一个实施例中，该可穿戴设备中还包括：降维处理单元，用于利用统计分析计算由一个或多个特征量组成的测试数据与用户运动状态之间的相关性，并根据测试数据与用户运动状态之间的相关性对测试数据进行筛选，得到筛选后的测试数据，将筛选后的测试数据与模板数据进行匹配。
在本发明的一个实施例中，状态监测单元，具体用于利用模板数据训练支持向量机SVM分类器，从模板数据中选择任意两种类别的模板数据训练一个两类分类器，得到训练好的能够将N种模板数据中任意两种模板数据区分开的SVM两类分类器，分别将测试数据与训练好的每个SVM两类分类器进行匹配，获取测试数据与每个SVM两类分类器的匹配结果，每一匹配结果对应一模板数据，并统计出现的模板数据的数目，将出现次数最多的模板数据作为与测试数据匹配成功的模板数据。
在本发明的一个实施例中，该可穿戴设备中还包括：折返动作确认单元，用于在确定出用户当前的游泳状态为折返状态后，判断本次折返状态发生的时间点与上一次折返状态发生的时间点之间的时间间隔是否大于预设的时间阈值，是则，确定本次判断出的折返状态有效，否则，确定本次判断出的折返状态无效；以及，当折返状态判断为有效时，保存本次折返状态发生的时间点并用本次折返状态发生的时间点更新存储的折返状态发生时间点。
在本发明的一个实施例中，该可穿戴设备中还包括：静止判断单元，用于在将测试数据与每个模板数据进行匹配之前，分别计算采集的各轴上传感器数据的标准差；将各轴上传感器数据的标准差与一个预设的标准差阈值进行比较，若各轴上传感器数据的标准差都小于该标准差阈值，则确定用户未处于运动状态，不做进一步的匹配处理。
需要说明的是，本实施例的这种可穿戴设备可应用到前述利用可穿戴设备进行运动状态的方法中，本实施例对可穿戴设备的工作过程的更多细节可以参见前述实施例中对利用可穿戴设备进行运动状态的方法部分的说明，在此不再赘述。
综上所述，本发明实施例的技术方案相比现有技术，如视频分析或专业的检测设备，通过采用内置在智能手表中的加速度传感器进行处理，携带方便，使用灵活，而且对游泳姿势可以进行实时的识别方便用户随时掌握自己的运动状况。而且，为了适应智能手表等设备的资源受限状况，本发明实施例中提取了有限的几种典型且有很好区别能力的时域特征，相比其它的频域或时-频域特征，避免了复杂的特征计算，减少了计算量。另一方面，采用具有很好泛化能力的支持向量机SVM进行识别，可以实现非用户受限的识别能力，即可以对不同用户的游泳姿势都可以很好地识别，避免了每个用户使用前都需要单独训练的情况，方便用户的使用，提高了用户体验。并且，本实施例的技术方案通过特征量提取和模板数据匹配相结合的技术手段，能够识别多种用户运动状态，经实验验证，本方案得到的监测结果的准确度较高。
以上所述，仅为本发明的具体实施方式，在本发明的上述教导下，本领域技术人员可以在上述实施例的基础上进行其他的改进或变形。本领域技术人员应该明白，上述的具体描述只是更好的解释本发明的目的，本发明的保护范围以权利要求的保护范围为准。

Claims (15)

