A terrain recognition method based on lower limb inertial information

Abstract

The present invention relates to a terrain recognition method based on lower limb inertial information, which belongs to the field of terrain recognition technology and solves the problems of high cost, low recognition accuracy and inapplicability to complex terrain in the existing terrain recognition methods. A self-oscillating sine curve is constructed based on the hip joint angle to obtain a first terrain recognition result, energy peak detection is performed based on the acceleration data output by the inertial measurement unit of the lower back, and a second terrain recognition result is obtained based on the detection result; the amplitude and offset of the hip joint angle curve, the knee joint angle curve and the forward lean angle curve are obtained, and compared with the amplitude range and offset range of the upper body forward lean angle, hip joint angle and knee joint angle corresponding to each terrain stored in the adaptive threshold knowledge base to obtain a third terrain recognition result; the final terrain recognition result is obtained based on the first, second and third recognition results. A low-cost and high-recognition-accuracy lower limb assisted exoskeleton terrain recognition method is realized.

Description

Technical Field

[0001] The present invention relates to the technical field of terrain recognition, and in particular to a terrain recognition method based on lower limb inertial information. Background Art

[0002] A lower-limb assistance exoskeleton is a mechanical device worn on the lower limbs. Using sensors and algorithms, it detects the type of terrain the user is walking on (e.g., flat ground, slopes, stairs, etc.) in real time and dynamically adjusts the exoskeleton's assistance strategy based on this recognition. Therefore, terrain recognition is crucial for lower-limb assistance exoskeletons. It allows the exoskeleton to adjust its assistance mode based on different terrain conditions, such as flat ground, slopes, and stairs, to optimize movement efficiency, reduce energy consumption, and improve safety and comfort.

[0003] In the existing technology, the terrain recognition methods of lower limb assisted exoskeleton mainly include: (1) the terrain recognition method of lower limb assisted exoskeleton based on visual sensors, which collects the ground image or three-dimensional point cloud data in front by equipping the exoskeleton with a camera or depth camera, and uses computer vision or deep learning models to identify the terrain category. However, this method depends on lighting and environment, has high computing resource requirements, and increases the complexity of load and wearing; (2) the method based on inertial measurement unit (IMU) and pressure sensor, which is based on the IMU data of bilateral calves and soles and the sole pressure sensor data, through gait cycle division, filtering and normalization processing, and realizes terrain recognition through neural network model or threshold classification. However, this method has low recognition accuracy and limited recognition ability for complex terrain.

[0004] Therefore, it is necessary to provide a terrain recognition method for lower limb assisted exoskeleton that is low in cost, has high recognition accuracy, and can be applied to complex terrain. Summary of the Invention

[0005] In view of the above analysis, an embodiment of the present invention aims to provide a terrain recognition method based on lower limb inertial information to solve the problems of existing terrain recognition methods such as high cost, low recognition accuracy, and inapplicability to complex terrain.

[0006] An embodiment of the present invention provides a terrain recognition method based on lower limb inertial information, comprising:

[0007] Inertial measurement units are arranged on the left and right thighs, left and right calves, and lower back of the human body to be measured;

[0008] calculating a hip joint angle based on output data of an inertial measurement unit of the left and right thighs and the lower back, constructing a self-oscillating sine curve based on the hip joint angle, and obtaining a first terrain recognition result based on the self-oscillating sine curve;

[0009] performing energy peak detection based on acceleration data output by the inertial measurement unit of the rear waist, and obtaining a second terrain recognition result based on the detection result;

[0010] Calculating the hip joint angle, knee joint angle, and forward lean angle of the human body at each moment during the current motion process based on output data from the inertial measurement units of the left and right thighs, left and right calves, and lower back, thereby constructing a hip joint angle curve, a knee joint angle curve, and a forward lean angle curve, obtaining amplitudes and offsets corresponding to the above curves, and comparing them with amplitude ranges and offset ranges of the upper body forward lean angle, hip joint angle, and knee joint angle corresponding to each terrain stored in an adaptive threshold knowledge base to obtain a third terrain recognition result;

[0011] A final terrain recognition result is obtained based on the first terrain recognition result, the second terrain recognition result, and the third terrain recognition result.

