WO2024087151A1 - 一种基于角速度传感器的步态时相识别方法 - Google Patents

一种基于角速度传感器的步态时相识别方法 Download PDF

Info

Publication number
WO2024087151A1
WO2024087151A1 PCT/CN2022/128174 CN2022128174W WO2024087151A1 WO 2024087151 A1 WO2024087151 A1 WO 2024087151A1 CN 2022128174 W CN2022128174 W CN 2022128174W WO 2024087151 A1 WO2024087151 A1 WO 2024087151A1
Authority
WO
WIPO (PCT)
Prior art keywords
gait
angular velocity
event
signal
velocity signal
Prior art date
Application number
PCT/CN2022/128174
Other languages
English (en)
French (fr)
Inventor
孙方敏
王皓
侯沛尧
李烨
Original Assignee
中国科学院深圳先进技术研究院
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 中国科学院深圳先进技术研究院 filed Critical 中国科学院深圳先进技术研究院
Priority to PCT/CN2022/128174 priority Critical patent/WO2024087151A1/zh
Publication of WO2024087151A1 publication Critical patent/WO2024087151A1/zh

Links

Images

Definitions

  • the present invention relates to the field of computer application technology, and more specifically, to a gait phase recognition method and system based on an angular velocity sensor.
  • Walking is the most basic motor function of human beings.
  • the body posture i.e. gait
  • This seemingly simple movement process contains complex kinematics, dynamics and biomechanics, and requires a high degree of coordination of the muscle system, skeletal system and nervous system.
  • Normal gait is an activity completed by the human central nervous system controlling bones, muscles and other parts of the human body. It has certain stability, coordination, periodicity, directionality and individual differences. Any abnormality or disorder in any link will lead to abnormal gait. Factors that lead to abnormal gait include congenital inheritance, acquired diseases and accidental injuries.
  • Gait is periodic and regular, and each gait cycle during walking contains a series of foot posture transfers. People usually divide a series of time periods based on the changes in typical foot postures, which are called gait phases. Gait phases play an important role in the analysis and detection of gait abnormalities. Compared with normal gait, abnormal gait is manifested by missing phases, chaotic timing, and disproportionate proportions. Therefore, gait phases are an important indicator that reflects factors that affect body coordination, such as gait habits, age, and health status. For gait analysis systems, gait phase recognition is the basis for gait cycle extraction, gait phase segmentation, and gait spatiotemporal parameter calculation. Gait analysis systems must be able to accurately and reliably detect gait phases to meet the needs of clinical applications.
  • the current research on identifying gait phases can be divided into three categories: machine vision-based methods, pressure sensor-based methods, and wearable inertial sensor-based methods.
  • the machine vision-based method uses high-precision cameras and image processing technology to track the motion trajectory of highly reflective markers on the subject, and then calculates the position information of fixed parts through the corresponding algorithm to obtain the characteristic parameters of human gait.
  • the pressure sensor-based method uses pressure sensors laid on the ground or built into the soles of shoes to measure the ground reaction force during walking to perform gait analysis.
  • the wearable inertial sensor-based method uses inertial sensors worn at different positions on the body to extract gait signals, and after coordinate conversion and posture angle solution, the gait characteristics are calculated through comprehensive operations on acceleration signals, angular velocity signals, and magnetometers.
  • patent application CN202210185789.7 provides a human gait parameter extraction method based on Kinect, which is a method based on machine vision.
  • This solution extracts joint point position sequence data through the Kinect somatosensory depth sensor, filters the data and calculates the gait parameter values.
  • the gait feature space is constructed according to the gait parameter values, and the gait phase division strategy based on the random forest model is adopted to obtain the preliminary judgment result of the unilateral state of each frame, and then the abnormal phase is corrected for misjudgment to obtain the final gait phase division result.
  • This scheme has a more comprehensive gait feature representation, but affected by the detection range of the Kinect depth sensor, data collection needs to be in a specific venue, and the range of motion is limited; and the image information is easily affected by lighting, obstructions, and background environment.
  • Patent application CN202010202430.7 provides a gait health assessment method and device, which obtains gait pressure signals through 8-way plantar pressure sensors, and uses fuzzy logic inference rules to identify gait phases based on gait pressure signals.
  • This method has the advantages of portability, dynamic acquisition, and accurate reflection of plantar pressure changes, but the sensor can only obtain gait signals in the support phase, and cannot reflect complete gait information when used alone.
  • the position of the sensor on the sole has a great influence on the accuracy of pressure measurement, and the position of the sensor needs to be adjusted according to the size of the feet of different subjects.
  • Patent application CN202111029801.7 proposes a gait event point detection method based on angular velocity signals. By collecting angular velocity data of the calf and comparing the angular velocity near the feature point with a preset threshold, the time point when the heel touches the ground and the time point when the toe leaves the ground when a person walks are determined, and the gait cycle is divided accordingly. This method is easy to use and simple to calculate, but the number of gait events detected is small. If gait analysis is required, the gait cycle needs to be divided into more stages.
  • the method based on machine vision is usually deployed in a professional gait analysis laboratory, which is costly and requires professional technicians to operate. It is also easily restricted by light, occlusion, distance, etc., and has not yet been widely used.
  • the advantage of the method based on pressure sensors is that it can accurately display the changes in plantar pressure, which is usually used as the gold standard for gait detection.
  • this method can only present the pressure changes when the foot contacts the ground during walking. When the foot is in the air, the pressure signal is zero, and the gait information of the complete gait cycle cannot be obtained.
  • the position of the pressure sensor has a great influence on the detection accuracy. Due to mechanical loss, the plantar pressure sensor is not suitable for long-term continuous collection.
  • the method based on wearable inertial sensors has many advantages. It can be fixed on various parts of the human body, so complete gait information can be obtained. It can also work in any environment, is not affected by light, weather, shadows and occlusions, does not need to store a large amount of video image information, has good portability and is easy to wear. With the rapid development of micro-electro-mechanical system (MEMS) technology, wearable inertial sensors have gradually become an important tool in gait analysis research, but the accuracy of gait recognition by this method still needs to be improved.
  • MEMS micro-electro-mechanical system
  • the purpose of the present invention is to overcome the defects of the prior art and provide a gait phase recognition method based on an angular velocity sensor.
  • the method comprises:
  • the sensor is used to collect the original angular velocity signal of the target walking and the plantar pressure signal synchronously;
  • gait events are detected and gait phases are divided, thereby obtaining a recognition result.
  • the advantages of the present invention are that it provides a gait phase recognition method based on an angular velocity sensor, uses a 3-axis angular velocity sensor to collect angular velocity data of the position of the foot surface of the human body during walking, and establishes a gait significant feature recognition method based on angular velocity by mining the significant features of the inertial sensor information corresponding to different gait events, so as to achieve accurate recognition of different gait events and different gait phases.
  • the present invention is a gait phase recognition method based on a wearable inertial sensor, which has the advantages of simple calculation, high recognition accuracy, low cost, and long-term continuous monitoring of long-term dimensional changes of gait features.
  • FIG1 is a flow chart of a gait phase recognition method based on an angular velocity sensor according to an embodiment of the present invention
  • FIG2 is a schematic diagram of a process of a gait phase recognition method based on an angular velocity sensor according to an embodiment of the present invention
  • FIG3 is a schematic diagram of a force-bearing area on the sole of a foot according to an embodiment of the present invention.
  • FIG4 is a schematic diagram of a raw inertial sensor signal and a plantar pressure signal according to an embodiment of the present invention
  • FIG5 is a schematic diagram of converting sensor coordinates into geographic coordinates according to an embodiment of the present invention.
  • FIG6 is a comparison diagram of an angular velocity signal before and after coordinate conversion according to an embodiment of the present invention.
  • FIG7 is an angular velocity signal and a spectrum diagram according to an embodiment of the present invention.
  • FIG8 is a schematic diagram of angular velocity signals before and after filtering according to an embodiment of the present invention.
  • FIG9 is a schematic diagram of a threshold algorithm for detecting gait events according to an embodiment of the present invention.
  • FIG10 is a schematic diagram of a gait event detection result based on plantar pressure according to an embodiment of the present invention.
  • FIG11 is a schematic diagram of flat area division according to an embodiment of the present invention.
  • FIG12 is a schematic diagram of gait event detection results based on angular velocity according to an embodiment of the present invention.
  • FIG13 is a schematic diagram of gait phase division according to an embodiment of the present invention.
  • FIG14 is a schematic diagram of the duration of normal human gait phases according to an embodiment of the present invention.
  • FIG15 is a schematic diagram of a phase duration of a simulated abnormal human gait according to an embodiment of the present invention.
  • FIG16 is a schematic diagram of an experimental data platform according to an embodiment of the present invention.
  • FIG17 is a comparison diagram of a gait event detection result and a reference result according to an embodiment of the present invention.
  • the present invention proposes a gait event recognition method based on angular velocity signals through synchronously collected plantar pressure information, and marks the angular velocity signals based on gait events detected by plantar pressure; further, the significant features of angular velocity signals in different gait events are mined through time and frequency analysis methods, and a gait event significant feature recognition algorithm based on flat area detection and double trough detection is established, and accurate detection of different gait events is achieved based on angular velocity signals; finally, the gait cycle is divided into multiple phases according to the detected gait events and prior knowledge, so as to achieve accurate division of gait phases.
  • the provided gait phase recognition method based on an angular velocity sensor includes the following steps.
  • Step S110 synchronously collecting angular velocity signals and plantar pressure signals.