  1. 一种利用可穿戴设备监测运动状态的方法，其中，该可穿戴设备中设置有传感器，所述方法包括：
    当一次监测过程开始时控制所述传感器采集用户的运动数据；
    从所述运动数据中提取用于识别用户运动状态的一个或多个特征量，得到测试数据；
    将所述测试数据与存储的代表预定运动状态的模板数据进行匹配，得到与测试数据匹配成功的模板数据，确定该匹配到的测试数据关联的模板数据对应的运动状态发生。
  2. 根据权利要求1所述的方法，其中，控制所述传感器采集用户的运动数据包括：
    控制所述传感器采集一个轴向或多个轴向上用户的运动数据；
    从所述运动数据中提取用于识别用户运动状态的一个或多个特征量包括：
    从每个轴向的运动数据中提取下述时域特征量中的一个或多个：
    均值、标准差、最小值、最大值、偏度、峰度和相关系数。
  3. 根据权利要求1所述的方法，其中，所述方法具体为：
    当一次监测过程开始时控制传感器采集用户的游泳动作数据；
    从所述游泳动作数据中提取用于识别用户游泳状态的一个或多个特征量，得到测试数据，
    将所述测试数据与每个代表游泳运动状态的模板数据进行匹配，得到与测试数据匹配成功的模板数据，识别用户的游泳状态为该测试数据关联的模板数据对应的游泳状态；
    其中，所述模板数据是由采集到的多个用户的标准游泳状态数据生成，并存储在可穿戴设备中，
    所述标准游泳状态数据至少包括如下类别数据：蛙泳数据、自由泳数据，蝶泳数据、仰泳数据以及折返状态数据；
    所述识别用户的游泳状态为该测试数据关联的模板数据对应的游泳状态包括：
    识别用户的游泳状态为该测试数据关联的模板数据对应的蛙泳泳姿、自由泳泳姿、蝶泳泳姿、仰泳泳姿或折返状态。
  4. 根据权利要求3所述的方法，其中，所述控制传感器采集用户的游泳动作数据包括：
    控制三轴加速度传感器采集用户游泳运动时的三轴加速度数据，将采集的三轴加速度数据保存到缓存中，
    在从所述游泳动作数据中提取用于识别用户游泳状态的特征量之前，对采集的三轴加速度数据进行如下预处理操作：
    按照预定的频率同时从所述缓存中采样，并以预定步长的滑动窗对采样数据进行加窗处理，得到预定长度的各轴向加速度数据，其中，所述滑动窗的移动步长需满足相邻滑动窗中的数据部分重叠的条件；
    以及，对得到的预定长度的各轴向加速度数据，分别采用K时间近邻均值滤波进行平滑滤波处理，以去除干扰噪声。
  5. 根据权利要求1所述的方法，其中，该方法还包括：
    利用统计分析计算由一个或多个特征量组成的测试数据与用户运动状态之间的相关性，并根据测试数据与用户运动状态之间的相关性对测试数据进行筛选，得到筛选后的测试数据，将筛选后的测试数据与所述模板数据进行匹配。
  6. 根据权利要求1所述的方法，其中，所述将所述测试数据与存储的代表预定运动状态的模板数据进行匹配，得到与测试数据匹配成功的模板数据，确定该匹配到的测试数据关联的模板数据对应的运动状态发生包括：
    利用模板数据训练支持向量机SVM分类器，从模板数据中选择任意两种类别的模板数据训练一个两类分类器，得到训练好的能够将N种模板数据中任意两种模板数据区分开的SVM两类分类器，
    分别将所述测试数据与训练好的每个SVM两类分类器进行匹配，获取所述测试数据与每个SVM两类分类器的匹配结果，每一匹配结果对应一模板数据，并统计出现的模板数据的数目，将出现次数最多的模板数据作为与所述测试数据匹配成功的模板数据。
  7. 根据权利要求4所述的方法，其中，该方法还包括：
    在确定出用户当前的游泳状态为折返状态后，判断本次折返状态发生的时间点与上一次折返状态发生的时间点之间的时间间隔是否大于预设的时间阈值，是则，确定本次判断出的折返状态有效，否则，确定本次判断出的折返状态无效；
    以及，当折返状态判断为有效时，保存本次折返状态发生的时间点并用本次折返状态发生的时间点更新存储的折返状态发生时间点。
  8. 根据权利要求2所述的方法，其中，该方法还包括：在将所述测试数据与每个模板数据进行匹配之前，
    分别计算采集的各轴上传感器数据的标准差；