[0012] Based on a further improvement of the above method, the hip joint angle includes: a first hip joint angle calculated based on the output data of the inertial measurement unit of the left thigh and the lower back; a second hip joint angle calculated based on the output data of the inertial measurement unit of the right thigh and the lower back; the self-oscillating sine curve includes: a first self-oscillating sine curve constructed based on the first hip joint angle, and a second self-oscillating sine curve constructed based on the second hip joint angle.

[0013] Based on a further improvement of the above method, the step of constructing a self-oscillating sinusoidal curve based on the hip joint angle includes:

[0014] S1: Initialize the amplitude, frequency, phase and bias of the self-oscillating sinusoid;

[0015] S2: obtaining the actual hip joint angle at the current moment, and calculating the tracking error based on the actual hip joint angle and the simulated hip joint angle output by the self-oscillating sine curve;

[0016] S3: If the tracking error is less than the first target threshold, the current self-oscillating sinusoidal curve is used as the self-oscillating sinusoidal curve corresponding to the human body to be measured; if the tracking error is greater than or equal to the first target threshold, the amplitude, frequency, phase and offset of the self-oscillating sinusoidal curve at the current moment are updated based on the tracking error, and the updated values ​​are used as the amplitude, frequency, phase and offset of the self-oscillating sinusoidal curve at the next moment, and the process returns to step S2.

[0017] Based on a further improvement of the above method, obtaining a first terrain recognition result based on the self-oscillating sine curve includes:

[0018] Comparing the amplitude, frequency, phase, and offset of the first self-oscillating sinusoidal curve with a preset first threshold classification table to obtain a first initial terrain recognition result, and comparing the amplitude, frequency, phase, and offset of the second self-oscillating sinusoidal curve with the preset first threshold classification table to obtain a second initial terrain recognition result;

[0019] If the first terrain initial recognition result is consistent with the second terrain initial recognition result, the first terrain initial recognition result or the second terrain initial recognition result is used as the first terrain recognition result; if they are inconsistent, the irregular terrain is used as the first terrain recognition result;

[0020] The first threshold classification table is used to store the amplitude range, frequency range, phase range and offset range corresponding to each type of terrain.

[0021] According to a further improvement of the above method, the updating of the amplitude, frequency, phase and offset of the self-oscillating sinusoidal curve at the current moment based on the tracking error includes:

[0022] ;

[0023] in, is the updated phase, is the updated frequency, is the updated amplitude, is the updated bias, is the horizontal learning parameter, is the longitudinal learning parameter, is the tracking error.

[0024] According to a further improvement of the above method, obtaining a final terrain recognition result based on the first terrain recognition result, the second terrain recognition result, and the third terrain recognition result includes:

[0025] If the first terrain recognition result, the second terrain recognition result, and the third terrain recognition result are the same, the first terrain recognition result, the second terrain recognition result, or the third terrain recognition result is used as the final terrain recognition result;

[0026] If the first terrain recognition result, the second terrain recognition result, and the third terrain recognition result are different, constructing a recognition result set based on the terrain types appearing in each terrain recognition result, and calculating the probability of each terrain type in the recognition result set based on the weights of the first terrain recognition result, the second terrain recognition result, and the third terrain recognition result, and selecting the terrain type with the highest probability as the final terrain recognition result;

[0027] The terrain recognition results include: flat land type, stair-up type, uphill type, downhill type, downhill type, and irregular terrain type.

[0028] Based on a further improvement of the above method, the calculating the probability of each terrain type in the recognition result set includes:

[0029] ,

[0030] Wherein, c is the terrain type in the recognition result set, is the probability value when the terrain type is c, 、 、 is the result value of terrain type c; when the terrain type c is consistent with the first terrain recognition result, ,otherwise, ; When the terrain type c is consistent with the second terrain recognition result, ,otherwise, ; When the terrain type c is consistent with the third terrain recognition result, ,otherwise, ; 、 、 are the weights corresponding to the first terrain recognition result, the second terrain recognition result, and the third terrain recognition result.

[0031] According to a further improvement of the above method, energy peak detection is performed based on the acceleration data output by the inertial measurement unit of the lower back, and a second terrain recognition result is obtained based on the detection result, including:

[0032] Acquiring the acceleration data, taking a period in which the acceleration value in the acceleration data exceeds a preset detection threshold as a peak detection interval, and performing energy peak detection based on the peak detection interval to determine a touchdown point moment;

[0033] Based on the output data of the inertial measurement unit at the time of contact and the limb lengths of the left and right thighs and left and right calves, a human lower limb model at the time of contact is constructed, the coordinates of the center of mass of the left and right feet are determined based on the human lower limb model, and a second terrain recognition result is obtained based on the coordinates of the center of mass of the left and right feet.