  • gait event detection methods require that gait events be labeled in a laboratory where high-speed motion capture equipment is installed.
  • a gait event detection method based on plantar pressure sensor information is proposed. Therefore, it is necessary to synchronously collect angular velocity sensor information and plantar pressure information during the modeling data collection process.
  • the inertial measurement unit (IMU) built into the wearable device Xsens is used to obtain the angular velocity signal during walking, and the sampling frequency of Xsens is set to 120Hz.
  • the plantar pressure distribution measurement system FS-INS-W99 is used to obtain the plantar pressure signal during exercise, and the sampling frequency is set to 50Hz.
  • the experimental data is collected and stored offline and used for subsequent modeling and analysis.
  • Two Xsens sensors are fixed to the left and right shoe uppers of the subject by straps or insulating tapes, and the plantar pressure distribution measurement system is fixed in the subject's shoe in the form of an insole.
  • the plantar pressure distribution measurement system FS-INS-W99 has 99 pressure sensing areas.
  • the 99-channel pressure data collected has high redundancy, which is not convenient for subsequent analysis and requires preliminary data screening.
  • FIG. 3 is a schematic diagram of the original angular velocity signal collected by the inertial sensor and the original plantar pressure signals of the four characteristic areas screened out.
  • Step S120 transforming the angular velocity signal from the sensor coordinate system to the geographic coordinate system.
  • the coordinate system conversion mainly involves the geographic coordinate system and the sensor coordinate system.
  • the "East-North-Sky" coordinate system can be selected. This coordinate system is stationary. The origin of this coordinate system is the center of mass of the carrier on the surface of the earth. One of the coordinate axes points in the opposite direction of the center of the earth, and the other two coordinate axes are parallel to the tangent directions of the longitude and latitude of the area, that is, the positive direction of the x-axis points to the east, the positive direction of the y-axis points to the north, and the positive direction of the z-axis points in the opposite direction of the gravity vector g.
  • the sensor frame is a coordinate system defined by the three orthogonal measurement axes of the inertial sensor (or inertial measurement unit) itself.
  • the output measurement values of the inertial sensor are all based on the sensor coordinate system.
  • the origin of the sensor coordinate system coincides with the origin of the geographic coordinate system; in the sensor coordinate system, the x-axis signal output measurement value reflects the inertial data in the front-to-back direction during walking; the y-axis signal output measurement value reflects the inertial data in the left-to-right direction during walking; and the z-axis signal output measurement value reflects the inertial data in the vertical direction during walking.
  • Common methods for describing the relationship between two coordinate systems include the Euler angle method, the direction cosine matrix method, the trigonometric function method, and the quaternion method.
  • the attitude matrix obtained does not need to be orthogonalized, but when the pitch angle of the carrier is ⁇ 90°, a singular point will appear, resulting in the loss of one degree of freedom. Therefore, this method cannot perform full attitude solution, and its use has certain limitations.
  • the direction cosine method uses the direction cosines of the vector to represent the attitude matrix. This method avoids the singular point problem encountered by the Euler angle method when solving the attitude matrix.
  • the direction cosine matrix has nine elements, so nine differential equations need to be solved, which has a large amount of calculation and is not practical in engineering.
  • the trigonometric function method is to equivalently represent the relationship between the two coordinate systems rotating around a fixed point with three rotations, and use the sine and cosine functions of the three rotation angles to represent the attitude function. This method can also avoid the singular point problem, but it also needs to solve six differential equations, which has a large amount of calculation.
  • the quaternion method needs to solve four differential equations. This method has small computational complexity, high precision, can avoid singularities, and provides smooth interpolation.
  • the results show that the quaternion method has the best performance. Therefore, preferably, the quaternion method is used for coordinate system conversion.
  • a quaternion consists of a real part and three imaginary parts.
  • the complex number representation of a unit quaternion can be written as:
  • R is the rotation matrix
  • R is the original data in the sensor coordinate system
  • FIG6 is a comparison diagram of the inertial sensor angular velocity signal before and after coordinate conversion.
  • Step S130 pre-processing the angular velocity signal and the plantar pressure signal.
  • the inertial sensor signal i.e., angular velocity signal
  • the inertial sensor signal has high-frequency noise, and the presence of noise will seriously interfere with the subsequent gait analysis results. Therefore, preferably, these noises need to be processed.
  • the frequency of the gait data signal is mainly concentrated below 10Hz.
  • high-frequency noise is filtered out at the same time, for example, a 2nd-order Butterworth low-pass filter with a cutoff frequency of 10Hz is used to filter the signal.
  • the filtered signal is reversely filtered again to achieve zero phase offset and reduce local noise.
  • the angular velocity signal before and after filtering is shown in Figure 8.
  • the plantar pressure signal can also be preprocessed to filter out noise to improve the accuracy of subsequent analysis.
  • Step S140 Detect different gait events based on the plantar pressure signal and determine the occurrence time of each gait event.
  • the plantar pressure sensor will detect the pressure value, and the gait event can be judged by setting the threshold Th.
  • Th For example, for the i-th sampling point, when the corresponding pressure value T(i) is less than the threshold Th, it is considered that this area is not in contact with the ground, that is, the closed state; when the pressure value T(i) is greater than or equal to the threshold Th, it is considered that this area is in contact with the ground, that is, the open state, expressed as:
  • S(i) represents the on/off state of the i-th sampling point
  • T(i) represents the pressure value of the i-th sampling point
  • off-ground means not in contact with the ground
  • on-ground means in contact with the ground.
  • the threshold Th is determined according to the following steps:
  • l is the number of gait cycles
  • Max k is the maximum value of the pressure signal in the kth gait cycle
  • Min k is the minimum value of the pressure signal in the kth gait cycle
  • Th MAX and Th MIN are the average maximum and average minimum values of the plantar pressure signals of all gait cycles.
  • the threshold for judging the on and off state of each feature area can be obtained by the average maximum value and the average minimum value, which can be uniformly expressed as:
  • Th ThMIN + ⁇ ( ThMAX - ThMIN ) (7)
  • the coefficient ⁇ can adjust the threshold Th to compensate for the difference in pressure levels of subjects with different weights.
  • ⁇ is set to 0.05.
  • the threshold Th is calculated by the average of the maximum and minimum values of the plantar pressure in each gait cycle.
  • the plantar pressure signals of the four characteristic areas screened out are Ta (i), Tb (i), Tc (i) and Td (i), and the thresholds for determining whether the characteristic areas are in contact with the ground are Tha , Thb , Thc and Thd . Then, four gait events are determined according to the contact between different areas of the plantar and the ground.
  • the point at which the pressure amplitude begins to be greater than the threshold value Th is taken as the time when the HS event occurs.
  • FF events refer to the moment when the entire sole of the foot touches the ground or at least the metatarsals touch the ground.
  • the heel already has a pressure signal, and it is necessary to judge based on the pressure signal of the metatarsals.
  • this time point is considered to be the FF time feature point.
  • the point at which the pressure amplitude begins to be less than the threshold value Th is taken as the time when the HO event occurs.
  • the point at which the pressure amplitude begins to be less than the threshold value Th is taken as the time when the TO event occurs.
  • step S140 according to the different force conditions of different parts of the sole of the foot in different gait events, the sole pressure information of four characteristic areas is selected, and a gait event detection method based on threshold analysis is proposed, thereby completing the marking of the occurrence time of different gait events.
  • Step S150 mapping the gait events detected based on the plantar pressure signal onto the angular velocity signal, thereby mining the significant features in the angular velocity signal curve.
  • the y-axis angular velocity signal of the gyroscope placed above the metatarsal bones of the forefoot takes the moment of the heel-touching event as the starting point of the gait cycle.
  • the angular velocity signal has a double trough feature and an obvious flat area.
  • Figure 10 shows the corresponding positions of the plantar pressure gait event detection results on the angular velocity curve.
  • the two troughs of the double trough feature are located in front and behind the flat area, respectively.
  • the time point of the HS event corresponds to the trough in front of the flat area in the angular velocity curve.
  • the time point corresponding to the TO event corresponds to the trough behind the flat area in the angular velocity curve.
  • the angular velocity amplitude and change rate in the flat area are close to zero, and FF and HO are the starting point and end point of the flat area, respectively.
  • the gait events in the angular velocity signal are time-series labeled, which can mine the significant features of the inertial sensing information under different gait events.
  • Step S160 detecting gait events by identifying significant features of the angular velocity signal curve.
  • step S160 the gait event point detection of the angular velocity of the inertial sensor is achieved by identifying the double trough feature and the flat area feature of the angular velocity curve.
  • two inertial sensors are used to collect angular velocity signals of the left and right feet respectively during walking, and then gait events are detected based on the significant features of the angular velocity signal curve.
  • the angular velocity signal has two troughs and a flat area. Since the angular velocity and the angular velocity change rate in the flat area are close to zero, the range of the flat area can be identified by setting a threshold.
  • the absolute value M and the variance S of the sliding average are obtained according to formula (8).
  • W is the window size, for example, it is set to 10
  • N is the total number of sampling points in the gait sampling process
  • yi is the gyroscope y-axis angular velocity data
  • Mi and Si are the sliding mean and variance of the window on yi
  • ⁇ M and ⁇ S are the thresholds of the sliding mean and variance
  • the step size of the sliding window is 1.
  • the sliding mean Mi is less than the threshold ⁇ M and the sliding variance Si is less than ⁇ S
  • the threshold ⁇ M is set to 15 and the threshold ⁇ S is set to 200.
  • the flat area detection result is shown in Figure 11.
  • the FF event is the starting point of the flat area
  • the HO event is the end point of the flat area.