    将各轴上传感器数据的标准差与一个预设的标准差阈值进行比较，若各轴上传感器数据的标准差都小于该标准差阈值，则确定用户未处于运动状态，不做进一步的匹配处理。
  9. 一种可穿戴设备，其中，该可穿戴设备中设置有传感器，可穿戴设备包括：
    数据采集单元，用于当一次监测过程开始时控制传感器采集用户的运动数据；
    特征提取单元，用于从所述运动数据中提取用于识别用户运动状态的一个或多个特征量，得到测试数据；
    状态监测单元，用于将所述测试数据与存储的代表预定运动状态的模板数据进行匹配，得到与测试数据匹配成功的模板数据，确定该匹配到的测试数据关联的模板数据对应的运动状态发生。
  10. 根据权利要求9所述的可穿戴设备，其中，
    所述数据采集单元，具体用于控制所述传感器采集一个轴向或多个轴向上用户的运动数据；
    所述特征提取单元，具体用于从每个轴向的运动数据中提取下述时域特征量中的一个或多个：均值、标准差、最小值、最大值、偏度、峰度和相关系数。
  11. 根据权利要求9所述的可穿戴设备，其中，所述数据采集单元，具体用于当一次监测过程开始时控制传感器采集用户的游泳动作数据；
    所述特征提取单元，具体用于从所述游泳动作数据中提取用于识别用户游泳状态的一个或多个特征量，得到测试数据，
    所述状态监测单元，具体用于将所述测试数据与每个代表游泳运动状态的模板数据进行匹配，得到与测试数据匹配成功的模板数据，识别用户的游泳状态为该测试数据关联的模板数据对应的游泳状态；
    其中，所述模板数据是由采集到的多个用户的标准游泳状态数据生成，并存储在可穿戴设备中，
    所述标准游泳状态数据至少包括如下类别数据：蛙泳数据、自由泳数据，蝶泳数据、仰泳数据以及折返状态数据；
    所述状态监测单元，具体用于识别用户的游泳状态为该测试数据关联的模板数据对应的蛙泳泳姿、自由泳泳姿、蝶泳泳姿、仰泳泳姿或折返状态。
  12. 根据权利要求11所述的可穿戴设备，其中，所述数据采集单元，具体用于控制三轴加速度传感器采集用户游泳运动时的三轴加速度数据，将采集的三轴加速度数据保存缓存中，
    在从所述游泳动作数据中提取用于识别用户游泳状态的特征量之前，对采集的三轴加速度数据进行如下预处理操作：
    按照预定的频率同时从所述缓存中采样，并以预定步长的滑动窗对采样数据进行加窗处理，得到预定长度的各轴向加速度数据，其中，所述滑动窗的移动步长需满足相邻滑动窗中的数据部分重叠的条件；
    以及，对得到的预定长度的各轴向加速度数据，分别采用K时间近邻均值滤波进行平滑滤波处理，以去除干扰噪声。
  13. 根据权利要求9所述的可穿戴设备，其中，还包括：降维处理单元，具体用于利用统计分析计算由一个或多个特征量组成的测试数据与用户运动状态之间的相关性，并根据测试数据与用户运动状态之间的相关性对测试数据进行筛选，得到筛选后的测试数据，将筛选后的测试数据与所述模板数据进行匹配。
  14. 根据权利要求9所述的可穿戴设备，其中，所述状态监测单元，具体用于利用模板数据训练支持向量机SVM分类器，从模板数据中选择任意两种类别的模板数据训练一个两类分类器，得到训练好的能够将N种模板数据中任意两种模板数据区分开的SVM两类分类器，分别将测试数据与训练好的每个SVM两类分类器进行匹配，获取测试数据与每个SVM两类分类器的匹配结果，每一匹配结果对应一模板数据，并统计出现的模板数据的数目，将出现次数最多的模板数据作为与测试数据匹配成功的模板数据。
  15. 根据权利要求9所述的可穿戴设备，其中，该可穿戴设备中还包括：折返动作确认单元，用于在确定出用户当前的游泳状态为折返状态后，判断本次折返状态发生的时间点与上一次折返状态发生的时间点之间的时间间隔是否大于预设的时间阈值，是则，确定本次判断出的折返状态有效，否则，确定本次判断出的折返状态无效；以及，当折返状态判断为有效时，保存本次折返状态发生的时间点并用本次折返状态发生的时间点更新存储的折返状态发生时间点。
PCT/CN2017/092739 2016-08-31 2017-07-13 可穿戴设备及利用其监测运动状态的方法 WO2018040757A1 (zh)