[0034] Based on a further improvement of the above method, the adaptive threshold knowledge base is used to store a first offset range of the upper body forward lean angle, a second amplitude range and a second offset range of the hip joint angle, and a third amplitude range and a third offset range of the knee joint angle corresponding to different terrains;

[0035] The first offset range, the second amplitude range and the second offset range of the hip joint angle, and the third amplitude range and the third offset range of the knee joint angle are obtained as follows:

[0036] For any terrain, obtain the upper body forward lean angle, hip joint angle, and knee joint angle of different people at different times, construct upper body forward lean angle curves, hip joint angle curves, and knee joint angle curves based on the upper body forward lean angle, hip joint angle, and knee joint angle curves at each time and their corresponding upper body forward lean angle, hip joint angle, and knee joint angle curves, and extract the amplitudes and offsets corresponding to the upper body forward lean angle curves, the hip joint angle curves, and the knee joint angle curves;

[0037] Obtaining a first offset range based on the offset corresponding to the upper body forward lean angle curve of all persons;

[0038] Obtaining a second amplitude range and a second offset range based on the amplitudes and offsets corresponding to the hip joint angle curves of all persons;

[0039] A third amplitude range and a third offset range are obtained based on the amplitudes and offsets corresponding to the knee joint angle curves of all persons.

[0040] Based on a further improvement of the above method, the third terrain recognition result is obtained by comparing it with the amplitude range and offset range of the upper body forward lean angle, hip joint angle, and knee joint angle stored in the adaptive threshold knowledge base, including:

[0041] For any terrain stored in the adaptive threshold knowledge base, if the bias of the forward tilt angle is within the first bias range, and the amplitude and bias of the first hip joint angle are respectively within the second amplitude range and the second bias range, or the amplitude and bias of the second hip joint angle are respectively within the second amplitude range and the second bias range, and the amplitude and bias of the first knee joint angle are respectively within the third amplitude range and the third bias range, or the amplitude and bias of the second knee joint angle are respectively within the third amplitude range and the third bias range, then the terrain is taken as the third terrain recognition result. If no matching terrain exists, the irregular terrain is taken as the third terrain recognition result.

[0042] Compared with the prior art, the present invention can achieve at least one of the following beneficial effects:

[0043] The present invention provides a terrain recognition method based on lower limb inertial information. The final terrain recognition result is determined by combining the recognition results of an adaptive oscillator-based terrain recognition algorithm, a ground contact detection-based terrain recognition algorithm, and a joint angle feature information-based terrain recognition algorithm. This reduces the error associated with obtaining the recognition result using only a single recognition method and improves the accuracy and reliability of the terrain recognition result. Furthermore, the method proposed in the present invention only requires data acquired by five inertial measurement units to achieve terrain recognition. Compared to the existing lower limb assisted exoskeleton terrain recognition method based on visual sensors, its hardware cost is lower. Furthermore, since all inertial measurement units are placed on the human body, the method is unaffected by the external environment. Its high-precision motion capture capability and efficient dynamic response capability make it better suited for complex terrain recognition.

[0044] In the present invention, the above-mentioned technical solutions can be combined with each other to achieve more preferred combinations. Other features and advantages of the present invention will be described in the following description, and some advantages will become apparent from the description or be learned through practice of the present invention. The objectives and other advantages of the present invention can be realized and obtained through the contents particularly pointed out in the description and drawings. BRIEF DESCRIPTION OF THE DRAWINGS

[0045] The accompanying drawings are only used for the purpose of illustrating specific embodiments and are not to be considered as limiting the present invention. Throughout the drawings, the same reference symbols denote the same components.

[0046] Figure 1 This is an example diagram of a terrain recognition method based on lower limb inertial information in an embodiment of the present invention;

[0047] Figure 2 This is an example diagram of energy peak detection in an embodiment of the present invention;

[0048] Figure 3 This is an example diagram of a human lower limb model in an embodiment of the present invention. DETAILED DESCRIPTION

[0049] The preferred embodiments of the present invention will be described in detail below in conjunction with the accompanying drawings, wherein the accompanying drawings constitute a part of this application and are used together with the embodiments of the present invention to illustrate the principles of the present invention, and are not used to limit the scope of the present invention.