  • the angular velocity signal there are two troughs in each gait cycle.
  • the first minimum point before the flat area, and the angular velocity value is less than 0, corresponds to the HS event; the first minimum point after the flat area, and the angular velocity value is less than 0, corresponds to the HO event.
  • the gait cycle of normal walking is about 1 to 1.32s, the same gait event needs to meet a certain time difference.
  • the gait cycle is known to be T, and the time difference is set to 0.8T.
  • HS and HO events can be obtained by judging the minimum point, angular velocity and time difference.
  • Figure 12 is a gait event detection result based on the angular velocity signal significant feature detection.
  • an inertial sensor information significant feature recognition method based on double trough detection and flat area detection method is established to realize gait event detection based on inertial sensor.
  • Step S170 dividing the detected gait events into time phases.
  • a gait cycle is from one heel touchdown to the next heel touchdown.
  • the gait cycle of a unilateral lower limb can be divided into four phases by the four gait events of HS, FF, HO, and TO. As shown in Figure 13, the definitions of the four gait phases are as follows:
  • Loading Response The phase between gait events HS and FF, accounting for about 10% of the time;
  • MS Mid Stance
  • Terminal Stance The phase between gait events HO and TO, accounting for about 20% of the time;
  • Swing phase The phase between gait events TO and HS, accounting for about 40% of the time.
  • the HS event is set as the starting point of a gait cycle
  • the LR phase is the first phase of the gait cycle.
  • the LR phase is considered to start. Then, the four gait events are retrieved in sequence on the angular velocity curve in chronological order.
  • the time phase division includes:
  • the current state is MS phase. If a HO event is detected next time, the gait phase is switched from the current MS phase to the TS phase. If no gait event is detected next time, the current gait phase is maintained unchanged.
  • the current state is TS phase. If a TO event is detected next time, the gait phase is switched from the current TS phase to the SW phase. If no gait event is detected next time, the current gait phase is maintained unchanged.
  • the current state is SW phase. If a HS event is detected next time, the gait phase is switched from the current SW phase to the LR phase. If no gait event is detected next time, the current gait phase is maintained unchanged.
  • Step S180 based on the time sequence characteristics of the normal gait and the typical abnormal gait, the gait phase is evaluated to achieve abnormal gait detection.
  • the gait time parameters of each gait cycle can be calculated. Assuming the time points of the four gait events in the kth cycle are HS(k), FF(k), HO(k), and TO(k), and HS is set as the starting point of each gait cycle, the gait cycle T can be expressed as:
  • the duration of the LR phase T LR is:
  • T MS duration of the MS phase
  • the duration of the TS phase T TS is:
  • T TS (k) TO (k) - HO (k) (13)
  • the duration of the SW phase T SW is:
  • T SW (k) HS (k + 1) - TO (k) (14)
  • the evaluation of the time dimension of gait quality is completed. That is, for the detected gait events, the division of gait phases and the detection of abnormal gait are achieved based on the time sequence characteristics of normal gait and typical abnormal gait.
  • an experiment was conducted. Six healthy subjects with athletic ability were recruited for the experiment. Each subject performed 5 experiments, with a total of 30 experimental data. The data acquisition platform is shown in Figure 16. Before the test, the subject's height, weight, foot length, age, gender and other individual basic information are measured and recorded. During the experiment, the subject wore special shoes for the experiment, and the sole was placed on the sole. The plantar pressure measurement system was placed on the sole, and the inertial sensor was fixed at the instep for synchronous collection. Among them, the plantar pressure measurement system has a sampling rate of 50Hz for the plantar pressure signal, and the inertial sensor has a sampling rate of 120Hz for the angular velocity signal. In addition, an experiment simulating abnormal gait was designed: the right legs of the two subjects were fixed with knee joint limiters to limit their lower limb motor ability, and the inertial sensor was used alone to record gait parameters at the same sampling rate.
  • Table 1 shows the difference between the gait event detection results based on the angular velocity signal and the gait event detection results based on the plantar pressure signal, given in the form of mean time difference ⁇ standard deviation of the time difference (Mean Difference ⁇ Standard Deviation).
  • Figure 17 is a comparison of the four gait events on the angular velocity curve and the plantar pressure signal. It can be seen from Table 1 that the time difference between the gait event detection result and the reference result is about 20ms, and the detection result is relatively accurate.
  • Table 2 is the duration of the gait phase of the healthy group
  • Table 3 is the duration comparison of the gait phase of the healthy group and the abnormal group. It can be concluded from Table 2 that the gait cycle of the healthy group is about 1.20s, the swing phase accounts for 41.44%, and the support phase accounts for 58.56%, wherein the load-bearing reaction period (LR) accounts for 10.81%, the mid-support phase (MS) accounts for 30.04%, and the terminal support phase accounts for 17.71%.
  • the experimental results are approximately consistent with the phase ratios provided in the data, verifying the accuracy and reliability of the gait phase detection method provided by the present invention.
  • the present invention proposes a simple and reliable gait event labeling method. According to the different contact conditions between the sole of the foot and the ground, a gait event labeling method based on the sole pressure information collected synchronously with the inertial sensor is proposed. Different detection thresholds are set according to the contact conditions between the sole of the foot and the ground under different gait events, so as to achieve accurate labeling of the occurrence time of the four gait events, and provide a reference standard for gait event detection based on the angular velocity of the inertial sensor.
  • the present invention explores new features for gait event detection based on angular velocity sensor information. Based on the time series information of gait events detected by plantar pressure, the present invention explores the significant features corresponding to specific gait events in the angular velocity sensor signal, including local minima and flat areas.
  • the present invention proposes a gait event detection method based on angular velocity, designs dual trough detection and flat area detection algorithms, establishes a gait event recognition method based on the above-mentioned significant feature recognition, and realizes automatic recognition of gait events based only on angular velocity sensors.
  • the present invention proposes a gait phase division method based on prior knowledge and gait events. On the basis of gait event detection, a gait phase division method is established based on the state transition timing information of normal gait and different types of abnormal gaits.
  • the present invention detects gait events based on plantar pressure data and uses it as a reference to mine the significant features corresponding to gait events in the angular velocity sensor signal, and accurately identifies four gait events and four gait phases through feature recognition methods.
  • the experimental results show that the error of the gait event detection result is 20ms compared with the gold standard time of gait time based on pressure, and the proportion of each gait phase is consistent with the proportion of each phase in the previous research results.
  • the present invention collects gait data of a human body when walking based on an angular velocity sensor, and realizes real-time detection of gait events and division of time phases based on waveform morphological feature detection, with low detection cost and simple calculation;
  • the present invention evaluates the gait in the time dimension according to the objective laws of human gait, combined with the results of gait event detection and gait phase recognition.
  • the evaluation results show that there is a significant difference in the duration of gait phases between normal gait and abnormal gait, and the abnormal gait can be effectively detected by gait time parameters.
  • the present invention may be a system, a method and/or a computer program product.
  • the computer program product may include a computer-readable storage medium carrying computer-readable program instructions for causing a processor to implement various aspects of the present invention.
  • Computer readable storage medium can be a tangible device that can keep and store the instructions used by the instruction execution device.
  • Computer readable storage medium can be, for example, but not limited to, an electrical storage device, a magnetic storage device, an optical storage device, an electromagnetic storage device, a semiconductor storage device, or any suitable combination thereof.
  • Non-exhaustive list of computer readable storage medium include: a portable computer disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), a static random access memory (SRAM), a portable compact disk read-only memory (CD-ROM), a digital versatile disk (DVD), a memory stick, a floppy disk, a mechanical encoding device, for example, a punch card or a convex structure in a groove on which instructions are stored, and any suitable combination thereof.
  • RAM random access memory
  • ROM read-only memory
  • EPROM or flash memory erasable programmable read-only memory
  • SRAM static random access memory
  • CD-ROM compact disk read-only memory
  • DVD digital versatile disk
  • memory stick a floppy disk
  • mechanical encoding device for example, a punch card or a convex structure in a groove on which instructions are stored, and any suitable combination thereof.
  • Computer readable storage medium used here is not interpreted as an instantaneous signal itself, such as a radio wave or other freely propagating electromagnetic waves, an electromagnetic wave propagated by a waveguide or other transmission medium (for example, a light pulse by an optical fiber cable), or an electrical signal transmitted by a wire.
  • These computer-readable program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, or other programmable data processing device, thereby producing a machine, so that when these instructions are executed by the processor of the computer or other programmable data processing device, a device that implements the functions/actions specified in one or more boxes in the flowchart and/or block diagram is generated.
  • These computer-readable program instructions can also be stored in a computer-readable storage medium, and these instructions cause the computer, programmable data processing device, and/or other equipment to work in a specific manner, so that the computer-readable medium storing the instructions includes a manufactured product, which includes instructions for implementing various aspects of the functions/actions specified in one or more boxes in the flowchart and/or block diagram.
  • each square frame in the flow chart or block diagram can represent a part for a module, program segment or instruction.
  • the function marked in the square frame can also occur in a sequence different from that marked in the accompanying drawings.
  • two continuous square frames can actually be performed substantially in parallel, and they can also be performed in reverse order sometimes, depending on the function involved.