Priority Applications (1)

Application Number Priority Date Filing Date Title
US16/327,173 US20190209050A1 (en) 2016-08-31 2017-07-13 Wearable device and method of using same to monitor motion state

Applications Claiming Priority (2)

Application Number Priority Date Filing Date Title
CN201610789754.9A CN106237604A (zh) 2016-08-31 2016-08-31 可穿戴设备及利用其监测运动状态的方法
CN201610789754.9 2016-08-31

Publications (1)

Publication Number Publication Date
WO2018040757A1 true WO2018040757A1 (zh) 2018-03-08

Family

ID=58081145

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/CN2017/092739 WO2018040757A1 (zh) 2016-08-31 2017-07-13 可穿戴设备及利用其监测运动状态的方法

Country Status (3)

Country Link
US (1) US20190209050A1 (zh)
CN (1) CN106237604A (zh)
WO (1) WO2018040757A1 (zh)

Cited By (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN110852260A (zh) * 2019-11-08 2020-02-28 青岛联合创智科技有限公司 一种基于加速度计的溺水行为识别方法
CN114170264A (zh) * 2021-12-09 2022-03-11 湖南大学 基于智能可穿戴设备的驾驶状态监测方法

Families Citing this family (55)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN106237604A (zh) * 2016-08-31 2016-12-21 歌尔股份有限公司 可穿戴设备及利用其监测运动状态的方法
DE102016225648A1 (de) * 2016-12-20 2018-06-21 Bundesdruckerei Gmbh Verfahren und System zur verhaltensbasierten Authentifizierung eines Nutzers
CN106874874A (zh) * 2017-02-16 2017-06-20 南方科技大学 一种运动状态的识别方法及装置
CN106971059B (zh) * 2017-03-01 2020-08-11 福州云开智能科技有限公司 一种基于神经网络自适应健康监测的可穿戴设备
CN109002189B (zh) * 2017-06-07 2021-09-07 斑马智行网络(香港)有限公司 一种运动识别方法、装置、设备和计算机存储介质
CN107270934A (zh) * 2017-06-08 2017-10-20 海能电子(深圳)有限公司 一种基于穿戴式智能设备的游泳信息监测识别方法
CN107469326B (zh) * 2017-07-04 2020-08-07 广东乐心医疗电子股份有限公司 一种用于可穿戴设备的游泳监测方法与装置及可穿戴设备
CN107616795A (zh) * 2017-08-17 2018-01-23 华东师范大学 一种基于摄像头的非接触式实时呼吸率检测方法
CN107609501A (zh) * 2017-09-05 2018-01-19 东软集团股份有限公司 人体相近动作识别方法及装置、存储介质、电子设备
CN109635617A (zh) * 2017-10-09 2019-04-16 富士通株式会社 动作状态的识别方法、装置和电子设备
CN108008151A (zh) * 2017-11-09 2018-05-08 惠州市德赛工业研究院有限公司 一种基于三轴加速度传感器的运动状态识别方法及系统
CN108009572A (zh) * 2017-11-22 2018-05-08 中国地质大学（武汉） 移动设备坠落检测方法及其模型形成方法与移动设备
CN108564100A (zh) * 2017-12-12 2018-09-21 惠州Tcl移动通信有限公司 移动终端及其生成动作分类模型的方法、存储装置
CA3087787A1 (en) * 2017-12-28 2019-07-04 Sleep Number Corporation Bed having presence detecting feature
CN108460322A (zh) * 2017-12-28 2018-08-28 惠州市德赛工业研究院有限公司 一种泳姿识别方法及应用
CN108073287B (zh) * 2018-01-22 2024-01-19 杨尚林 运动监测方法、装置和可穿戴运动设备
WO2019185244A1 (en) * 2018-03-26 2019-10-03 Sony Corporation A method for providing information about an object and an object providing information
CN108830299A (zh) * 2018-05-21 2018-11-16 千寻位置网络有限公司 基于svm的智能设备佩戴模式的识别方法及系统、智能设备
CN109091848A (zh) * 2018-05-31 2018-12-28 深圳还是威健康科技有限公司 挥动动作识别方法、装置、终端及计算机可读存储介质