[0050] A specific embodiment of the present invention discloses a terrain recognition method based on lower limb inertial information, such as Figure 1 Shown, including:

[0051] S1: Inertial measurement units are placed on the left and right thighs, left and right calves, and lower back of the human body to be measured.

[0052] Among them, the layout position and angle of the inertial measurement unit (IMU) directly determine the validity of sensor data and the accuracy of terrain classification.

[0053] For both thighs, the IMU can be placed on the front of the thigh, which makes it easier to hide and fix the sensor cable and improves wearing stability. To effectively reduce the impact of muscle contraction or soft tissue movement on the sensor and ensure measurement stability, it can be placed on the midline of the front thigh (near the rectus femoris), about 15-20 cm away from the hip joint (or 1 / 3 of the thigh length).

[0054] For the left and right calves, the IMU can be placed on the front of the calf. In order to reduce the sensor shaking caused by the swing of the calf during walking, avoid the tendon activity area, and avoid the sensor displacement caused by the violent contraction of the tibialis anterior muscle, it can be placed on the midline of the front of the calf (near the tibialis anterior muscle), about 10-15 cm away from the knee joint (or 1 / 2 of the calf length).

[0055] For the lower back, because it provides a global motion reference (such as the direction of gravitational acceleration), it can also be used to calibrate the coordinate system of the lower limb sensors. It is usually placed at the L3 or L4 position of the lumbar spine, near the midline of the spine. It can effectively reduce the coupling interference of trunk rotation on lower limb movement, so that the output data of the IMU can represent the movement of the pelvis.

[0056] After placing the inertial measurement units on the left and right thighs, left and right calves, and lower back of the human body, the IMU installation angle needs to be calibrated to ensure the accuracy of kinematic calculations. For example, the calibration includes:

[0057] (1) Alignment of bone axes, that is, the coordinate system of the thigh and calf IMU must be strictly aligned with the anatomical axes of the femur and tibia (such as the Y axis along the long axis of the bone) to ensure that the measured angular velocity / acceleration reflects the actual limb movement.

[0058] Calibration methods include: static calibration, where the sensor coordinate system is aligned using the direction of gravity (Z-axis vertically downward) when the user is standing upright; and dynamic verification, where the user verifies that the coordinate system rotation matches the joint motion plane by walking or lifting the leg.

[0059] (2) Establish a multi-IMU synchronous reference system, with the rear IMU as the origin of the global reference system. Other IMUs are converted to this coordinate system through initial posture transformation to ensure data spatial consistency.

[0060] Alternatively, for the front thigh, align the IMU's x-axis along the long axis of the thigh (from the hip joint to the knee joint), with the y-axis pointing to the side of the body, and the z-axis perpendicular to the first two axes. For the front calf, align the IMU's x-axis along the long axis of the calf (from the knee joint to the ankle joint), with the y-axis pointing to the side of the body, and the z-axis perpendicular to the first two axes. For the lower back, align the IMU's x-axis roughly along the front-to-back direction of the body (from the front to the back of the pelvis), with the y-axis along the left-right direction of the body, and the z-axis vertically along the body.

[0061] S2: Calculating a hip joint angle based on output data of the inertial measurement units of the left and right thighs and the lower back, constructing a self-oscillating sine curve based on the hip joint angle, and obtaining a first terrain recognition result based on the self-oscillating sine curve.

[0062] Output data refers to the triaxial acceleration, triaxial angular velocity, and triaxial magnetic field intensity data output by the IMU sensor, if a magnetometer is included. Before calculating the hip joint angle, it is necessary to verify that the collected IMU data is synchronized in time to facilitate subsequent accurate correlation analysis and fusion calculation of data at different locations. For example, data synchronization can be achieved by setting a unified timestamp in the data acquisition system or using an external synchronization signal.

[0063] The calculating of the hip joint angle based on output data of the inertial measurement units of the left and right thighs and the lower back includes:

[0064] A1: Filter the three output data to remove noise and unnecessary high-frequency interference. For example, a low-pass filter can remove high-frequency noise and retain low-frequency signals related to human motion.