Abstract

本发明公开一种基于角速度传感器的步态时相识别方法。该方法包括：利用传感器采集目标走路的原始角速度信号并同步采集足底压力信号；将所述原始角速度信号从传感器坐标系变换到地理坐标系，获得对应的角速度信号，其中所述传感器坐标系是由传感器自身的三个正交测量轴所定义的坐标系；基于所述足底压力信号检测步态事件并确定各类步态事件的发生时刻；将基于所述足底压力信号检测的步态事件映射到所述角速度信号上，挖掘所述角速度信号的显著性特征；通过识别所述角速度信号的显著性特征，检测步态事件并对步态时相进行划分，进而获得识别结果。本发明能够有效识别不同的步态事件及包含的时相，并且计算简单，识别准确率高，成本低。

Description

一种基于角速度传感器的步态时相识别方法 技术领域
本发明涉及计算机应用技术领域，更具体地，涉及一种基于角速度传感器的步态时相识别方法和系统。
背景技术
行走是人类最基本的运动功能，行走过程的身体姿态(即步态)包含了大量的人体运动学信息。这种看似简单却蕴含复杂运动学、动力学和生物力学的运动过程，要求肌肉系统，骨骼系统及神经系统的高度协调。正常的步态是人体中枢神经系统控制骨骼、肌肉等人体部位完成的一项活动，它具有一定的稳定性、协调性、周期性、方向性以及个体差异性，任何一个环节出现异常或失调都会导致步态异常。导致人体步态异常的因素包括先天遗传、后天疾病以及意外伤害等。长时间以异常步态行走，不仅会严重影响生活质量，还会导致步态的异常程度加重，表现为病态的步行姿势，甚至丧失步行的能力。因此，对异常步态的检测和识别具有十分重要的意义。
步态具有周期性和规律性，行走中每个步态周期都包含一系列足姿位的转移。人们通常依据典型足姿位的变化划分出一系列时段，称之为步态时相。步态时相对步态异常的分析和检测具有重要作用。相比正常步态，异常步态表现为时相缺失、时序混乱、比例失调等。因此，步态时相是反映步态习惯、年龄、健康状况等影响身体协调性因素的重要指标。对于步态分析系统，步态时相识别是步态周期提取、步态相位分割和步态时空参数计算的基础。步态分析系统必须能够准确、可靠地检测出步态时相，才能满足临床应用的需要。
根据所使用的感知方法，目前识别步态时相的研究主要分为三类：基于机器视觉的方法，基于压力传感器的方法和基于可穿戴惯性传感器的方 法。基于机器视觉的方法是利用高精度摄像机和图像处理技术，追踪实验者身上高反射标志点的运动轨迹，再经过相应算法计算出固定部位的位置信息，获取人体步态的特征参数。基于压力传感器的方法是通过地面铺设或内置鞋底的压力传感器测量走路过程中的地面反作用力来进行步态分析。基于可穿戴惯性传感器的方法利用佩戴于身体不同位置的惯性传感器提取步态信号，经过坐标转换及姿态角解算，通过对加速度信号、角速度信号及磁力计的综合运算，计算得到步态特征。
在现有技术中，专利申请CN202210185789.7提供一种基于Kinect的人体步态参数提取方法，是基于机器视觉的方法。该方案通过Kinect体感深度传感器提取关节点位置序列数据，对数据进行滤波处理并计算步态参数值。根据步态参数值构造步态特征空间，采用基于随机森林模型的步态相位划分策略得到每帧单侧状态的初步判定结果，再对异常相位进行误判修正得到最终步态相位划分结果。该方案具有较全面的步态特征表示，但是受Kinect深度传感器检测范围的影响，数据采集需要在特定的场地，运动范围有限；并且图像信息容易受光照、遮挡物、背景环境的影响。
专利申请CN202010202430.7提供一种步态健康评估方法及装置，通过8路足底压力传感器获取步态压力信号，基于步态压力信号采用模糊逻辑推理规则识别步态相位。该方法具有便携、动态采集和精确反映足底压力变化情况的优点，但是传感器只能获取支撑相的步态信号，单独使用时无法体现完整步态信息，并且放置在足底的位置对压力测量准确性有很大影响，需要根据不同受试者脚的大小调整传感器的位置。
专利申请CN202111029801.7提出了一种基于角速度信号的步态事件点检测方法，通过采集小腿的角速度数据，将特征点附近的角速度与预设阈值相比较，来判断人走路时的脚跟着地时间点和脚尖离地时间点，并以此划分步态周期。该方法使用方便，计算简单，但是检测的步态事件数量较少，如果需要对步态进行分析，需要对步态周期划分更多的阶段。
在上述现有方案中，基于机器视觉的方法通常部署在专业的步态分析实验室中，成本较高，需要专业技术人员操作，且容易受到光，遮挡，距离等的限制，尚未得到广泛应用。基于压力传感器的方法优势在于能够精 确显示足底压力变化情况，通常作为步态检测的金标准，但是此方法只能呈现行走过程中足部与地面接触时的压力变化，而在足部腾空的过程中，压力信号为零，无法得到完整步态周期的步态信息，并且压力传感器的位置对检测准确性有很大的影响，由于存在机械损耗，足底压力传感器不适合长时间连续采集。基于可穿戴惯性传感器的方法与视频图像、压力传感器相比具有很多优点，它可以固定在人体的各个部位，因此可以获取完整的步态信息，也可以在任意环境中工作，不受光照、天气、阴影及遮挡的影响，无需存储大量的视频图像信息，便携性好、穿戴方便。随着微机电系统(Micro-Electro-Mechanical，MEMS)技术的快速发展，可穿戴惯性传感器逐渐成为步态分析研究中的重要工具，但目前这种方式的步态识别精确度还有待提高。
发明内容
本发明的目的是克服上述现有技术的缺陷，提供一种基于角速度传感器的步态时相识别方法。该方法包括：
利用传感器采集目标走路的原始角速度信号并同步采集足底压力信号；
将所述原始角速度信号从传感器坐标系变换到地理坐标系，获得对应的角速度信号，其中所述传感器坐标系是由传感器自身的三个正交测量轴所定义的坐标系；
基于所述足底压力信号检测步态事件并确定各类步态事件的发生时刻；
将基于所述足底压力信号检测的步态事件映射到所述角速度信号上，挖掘所述角速度信号的显著性特征，该显著性特征表明，在每个步态周期中，所述角速度信号具有双波谷特征和一段平坦区域；
通过识别所述角速度信号的显著性特征，检测步态事件并对步态时相进行划分，进而获得识别结果。
与现有技术相比，本发明的优点在于，提供一种基于角速度传感器的步态时相识别方法，利用3轴角速度传感器采集人体步行过程中脚面位置 的角速度数据，通过挖掘不同步态事件对应的惯性传感器信息的显著性特征，建立基于角速度的步态显著性特征识别方法，实现对不同步态事件、不同步态时相的准确识别。此外，本发明是一种基于可穿戴惯性传感器的步态时相识别方法，具有计算简单，识别准确率高，成本低，可以长期连续监测步态特征长时间维度变化等优点。
通过以下参照附图对本发明的示例性实施例的详细描述，本发明的其它特征及其优点将会变得清楚。
附图说明
被结合在说明书中并构成说明书的一部分的附图示出了本发明的实施例，并且连同其说明一起用于解释本发明的原理。
图1是根据本发明一个实施例的基于角速度传感器的步态时相识别方法的流程图；
图2是根据本发明一个实施例的基于角速度传感器的步态时相识别方法的过程示意图；
图3是根据本发明一个实施例的足底受力区域的示意图；
图4是根据本发明一个实施例的原始惯性传感器信号和足底压力信号的示意图；
图5是根据本发明一个实施例的传感器坐标转换为地理坐标的示意图；
图6是根据本发明一个是实施例的角速度信号坐标转换前后的对比图；
图7是根据本发明一个是实施例的角速度信号和频谱图；
图8是根据本发明一个实施例的滤波前后角速度信号的示意图；
图9是根据本发明一个实施例的阈值算法检测步态事件的示意图；
图10是根据本发明一个实施例的基于足底压力的步态事件检测结果的示意图；
图11是根据本发明一个实施例的平坦区域划分示意图；
图12是根据本发明一个实施例的基于角速度的步态事件检测结果示 意图；
图13是根据本发明一个实施例的步态相位划分示意图；
图14是根据本发明一个实施例的正常人步态时相持续时间示意图；