CN108837476B (zh) * 2018-06-08 2020-09-04 歌尔科技有限公司 一种检测游泳起点的方法、装置和智能穿戴设备
CN108968918A (zh) * 2018-06-28 2018-12-11 北京航空航天大学 早期帕金森的可穿戴辅助筛查设备
CN111388987A (zh) * 2018-06-29 2020-07-10 安徽省徽腾智能交通科技有限公司 游泳监测系统
CN108962394B (zh) * 2018-07-05 2021-04-16 广东工业大学 一种医疗数据决策支持方法及系统
CN109620241B (zh) * 2018-11-16 2021-10-08 歌尔科技有限公司 一种可穿戴设备及基于其的动作监测方法
CN109582713B (zh) * 2018-11-30 2023-05-19 歌尔科技有限公司 一种运动状态的识别方法、装置及终端
CN109663323A (zh) * 2018-12-20 2019-04-23 出门问问信息科技有限公司 一种基于可穿戴设备的运动状态监测方法及可穿戴设备
CN109621355A (zh) * 2018-12-29 2019-04-16 曲大方 武术对抗模拟训练设备
WO2020140266A1 (zh) * 2019-01-04 2020-07-09 深圳大学 一种智能手表的交互方法及交互系统
CN109758154B (zh) * 2019-01-09 2022-02-25 北京卡路里信息技术有限公司 一种运动状态确定方法、装置、设备及存储介质
CN110008847B (zh) * 2019-03-13 2021-07-20 华南理工大学 一种基于卷积神经网络的泳姿识别方法
CN109985369B (zh) * 2019-03-13 2020-06-19 华南理工大学 一种基于智能腕戴式设备的自适应泳姿识别方法
CN109948686B (zh) * 2019-03-13 2021-06-08 华南理工大学 一种基于九轴传感信号统计特征的泳姿识别方法
CN111803902B (zh) * 2019-04-10 2022-08-12 北京卡路里信息技术有限公司 泳姿识别方法、装置、可穿戴设备及存储介质
CN111814523B (zh) * 2019-04-12 2024-06-18 北京京东振世信息技术有限公司 人体活动识别方法及装置
CN110146077A (zh) * 2019-06-21 2019-08-20 台州知通科技有限公司 移动机器人姿态角解算方法
CN110180158B (zh) * 2019-07-02 2021-04-23 乐跑体育互联网（武汉）有限公司 一种跑步状态识别方法、系统及终端设备
CN110458033B (zh) * 2019-07-17 2023-01-03 哈尔滨工程大学 一种基于可穿戴位置传感器的人体行为序列识别方法
CN110751060B (zh) * 2019-09-29 2021-02-19 西安交通大学 一种基于多源信号的便携式运动模式实时识别系统
CN112580403A (zh) * 2019-09-29 2021-03-30 北京信息科技大学 一种跌倒检测的时频特征提取方法
CN110793539A (zh) * 2019-11-12 2020-02-14 浙江大学城市学院 基于运动分量阈值和机器学习的溺水检测方法和系统
CN111141284A (zh) * 2019-12-28 2020-05-12 西安交通大学 智能建筑人员热舒适度及热环境管理系统和方法
CN111166340B (zh) * 2019-12-31 2022-11-18 石家庄学院 基于自适应加速度信号分割的人体姿势实时识别方法
WO2021170384A1 (en) * 2020-02-27 2021-09-02 Robert Bosch Gmbh A controller to determine swim characteristics of a swimmer and method thereof
CN113706847B (zh) * 2020-05-21 2022-10-21 安徽华米健康科技有限公司 一种数据处理方法、装置、储存介质及可穿戴设备
CN111717210B (zh) * 2020-06-01 2022-11-11 重庆大学 一种驾驶员手部相对静止状态下脱离方向盘检测方法
CN111870248A (zh) * 2020-06-05 2020-11-03 安徽建筑大学 基于3d加速度信号的运动状态特征提取及识别方法
CN112115978A (zh) * 2020-08-24 2020-12-22 中国人民健康保险股份有限公司 一种运动识别方法、装置及存储介质
CN112274902B (zh) * 2020-10-16 2022-06-10 深圳卡路里科技有限公司 游泳数据获取方法及装置、游泳数据采集设备的控制方法
CN114533010A (zh) * 2020-11-25 2022-05-27 安徽华米健康科技有限公司 心率检测方法和装置
CN114663905A (zh) * 2020-12-08 2022-06-24 华为技术有限公司 活动识别方法、显示方法及电子设备
CN113057628A (zh) * 2021-04-04 2021-07-02 北京泽桥传媒科技股份有限公司 一种基于惯性传感器式动作捕捉方法
CN113780159A (zh) * 2021-09-08 2021-12-10 广州偶游网络科技有限公司 触地动作识别方法与存储介质
CN114237397A (zh) * 2021-12-16 2022-03-25 厦门大学 一种拳击动作识别方法、系统及电子设备
CN114344873B (zh) * 2022-01-14 2023-04-07 北京数感科技有限公司 一种预摆动作类型识别方法、系统及设备
CN116954395B (zh) * 2023-09-21 2023-11-21 深圳博骏科技有限公司 基于多传感器融合的鼠标位移调控系统