[0065] A2: Convert the local coordinate systems of the three IMUs to a unified reference coordinate system. Usually, the rear IMU is used as the origin of the global reference system. Coordinate system alignment is achieved through mathematical tools such as coordinate transformation matrices or quaternions, ensuring that data at different locations are compared and calculated in the same reference system.

[0066] A3: Use data fusion algorithms to estimate the posture of each IMU location. Common data fusion algorithms include extended Kalman filtering (EKF), complementary filtering, etc. These algorithms can integrate data from accelerometers, gyroscopes, and magnetometers (if present) to obtain more accurate and stable posture information of human body parts. The estimated posture information is expressed in a suitable form, usually using quaternions or Euler angles. Quaternions have the advantages of computational stability and no singularities when representing three-dimensional rotations, and are suitable for subsequent posture calculations and fusion; while Euler angles more intuitively reflect the angle changes of human joints, which facilitates the final calculation and interpretation of the hip joint angle;

[0067] A4: Using the IMU posture data of the left and right thighs and lower back, calculate the posture difference of the thigh relative to the lower back (pelvis), that is, calculate the relative rotation matrix of the thigh posture relative to the torso posture, and convert the relative rotation matrix into Euler angles (i.e., flexion and extension angles), which correspond to the movement angles of the hip joint.

[0068] Optionally, after obtaining the hip joint angle, error correction and compensation can be performed to reduce angle calculation deviations caused by factors such as sensor errors, inaccurate model assumptions, and cumulative errors in data processing, thereby improving the accuracy and reliability of hip joint angle calculation.

[0069] The constructing of a self-oscillating sinusoidal curve based on the hip joint angle comprises:

[0070] N1: Initializes the amplitude, frequency, phase, and offset of the self-oscillating sinusoid.

[0071] The mathematical form of the self-oscillating sinusoid is:

[0072] ,

[0073] in, is the oscillator amplitude, is the oscillator frequency, is the oscillator phase, is the oscillator bias, Is the simulated hip joint angle.

[0074] When initializing the various parameters of the self-oscillating sine curve, they can be set randomly or based on experience or a historical template. Each piece of data in the historical template includes factors such as a person's height, weight, age, and gender, as well as the corresponding values ​​of the self-oscillating sine curve's amplitude, frequency, phase, and offset. Users can initialize the parameters based on the values ​​of the stored templates that roughly match the height, weight, age, and gender of the person being measured.

[0075] N2: Obtain the actual hip joint angle at the current moment, and calculate the tracking error based on the actual hip joint angle and the simulated hip joint angle output by the self-oscillating sine curve.

[0076] The tracking error calculated based on the actual hip joint angle and the simulated hip joint angle outputted by the self-oscillating sinusoidal curve can be expressed as:

[0077] ,

[0078] in, is the actual hip joint angle.

[0079] N3: If the tracking error is less than the first target threshold, the current self-oscillating sinusoidal curve is used as the self-oscillating sinusoidal curve corresponding to the human body to be measured; if the tracking error is greater than or equal to the first target threshold, the amplitude, frequency, phase and offset of the self-oscillating sinusoidal curve at the current moment are updated based on the tracking error, and the updated values ​​are used as the amplitude, frequency, phase and offset of the self-oscillating sinusoidal curve at the next moment, and the process returns to step N2.

[0080] The updating of the amplitude, frequency, phase and offset of the self-oscillating sinusoidal curve at the current moment based on the tracking error includes:

[0081] ;

[0082] in, is the updated phase, is the updated frequency, is the updated amplitude, is the updated bias, is the horizontal learning parameter, is the longitudinal learning parameter, is the tracking error.

[0083] For example, the first target threshold can be set according to actual needs or experience. Preferably, the first target threshold is 0.01.

[0084] As can be seen from step S1, the terrain recognition method proposed in the present invention deploys inertial measurement units (IMUs) on the left and right thighs, left and right calves, and lower back, for a total of five IMUs. The calculation of the hip joint angle requires consideration of the hip joint angles between the left thigh and lower back, as well as the hip joint angle between the right thigh and lower back. Therefore, the actual hip joint angles include: a first hip joint angle calculated based on the output data of the inertial measurement units for the left thigh and lower back; and a second hip joint angle calculated based on the output data of the inertial measurement units for the right thigh and lower back. Because two hip joint angles are obtained, two self-oscillating sinusoidal curves are generated: a first self-oscillating sinusoidal curve constructed based on the first hip joint angle, and a second self-oscillating sinusoidal curve constructed based on the second hip joint angle. Because two self-oscillating sinusoidal curves are obtained, two terrain recognition results are also generated: a first initial terrain recognition result corresponding to the left leg, and a second initial terrain recognition result corresponding to the right leg.