图15是根据本发明一个实施例的模拟异常人步态相位持续时间示意图；
图16是根据本发明一个实施例的实验数据平台示意图；
图17是根据本发明一个实施例的步态事件检测结果与参考结果对比图；
附图中，Angular Velocity-角速度；Amplitude-幅值；Samples-采样点；Sampling Rate-采样率；Gyroscope-陀螺仪。
具体实施方式
现在将参照附图来详细描述本发明的各种示例性实施例。应注意到：除非另外具体说明，否则在这些实施例中阐述的部件和步骤的相对布置、数字表达式和数值不限制本发明的范围。
以下对至少一个示例性实施例的描述实际上仅仅是说明性的，决不作为对本发明及其应用或使用的任何限制。
对于相关领域普通技术人员已知的技术、方法和设备可能不作详细讨论，但在适当情况下，所述技术、方法和设备应当被视为说明书的一部分。
在这里示出和讨论的所有例子中，任何具体值应被解释为仅仅是示例性的，而不是作为限制。因此，示例性实施例的其它例子可以具有不同的值。
应注意到：相似的标号和字母在下面的附图中表示类似项，因此，一旦某一项在一个附图中被定义，则在随后的附图中不需要对其进行进一步讨论。
本发明通过同步采集的足底压力信息，提出了基于角速度信号的步态事件识别方法，并基于足底压力检测的步态事件，对角速度信号进行标注；进一步，通过时、频分析方法挖掘角速度信号在不同步态事件下的显著性特征，建立基于平坦区域检测和双波谷检测的步态事件显著特征识别算法， 基于角速度信号实现对不同步态事件的准确检测；最后，根据检测出的步态事件及先验知识将步态周期划分为多个相位，实现步态时相的准确划分。
结合图1和图2所示，所提供的基于角速度传感器的步态时相识别方法包括以下步骤。
步骤S110，同步采集角速度信号和足底压力信号。
现有的步态事件检测方法需要在安装高速动捕设备的实验室对步态事件进行标注，在本发明一个实施例中，提出了基于足底压力传感信息的步态事件检测方法，因而在建模数据采集过程中需要同步采集角速度传感器信息和足底压力信息。
例如，采用可穿戴设备Xsens中内置的惯性传感单元(Inertial Measurement Unit，IMU)来获取走路过程中的角速度信号，Xsens的采样频率设为120Hz。采用足底压力分布测量系统FS-INS-W99来获取运动时的足底压力信号，采样频率设为50Hz，实验数据离线采集存储并用于后续建模分析。两个Xsens传感器通过绑带或绝缘胶带分别固定在受试者左右的鞋面上，足底压力分布测量系统以鞋垫的方式固定于受试者的鞋内。
足底压力分布测量系统FS-INS-W99具有99个压力感应区域，所采集的99通道压力数据冗余度较高，不便于后续分析，需要进行前期的数据筛选。
具体地，结合实验数据与步态生物力学的理论分析，人体站立或运动时，体重通过骨骼转移到脚上，脚上的骨骼向地面施力，脚底受力主要有四个位置：脚趾、第一跖骨、第五跖骨和脚跟，分别对应压力分布测量系统的四个区域，参见图3所示的(a)～(d)四个区域。因此，在一个实施例中，筛选出与足底压力鞋垫上这四个特征区域相对应的足底压力数据进行后续分析。图4是惯性传感器采集的原始角速度信号和所筛选出的四个特征区域的足底压力原始信号的示意图。
步骤S120，将角速度信号从传感坐标系变换到地理坐标系。
参见图5所示，坐标系转换主要涉及地理坐标系和传感器坐标系。
对于地理坐标系(Ground Frame，G系)，可选用“东-北-天”坐标系，该坐标系是静止不动的。该坐标系以地球表面的载体质心为原点，其 中一个坐标轴指向地心的反方向，其余两个坐标轴分别与所在区域的经线、纬线的切线方向平行，即x轴正方向指向东，y轴正方向指向北，z轴正方向指向重力向量g的反方向。
对于传感器坐标系(Sensor Frame，S系)，是由惯性传感器(或称惯性测量单元)自身的三个正交测量轴所定义的坐标系。惯性传感器的输出测量值均以传感器坐标系为参考系。例如，定义传感器坐标系的原点与地理坐标系的原点重合；在传感器坐标系上，x轴信号输出测量值反映步行过程中前后方向上的惯性数据；y轴信号输出测量值反映步行过程中左右方向上的惯性数据；z轴信号输出测量值反映步行过程中竖直方向上的惯性数据。
在步行过程中，由于惯性传感器可采用鞋绑式，每次实验中固定的位置无法达到完全相同的理想情况，因此传感器坐标系与地理坐标系的相对位置是时刻变化的，无法确保惯性传感器的z轴始终处于重力加速度的方向上，该感知轴与竖直方向的重力加速度之间存在一定的夹角，导致测量的惯性数据产生误差。如图5所示，两个坐标系之间存在不同的坐标轴指向，因此需要进行S系到G系的坐标系变换，得到在参考坐标系中具有实际意义的物理量。
描述两个坐标系之间关系的常用方法主要有欧拉角法、方向余弦矩阵法、三角函数法和四元数法。使用欧拉角法进行坐标转换后，得到的姿态矩阵不需要进行正交化处理，但是当载体的俯仰角为±90°时，将出现奇异点，导致丢失一个自由度，因此该方法不能进行全姿态解算，其使用存在一定的局限。方向余弦法用矢量的方向余弦来表示姿态矩阵的方法。该方法求解姿态矩阵时避免了欧拉角法所遇到的奇异点问题。但方向余弦矩阵具有九个元素，所以需要解九个微分方程，计算工作量较大，在工程上并不实用。三角函数法是将绕定点转动的两个坐标系之间的关系用三次转动等效地表示，用三次转动角度的正、余弦函数来表示姿态函数。该方法同样可以避免奇异点问题，但是也需要解六个微分方程，计算量较大。四元数法需要求解四个微分方程，该方法计算量小、精度高、可避免奇异性，并提供平滑插值。通过从不同角度对欧拉角法、方向余弦法、三角函数法 和四元数法进行对比，结果表明四元数法具有最佳的性能。因此，优选地，采用四元数法进行坐标系转换。
具体地，四元数由一个实部，三个虚部组成。一个单位四元数复数式表示形式可写为：
Figure PCTCN2022128174-appb-000001
其中q 0是实部，代表旋转角度；若q 0＝0，则为纯四元数；q 1i+q 2j+q 3k是虚部，描述了三维空间中的旋转轴。
在进行坐标系转换时，需要建立表示地理坐标系与传感器坐标系之间相对位置的矩阵，准确描述地理坐标系中三维空间向量与传感器坐标系中三维空间向量之间的转换关系，如公式(2)和公式(3)所示。利用四元数可以求解出旋转矩阵R，再用旋转矩阵乘以原始传感器的数据，就可以得到地理坐标系参数。
Figure PCTCN2022128174-appb-000002
Figure PCTCN2022128174-appb-000003
在公式(3)中，R为旋转矩阵，
Figure PCTCN2022128174-appb-000004
为传感器坐标系下的原始数据，
Figure PCTCN2022128174-appb-000005
为转换后的地理坐标系参数。
图6是惯性传感器角速度信号坐标转换前后的对比图。
步骤S130，对角速度信号和足底压力信号进行预处理。
通过对惯性传感器信号(即角速度信号)的初步分析，发现惯性传感器信号具有高频噪声，噪声的存在会严重干扰后续步态分析结果，因此，优选地，需要对这些噪声进行处理。参见图7所示的角速度信号和对应的频谱图，通过对惯性传感器信号进行频域分析，步态数据信号的频率主要集中在10Hz以下。为了尽可能地保留有用信息不使数据失真，同时滤除高 频噪音，例如使用一个截止频率为10Hz的2阶巴特沃斯低通滤波器对信号进行滤波。但是由于一次滤波会导致信号相位的移动，影响步态时相识别的准确率，因而对滤波后的信号再进行一次反向滤波，实现零相位偏移，并且降低局部噪声。滤波前后的角速度信号如图8所示。类似地，也可通过对足底压力信号进行预处理，滤除噪声，以提升后续分析的准确性。
步骤S140，基于足底压力信号检测不同的步态事件并确定各类步态事件的发生时刻。