Citations (10)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20060161053A1 (en) * 2005-01-19 2006-07-20 Polar Electro Oy System for monitoring an individual's physiological state, and a computer software product
CN101520815A (zh) * 2008-07-05 2009-09-02 杭州义盛祥通信技术有限公司 个人运动管理系统及管理方法
CN101518442A (zh) * 2008-07-05 2009-09-02 杭州义盛祥通信技术有限公司 运动量化腕表及运动计量分析方法
CN102481479A (zh) * 2009-04-01 2012-05-30 达腾科技有限公司 游泳监测器
CN103908259A (zh) * 2014-04-03 2014-07-09 深圳市德凯瑞科技有限公司 一种智能穿戴设备及人体运动的监测与识别方法
US20150164377A1 (en) * 2013-03-13 2015-06-18 Vaidhi Nathan System and method of body motion analytics recognition and alerting
CN105159441A (zh) * 2015-07-28 2015-12-16 东华大学 基于自主式运动识别技术的私人教练智能手环
CN205145530U (zh) * 2015-10-27 2016-04-13 杭州镜之镜科技有限公司 一种检测游泳运动特性指标的设备
CN105892674A (zh) * 2016-04-26 2016-08-24 哈尔滨工程大学 一种基于智能手环的泳姿识别方法和基于智能手环的运动方案
CN106237604A (zh) * 2016-08-31 2016-12-21 歌尔股份有限公司 可穿戴设备及利用其监测运动状态的方法

Family Cites Families (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN105268171B (zh) * 2015-09-06 2018-09-18 安徽华米信息科技有限公司 步态监测方法、装置及可穿戴设备
CN105787434A (zh) * 2016-02-01 2016-07-20 上海交通大学 基于惯性传感器的人体运动模式识别方法
US11402905B2 (en) * 2018-01-09 2022-08-02 Holland Bloorview Kids Rehabilitation Hospital EEG brain-computer interface platform and process for detection of changes to mental state
US11263480B2 (en) * 2018-10-25 2022-03-01 The Boeing Company Machine learning model development with interactive model evaluation