在一个完整的步态周期中，根据足底与地面的接触情况，存在四种关键步态事件，分别为脚跟撞击(Heel-Strike，HS)、脚掌着地(Foot-Flat，FF)、脚跟离地(Heel-Off，HO)、脚趾离地(Toe-Off，TO)。足底与地面接触时，足底压力传感器会检测到压力值，可以通过设置阈值Th来判断步态事件。例如，对于第i个采样点，当对应的压力值T(i)小于阈值Th时，认为此区域没有与地面接触，即关闭状态；当压力值T(i)大于等于阈值Th时，认为此区域与地面接触，即开启状态，表示为：
Figure PCTCN2022128174-appb-000006
其中，S(i)表示第i个采样点的开、关状态，T(i)表示第i个采样点的压力值大小，off-ground表示不与地面接触，on-ground表示与地面接触。
在一个实施例中，根据以下步骤确定阈值Th：
首先，针对所有步态周期，计算足底压力信号的平均最大值和平均最小值，表示为：
Figure PCTCN2022128174-appb-000007
Figure PCTCN2022128174-appb-000008
其中，l是步态周期数，Max k是第k个步态周期中压力信号的最大值，Min k是第k个步态周期中压力信号的最小值，Th MAX和Th MIN是所有步态周期的足底压力信号的平均最大值和平均最小值。
接下来，可以通过平均最大值和平均最小值获得每个特征区域判断开启与关闭状态的阈值，统一表示为：
Th＝Th MIN+α(Th MAX-Th MIN)  (7)
其中，系数α可以调整阈值Th的大小，以补偿体重不同的受试者压力水平的差异。例如，α取值为0.05。由公式(7)可知，该阈值Th通过足底压力的每个步态周期中的最大值和最小值的平均值计算得到。
基于上述过程，筛选出的四个特征区域的足底压力信号分别为T a(i)、T b(i)、T c(i)和T d(i)，计算特征区域与地面是否接触的判别阈值分别为Th a、Th b、Th c和Th d。进而，根据足底不同区域与地面接触情况确定四种步态事件。
具体地，参见图9所示，对于HS事件，脚跟的压力信号幅值处于增加阶段时，压力幅值开始大于阈值Th的点，作为HS事件的发生时刻。对于FF事件，FF事件是指整个足底着地或者至少跖骨着地的瞬间，此时脚跟已经有压力信号，需要根据跖骨的压力信号来判断，两个跖骨区域中任意一个压力信号幅值开始大于Th，该时间点就认为是FF时间特征点。对于HO事件，脚跟的压力信号幅值处于减小阶段时，压力幅值开始小于阈值Th的点，作为HO事件的发生时刻。对于TO事件，脚趾的压力信号幅值处于减小阶段时，压力幅值开始小于阈值Th的点，作为TO事件的发生时刻。
在该步骤S140中，根据不同步态事件下足底不同部位受力情况的不同，选取四个特征区域的足底压力信息，提出一种基于阈值分析的步态事件检测方法，进而完成不同步态事件发生时刻的标注。
步骤S150，通过将基于足底压力信号检测的步态事件映射到角速度信号上，挖掘出角速度信号曲线中的显著性特征。
通过将基于足底压力信息检测到的步态事件映射到角速度信号上，发现放置在前脚掌跖骨上方的陀螺仪y轴角速度信号，在沿直线向前行走时，以脚跟着地事件的时刻作为步态周期的开始点，在每个步态周期中，角速度信号具有双波谷特征和一段明显的平坦区域。
图10是足底压力步态事件检测结果在角速度曲线上的对应位置，通过对比发现，双波谷特征的两个波谷分别位于平坦区域的前方和后方，HS事件的时间点对应角速度曲线中平坦区域前方的波谷。TO事件对应的时间点对应角速度曲线中平坦区域后方的波谷。平坦区域的角速度幅值和变化率都接近零，FF和HO分别是平坦区域的起点和终点。通过这种方式，对 角速度信号中的步态事件进行时序标注，能够挖掘不同步态事件下惯性传感信息中的显著性特征。
步骤S160，通过识别角速度信号曲线的显著性特征检测步态事件。
在该步骤S160中，通过识别角速度曲线的双波谷特征和平坦区域特征，实现对惯性传感器角速度的步态事件点检测。
例如，使用两个惯性传感器分别采集行走时左右脚的角速度信号，进而基于角速度信号曲线显著性特征来检测步态事件检测。
1)平坦区域特征识别
在每个步态周期中，角速度信号具有两个波谷和一段平坦区域，由于平坦区域的角速度和角速度变化率都接近零，可以利用设置阈值的方法识别出平坦区域的范围。
在一个实施例中，针对角速度信号，按照公式(8)求解取滑动平均值的绝对值M和方差S。
Figure PCTCN2022128174-appb-000009
Figure PCTCN2022128174-appb-000010
在公式(8)中，W为设置窗口大小，例如设置为10，N为步态采样过程中总的采样点数，y i为陀螺仪y轴角速度数据，M i和S i分别为窗口在y i上的滑动均值和方差，γ M和γ S分别为滑动均值和方差的阈值，滑动窗口的步长为1，当滑动均值M i小于阈值γ M，且滑动方差S i小于γ S时，就判断其为平坦区域，例如阈值γ M设置为15，阈值γ S设置为200。平坦区域检测结果如图11所示。在每个步态周期中，FF事件为平坦区域的起点；HO事件为平坦区域的终点。
2)双波谷特征识别
在角速度信号中，每个步态周期中有两个波谷，平坦区域前出现的第一个极小值点，且角速度值小于0，对应可得到HS事件；平坦区域后出现的第一个极小值点，且角速度值小于0，对应可得到HO事件，由于正常行走的步态周期约为1～1.32s，相同的步态事件需要符合一定的时间差，已知步态周期为T，将时间差设定为0.8T。通过判断极小值点、角速度大小和 时间差可以得到HS和HO事件。
图12是基于角速度信号显著性特征检测得出的步态事件检测结果。通过这种方式，建立基于双波谷检测和平区检测方法的惯性传感器信息显著性特征识别方法，实现基于惯性传感器的步态事件检测。
步骤S170，对于所检测出的步态事件进行时相划分。
从一次足跟触地到下一次足跟触地为一个步态周期。为了便于对正常和异常步态的分析，可以由HS、FF、HO、TO四个步态事件将单侧下肢的步态周期划分为四个时相。如图13所示，四个步态时相的定义如下：
承重反应期(Loading Response，LR)：步态事件HS到FF之间的阶段，时间占比约为10％；
支撑相中期(Mid Stance，MS)：步态事件FF到HO之间的阶段，时间占比约为30％；
支撑相末期(Terminal Stance，TS)：步态事件HO到TO之间的阶段，时间占比约为20％；
摆动相(Swing，SW)：步态事件TO到HS之间的阶段，时间占比约为40％。在一个实施例中，将HS事件设为一个步态周期的起始点，LR相位为步态周期的第一个相位，当第一次检测到HS事件时，认为LR相位开始。然后按照时间顺序在角速度曲线上依次检索四种步态事件。
具体地，时相划分包括：
1)当前状态是LR相位，则如果下一次检测到FF事件，则步态相位从当前的LR相位转换MS相位。如果下一次没有检测到任何步态事件，则维持当前的步态相位不变。
2)当前状态是MS相位，如果下一次检测到HO事件，则步态相位从当前的MS相位转换到TS相位。如果下一次没有检测到任何步态事件，则维持当前的步态相位不变。
3)当前状态是TS相位，如果下一次检测到TO事件，则步态相位从当前的TS相位转换到SW相位。如果下一次没有检测到任何步态事件，则维持当前的步态相位不变。
4)当前状态是SW相位，如果下一次检测到HS事件，则步态相位从 当前的SW相位转换到LR相位。如果下一次没有检测到任何步态事件，则维持当前的步态相位不变。
步骤S180，基于正常步态和典型异常步态的时序特征，评估步态时相，进而实现异常步态检测。
利用检测出的步态事件和步态时相，可以计算出每一个步态周期的步态时间参数。设第k个周期的四个步态事件的时间点为HS(k)、FF(k)、HO(k)、TO(k)，将HS设为每一个步态周期的起点，则步态周期T可以表示为：
T(k)＝HS(k+1)-HS(k)  (10)
LR相位的持续时间T LR为：
T LR(k)＝F(k)-HS(k)  (11)
MS相位的持续时间T MS为：
T MS(k)＝HO(k)-FF(k)  (12)
TS相位的持续时间T TS为：
T TS(k)＝TO(k)-HO(k)  (13)
SW相位的持续时间T SW为：
T SW(k)＝HS(k+1)-TO(k)  (14)
通过对上述四个步态时相的持续时间的计算，完成对步态质量时间维度的评估。即对于检测出的步态事件，基于正常步态和典型异常步态的时序特征，实现对步态时相的划分和异常步态的检测。
此外，设计了模拟异常步态的实验。将两位受试者的右腿用膝关节限位器固定，限制其下肢运动能力，单独用惯性传感器记录步态参数。对比同一受试者正常与模拟异常步态的相位持续时间如图14和图15所示。
由图可知，正常步态与异常步态在步态时相的持续时间上有显著的差异，正常步态下每一步的持续时间以及各个时相的时长呈现规律性，而在异常步态下，步与步之间的步态参数呈现散乱、无规律性且波动范围大。进一步地，基于正常步态和典型性异常步态的时序特征，对步态对称性、变异性及稳定性等步态特征进行分析，从而实现对异常步态的检测。
为了进一步验证本发明的效果，进行了实验。实验招募了6名身体健康的具有运动能力的受试者。每名受试者进行5次实验，总共30次实验数 据。数据采集平台如图16所示，测试前测量并记录受试者的身高、体重、脚长、年龄、性别等个体基本信息，实验过程中受试者穿着实验专用鞋，鞋底放置足底压力测量系统，脚面处固定惯性传感器同步采集。其中，足底压力测量系统对足底压力信号采样率为50Hz，惯性传感器对角速度信号采样率为120Hz。此外，设计了模拟异常步态的实验：将两位受试者的右腿用膝关节限位器固定，限制其下肢运动能力，单独用惯性传感器，以相同的采样率记录步态参数。
具体的实验结果如下。
以足底压力测量结果为参考，计算基于惯性传感器角速度的检测方法与参考结果的时间差值，差值越小，证明检测方法越准确。表1为基于角速度信号的步态事件检测结果与基于足底压力信号的步态事件检测结果的差值，以平均时间差±时间差的标准差(Mean Difference±Standard Deviation)的形式给出。图17为四种步态事件在角速度曲线和足底压力信号上的对比图。由表1可得，步态事件检测结果与参考结果时间差在20ms左右，检测结果较为准确。
表1步态事件检测结果与参考结果时间差(单位：ms)
Figure PCTCN2022128174-appb-000011
表2为健康组步态时相的持续时间，表3为健康组与异常组步态时相的持续时间对比。从表2可以得出，健康组的步态周期约1.20s，摆动相占41.44％，支撑相占58.56％，其中承重反应期(LR)占10.81％，支撑相中期(MS)占30.04％，支撑相末期占17.71％，实验结果与资料中提供的各 时相比例近似一致，验证了本发明提供的步态时相检测方法的准确性和可靠性。从表3可以得出，健康组左右两侧步态时相的百分比和持续时间一致，而异常组左右两侧的支撑相中期及摆动相的时间有明显差异，且步态周期更长，直观反映了其肢体的不对称性。可以通过步态时间参数有效地检测出异常步态。
表2健康组步态时间参数
Figure PCTCN2022128174-appb-000012
表3健康组与异常组步态时间参数对比
Figure PCTCN2022128174-appb-000013
综上所述，相对于现有技术，本发明具有以下技术效果：
1)本发明提出了一种简单、可靠的步态事件标注方法。根据足底与地面的接触情况的不同，提出了基于与惯性传感器同步采集的足底压力信息进行步态事件标注方法，针对不同步态事件下足底与地面的接触情况，设置不同的检测阈值，实现对四种步态事件发生时刻的准确标注，为基于惯性传感器角速度的步态事件检测提供参考标准。
2)本发明挖掘了基于角速度传感信息进行步态事件检测的新特征。基 于足底压力检测出来的步态事件的时序信息，挖掘出了角速度传感器信号中与特定步态事件对应的显著性特征，包括局部极小值和平坦区等。
3)本发明提出了基于角速度的步态事件检测方法，设计双波谷检测和平坦区检测算法，建立基于上述显著性特征识别的步态事件识别方法，实现了仅基于角速度传感器的步态事件自动识别。
4)本发明提出了基于先验知识和步态事件的步态时相划分方法。在步态事件检测的基础上，基于正常步态和不同类型异常步态的状态转移时序信息，建立步态时相划分方法。
5)本发明基于足底压力数据检测步态事件并以此作为参考，挖掘角速度传感器信号中步态事件相应的显著性特征，通过特征识别方法实现对4中步态事件和4种步态时相的准确识别。实验结果表明，步态事件检测结果与基于压力的步态时间金标准时间相比，误差在20ms，各步态时相比例与前期研究结果的各时相比例一致。
6)本发明基于角速度传感器采集人体走路时的步态数据，基于波形形态特征检测实现步态事件的实时检测和时相的划分，检测成本低且计算简单；
7)本发明根据人体步态的客观规律，结合步态事件检测结果和步态时相识别的结果，对步态进行时间维度的评估，评估结果得出，正常步态与异常步态在步态时相的持续时间上有显著的差异，可以通过步态时间参数有效地检测出异常步态。
本发明可以是系统、方法和/或计算机程序产品。计算机程序产品可以包括计算机可读存储介质，其上载有用于使处理器实现本发明的各个方面的计算机可读程序指令。
计算机可读存储介质可以是可以保持和存储由指令执行设备使用的指令的有形设备。计算机可读存储介质例如可以是但不限于电存储设备、磁存储设备、光存储设备、电磁存储设备、半导体存储设备或者上述的任意合适的组合。计算机可读存储介质的更具体的例子(非穷举的列表)包括：便携式计算机盘、硬盘、随机存取存储器(RAM)、只读存储器(ROM)、可擦式可编程只读存储器(EPROM或闪存)、静态随机存取存储器 (SRAM)、便携式压缩盘只读存储器(CD-ROM)、数字多功能盘(DVD)、记忆棒、软盘、机械编码设备、例如其上存储有指令的打孔卡或凹槽内凸起结构、以及上述的任意合适的组合。这里所使用的计算机可读存储介质不被解释为瞬时信号本身，诸如无线电波或者其他自由传播的电磁波、通过波导或其他传输媒介传播的电磁波(例如，通过光纤电缆的光脉冲)、或者通过电线传输的电信号。
这里参照根据本发明实施例的方法、装置(系统)和计算机程序产品的流程图和/或框图描述了本发明的各个方面。应当理解，流程图和/或框图的每个方框以及流程图和/或框图中各方框的组合，都可以由计算机可读程序指令实现。
这些计算机可读程序指令可以提供给通用计算机、专用计算机或其它可编程数据处理装置的处理器，从而生产出一种机器，使得这些指令在通过计算机或其它可编程数据处理装置的处理器执行时，产生了实现流程图和/或框图中的一个或多个方框中规定的功能/动作的装置。也可以把这些计算机可读程序指令存储在计算机可读存储介质中，这些指令使得计算机、可编程数据处理装置和/或其他设备以特定方式工作，从而，存储有指令的计算机可读介质则包括一个制造品，其包括实现流程图和/或框图中的一个或多个方框中规定的功能/动作的各个方面的指令。
附图中的流程图和框图显示了根据本发明的多个实施例的系统、方法和计算机程序产品的可能实现的体系架构、功能和操作。在这点上，流程图或框图中的每个方框可以代表一个模块、程序段或指令的一部分。方框中所标注的功能也可以以不同于附图中所标注的顺序发生。例如，两个连续的方框实际上可以基本并行地执行，它们有时也可以按相反的顺序执行，这依所涉及的功能而定。
在不偏离所说明的各实施例的范围和精神的情况下，对于本技术领域的普通技术人员来说许多修改和变更都是显而易见的。本文中所用术语的选择，旨在最好地解释各实施例的原理、实际应用或对市场中的技术改进，或者使本技术领域的其它普通技术人员能理解本文披露的各实施例。本发明的范围由所附权利要求来限定。