Patent Citations (10)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20060161053A1 (en) * 2005-01-19 2006-07-20 Polar Electro Oy System for monitoring an individual's physiological state, and a computer software product
CN101520815A (zh) * 2008-07-05 2009-09-02 杭州义盛祥通信技术有限公司 个人运动管理系统及管理方法
CN101518442A (zh) * 2008-07-05 2009-09-02 杭州义盛祥通信技术有限公司 运动量化腕表及运动计量分析方法
CN102481479A (zh) * 2009-04-01 2012-05-30 达腾科技有限公司 游泳监测器
US20150164377A1 (en) * 2013-03-13 2015-06-18 Vaidhi Nathan System and method of body motion analytics recognition and alerting
CN103908259A (zh) * 2014-04-03 2014-07-09 深圳市德凯瑞科技有限公司 一种智能穿戴设备及人体运动的监测与识别方法
CN105159441A (zh) * 2015-07-28 2015-12-16 东华大学 基于自主式运动识别技术的私人教练智能手环
CN205145530U (zh) * 2015-10-27 2016-04-13 杭州镜之镜科技有限公司 一种检测游泳运动特性指标的设备
CN105892674A (zh) * 2016-04-26 2016-08-24 哈尔滨工程大学 一种基于智能手环的泳姿识别方法和基于智能手环的运动方案
CN106237604A (zh) * 2016-08-31 2016-12-21 歌尔股份有限公司 可穿戴设备及利用其监测运动状态的方法

Cited By (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN110852260A (zh) * 2019-11-08 2020-02-28 青岛联合创智科技有限公司 一种基于加速度计的溺水行为识别方法
CN110852260B (zh) * 2019-11-08 2023-05-30 青岛联合创智科技有限公司 一种基于加速度计的溺水行为识别方法
CN114170264A (zh) * 2021-12-09 2022-03-11 湖南大学 基于智能可穿戴设备的驾驶状态监测方法

Also Published As

Publication number Publication date
US20190209050A1 (en) 2019-07-11
CN106237604A (zh) 2016-12-21

Similar Documents

Publication Publication Date Title
WO2018040757A1 (zh) 可穿戴设备及利用其监测运动状态的方法
Hsu et al. Human daily and sport activity recognition using a wearable inertial sensor network
Xia et al. LSTM-CNN architecture for human activity recognition
CN105184325B (zh) 一种移动智能终端
WO2017050140A1 (zh) 一种人体动作识别方法、识别用户动作的方法和智能终端
Hsu et al. Wearable sport activity classification based on deep convolutional neural network
CN104200234B (zh) 人体动作建模和识别方法
CN106919958B (zh) 一种基于智能手表的人体手指动作识别方法
Jensen et al. Classification of kinematic swimming data with emphasis on resource consumption
WO2021139337A1 (zh) 基于深度学习模型的步态识别方法、装置和计算机设备
Zhang et al. SmartHandwriting: Handwritten Chinese character recognition with smartwatch
CN112801000B (zh) 一种基于多特征融合的居家老人摔倒检测方法及系统
CN109260673A (zh) 一种运动计数方法、装置、设备及存储介质
CN104484644A (zh) 一种手势识别方法和装置
CN109603142A (zh) 一种哑铃运动的识别计数方法及其装置
Zheng et al. L-sign: Large-vocabulary sign gestures recognition system
CN104586402A (zh) 一种人体活动的特征提取方法
Geng Research on athlete’s action recognition based on acceleration sensor and deep learning
Xu et al. Wearable muscle movement information measuring device based on acceleration sensor
Zhen et al. Hybrid Deep‐Learning Framework Based on Gaussian Fusion of Multiple Spatiotemporal Networks for Walking Gait Phase Recognition
Zhou et al. A Swarm Intelligence Assisted IoT-Based Activity Recognition System for Basketball Rookies
Peng Research on Emotion Recognition Based on Deep Learning for Mental Health
Fu et al. Ping pong motion recognition based on smart watch
CN111028488A (zh) 一种基于pso-svm的智能久坐提醒的方法
CN110694252A (zh) 一种基于六轴传感器的跑步姿态检测方法

Legal Events

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

Ref document number: 17845060

Country of ref document: EP

Kind code of ref document: A1

NENP Non-entry into the national phase

Ref country code: DE

122 Ep: pct application non-entry in european phase

Ref document number: 17845060

Country of ref document: EP

Kind code of ref document: A1