Claims (10)

  1. 一种基于角速度传感器的步态时相识别方法，包括以下步骤：
    利用传感器采集目标走路的原始角速度信号并同步采集足底压力信号；
    将所述原始角速度信号从传感器坐标系变换到地理坐标系，获得对应的角速度信号，其中所述传感器坐标系是由传感器自身的三个正交测量轴所定义的坐标系；
    基于所述足底压力信号检测步态事件并确定各类步态事件的发生时刻；
    将基于所述足底压力信号检测的步态事件映射到所述角速度信号上，挖掘所述角速度信号的显著性特征，该显著性特征表明，在每个步态周期中，所述角速度信号具有双波谷特征和一段平坦区域；
    通过识别所述角速度信号的显著性特征，检测步态事件并对步态时相进行划分，进而获得识别结果。
  2. 根据权利要求1所述的方法，其特征在于，基于所述足底压力信号检测步态事件并确定各类步态事件的发生时刻包括：
    从所述足底压力信号中筛选出四个特征区域的压力信号，分别对应脚趾、第一跖骨、第五跖骨和脚跟；
    将所述四个特征区域的压力信号与设定阈值Th比较，进而检测出四种步态事件，包括脚跟撞击事件、脚掌着地事件、脚跟离地事件和脚趾离地事件；
    其中，对于脚跟撞击事件，脚跟的压力信号幅值处于增加阶段时，将压力幅值开始大于阈值Th的点，作为脚跟撞击事件的发生时刻；
    对于脚掌着地事件，第一跖骨和第五跖骨区域中任意一个压力信号幅值开始大于Th时，该时间点是脚掌着地事件的发生时刻；
    对于脚跟离地事件，脚跟的压力信号幅值处于减小阶段时，将压力幅值开始小于阈值Th的点，作为脚跟离地事件的发生时刻；
    对于脚趾离地事件，脚趾的压力信号幅值处于减小阶段时，将压力幅值开始小于阈值Th的点，作为脚趾离地事件的发生时刻。
  3. 根据权利要求2所述的方法，其特征在于，针对一个特征区域，根据以下步骤确定所述阈值Th：
    计算所有步态周期的足底压力的平均最大值和平均最小值，表示为：
    Figure PCTCN2022128174-appb-100001
    Figure PCTCN2022128174-appb-100002
    其中，l是步态周期数，Max k是第k个步态周期中压力信号的最大值，Min k是第k个步态周期中压力信号的最小值；
    设置阈值Th，表示为：
    Th＝Th MIN+α(Th MAX-Th MIN)
    其中，α是设定参数。
  4. 根据权利要求2所述的方法，其特征在于，通过识别所述角速度信号的显著性特征，检测步态事件并对步态时相进行划分包括：
    识别所述角速度信号的平坦区域；
    识别所述角速度信号的双波谷特征；
    根据所述平坦区域和所述双波谷特征检测步态事件，其中脚跟撞击事件的时间点对应平坦区域前方的波谷，脚趾离地事件的时间点对应平坦区域后方的波谷，脚掌着地事件和脚跟离地事件分别是平坦区域的起点和终点；
    对所检测出的步态事件进行时相划分，其中承重反应期是脚跟撞击事件到脚掌着地事件之间的阶段，支撑相中期是脚掌着地事件到脚跟离地事件之间的阶段，支撑相末期是脚跟离地事件到脚趾离地事件之间的阶段，摆动相是脚趾离地事件到脚跟撞击事件之间的阶段。
  5. 根据权利要求4所述的方法，其特征在于，识别所述角速度信号的平坦区域包括：
    针对所述角速度信号，求取滑动平均值的绝对值和方差，表示为：
    Figure PCTCN2022128174-appb-100003
    Figure PCTCN2022128174-appb-100004
    其中，W为设置窗口大小，N为步态采样过程中总的采样点数，y i为惯性传感器y轴角速度数据，M i和S i分别是窗口在y i上的滑动均值和方差，γ M和γ S分别为滑动均值和方差的阈值，滑动窗口的步长为1，当滑动均值M i小于阈值γ M，且滑动方差S i小于γ S时，判断其为平坦区域。
  6. 根据权利要求4所述的方法，其特征在于，识别所述角速度信号的双波谷特征包括：
    平坦区域前出现的第一个极小值点，且角速度值小于0，对应第一波谷特征；平坦区域后出现的第一个极小值点，且角速度值小于0，对应第二波谷特征。
  7. 根据权利要求2所述的方法，其特征在于，在检测步态事件并对步态时相进行划分之后，还包括：根据以下方式计算每一个步态周期的步态时间参数：设第k个周期的四个步态事件的时间点为HS(k)、FF(k)、HO(k)、TO(k)，将脚跟撞击事件设为每一个步态周期的起点，则步态周期T表示为：
    T(k)＝HS(k+1)-HS(k)
    承重反应期相位的持续时间T LR为：
    T LR(k)＝F(k)-HS(k)
    支撑相中期相位的持续时间T MS为：
    T MS(k)＝HO(k)-FF(k)
    支撑相末期相位的持续时间T TS为：
    T TS(k)＝TO(k)-HO(k)
    摆动相相位的持续时间T SW为：
    T SW(k)＝HS(k+1)-TO(k)
    其中，HS表示脚跟撞击事件、FF表示脚掌着地事件、HO表示脚跟离地事件、TO表示脚趾离地事件。
  8. 根据权利要求1所述的方法，其特征在于，所述原始角速度信号利用设置在脚面处的惯性传感器采集，所述足底压力信号利用设置在鞋底的足底压力测量系统采集，并且利用四元数法将所述原始角速度信号从传感器坐标系变换到地理坐标系。
  9. 一种计算机可读存储介质，其上存储有计算机程序，其中，该计算 机程序被处理器执行时实现根据权利要求1至8中任一项所述方法的步骤。
  10. 一种计算机设备，包括存储器和处理器，在所述存储器上存储有能够在处理器上运行的计算机程序，其特征在于，所述处理器执行所述计算机程序时实现权利要求1至8中任一项所述的方法的步骤。
PCT/CN2022/128174 2022-10-28 2022-10-28 一种基于角速度传感器的步态时相识别方法 WO2024087151A1 (zh)

Priority Applications (1)

Application Number Priority Date Filing Date Title
PCT/CN2022/128174 WO2024087151A1 (zh) 2022-10-28 2022-10-28 一种基于角速度传感器的步态时相识别方法

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
PCT/CN2022/128174 WO2024087151A1 (zh) 2022-10-28 2022-10-28 一种基于角速度传感器的步态时相识别方法

Publications (1)

Publication Number Publication Date
WO2024087151A1 true WO2024087151A1 (zh) 2024-05-02

Family

ID=90829711

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/CN2022/128174 WO2024087151A1 (zh) 2022-10-28 2022-10-28 一种基于角速度传感器的步态时相识别方法

Country Status (1)

Country Link
WO (1) WO2024087151A1 (zh)

Similar Documents

Publication Publication Date Title
CN110916679B (zh) 一种人体下肢位姿步态检测装置及方法
EP3468450B1 (en) Method and system for analyzing human gait
US9119568B2 (en) Sensing device for detecting a wearing position
Bötzel et al. Gait recording with inertial sensors–How to determine initial and terminal contact
US9307932B2 (en) System and method for 3D gait assessment
Mannini et al. Walking speed estimation using foot-mounted inertial sensors: Comparing machine learning and strap-down integration methods
Barth et al. Subsequence dynamic time warping as a method for robust step segmentation using gyroscope signals of daily life activities
Kanzler et al. Inertial sensor based and shoe size independent gait analysis including heel and toe clearance estimation
Wang et al. Quantitative assessment of dual gait analysis based on inertial sensors with body sensor network
Lee et al. Detection of spatio-temporal gait parameters by using wearable motion sensors
CN108836344A (zh) 步长步频估算方法和装置及步态检测仪
JP2021030051A (ja) 転倒リスク評価方法、転倒リスク評価装置及び転倒リスク評価プログラム
CN114271812A (zh) 基于惯性传感器的三维步态分析系统及方法
CN115530815A (zh) 一种基于角速度传感器的步态时相识别方法
WO2024087151A1 (zh) 一种基于角速度传感器的步态时相识别方法
Yang et al. Empowering a gait feature-rich timed-up-and-go system for complex ecological environments
US20240049987A1 (en) Gait measurement system, gait measurement method, and program recording medium
CN112438723A (zh) 认知功能评估方法、认知功能评估装置以及存储介质
CN104331705B (zh) 融合时空信息的步态周期自动检测方法
Boutaayamou et al. Algorithm for temporal gait analysis using wireless foot-mounted accelerometers
JP7473354B2 (ja) サルコペニア評価方法、サルコペニア評価装置及びサルコペニア評価プログラム
Wu et al. Recognization of stance phase using flexible pressure sensors
Salis et al. A multi-sensor wearable system for gait assessment in real-world conditions: performance in individuals with impaired mobility
Boutaayamou et al. Extraction of Temporal Gait Parameters using a Reduced Number of Wearable Accelerometers.
CN110693501A (zh) 一种基于多传感器融合的行走步态无线检测系统