[0085] The obtaining of a first terrain recognition result based on the self-oscillating sinusoidal curve includes:

[0086] Comparing the amplitude, frequency, phase, and offset of the first self-oscillating sinusoidal curve with a preset first threshold classification table to obtain a first initial terrain recognition result, and comparing the amplitude, frequency, phase, and offset of the second self-oscillating sinusoidal curve with the preset first threshold classification table to obtain a second initial terrain recognition result;

[0087] If the first terrain initial recognition result is consistent with the second terrain initial recognition result, the first terrain initial recognition result or the second terrain initial recognition result is used as the first terrain recognition result; if they are inconsistent, the irregular terrain is used as the first terrain recognition result.

[0088] The first threshold classification table is used to store the amplitude range, frequency range, phase range, and offset range corresponding to each type of terrain. The first threshold classification table is shown in Table 1.

[0089] Comparing the amplitude, frequency, phase, and offset of the first self-oscillating sinusoidal curve or the amplitude, frequency, phase, and offset of the second self-oscillating sinusoidal curve with a preset first threshold classification table to obtain an initial terrain recognition result means:

[0090] M1: Determine whether the phase of the current self-oscillating sinusoidal curve meets the phase range specified by each terrain. If so, it indicates that the adaptive oscillator is operating normally, and execute step M2. If not, the irregular terrain is used as the initial terrain recognition result, and the weight of the adaptive oscillator recognition in the subsequent step S5 is reduced. At the same time, an error message is output to indicate that the current adaptive oscillator has failed to be activated normally, and jump to M5.

[0091] M2: Determine whether the frequency of the current self-oscillation sine curve meets the frequency range specified by each terrain. If the current frequency is within the specified frequency range, maintain the amplitude range and offset range specified by each terrain. If the current frequency is not within the specified frequency range, reduce the amplitude range and offset range specified by each terrain by 10%.

[0092] For example, the frequency range is set to 0-0.7. If the current frequency is less than or equal to 0.7, the amplitude range and offset range specified by each terrain remain unchanged. If the current frequency is greater than 0.7, the amplitude range and offset range specified by each terrain are reduced by 10%.

[0093] M3: Determine whether the amplitude of the current self-oscillating sine curve meets the amplitude range of any terrain type. If so, execute step M4. If not, take the irregular terrain as the initial terrain recognition result and jump to M5.

[0094] M4: Determine whether the bias value of the current self-oscillating sine curve meets the bias range specified by the terrain (i.e., the terrain type determined in step M3). If so, execute step M5. If not, take the irregular terrain as the initial terrain recognition result and jump to M5.

[0095] M5: Output the initial terrain recognition results.

[0096] Table 1

[0097] ;

[0098] S3: performing energy peak detection based on the acceleration data output by the inertial measurement unit of the lower back, and obtaining a second terrain recognition result based on the detection result, including:

[0099] Acquiring the acceleration data, taking a period in which the acceleration value in the acceleration data exceeds a preset detection threshold as a peak detection interval, and performing energy peak detection based on the peak detection interval to determine a touchdown point moment;

[0100] Based on the output data of the inertial measurement unit at the time of contact and the limb lengths of the left and right thighs and left and right calves, a human lower limb model at the time of contact is constructed, the coordinates of the center of mass of the left and right feet are determined based on the human lower limb model, and a second terrain recognition result is obtained based on the coordinates of the center of mass of the left and right feet.

[0101] During walking, the moment the foot touches the ground, the lower back IMU node generates a significant acceleration spike. This spike can be detected to identify the moment of foot contact. After detecting the moment of foot contact, a lower limb posture model is established based on the lower limb joint angles. This lower limb posture model is used to determine the spatial position of both feet at the moment of contact and identify the current terrain.

[0102] In touchdown point detection, a combination of threshold detection and energy spike detection is used to avoid false detection caused by glitches. First, a preset detection threshold is defined. When the absolute value of the acceleration exceeds the threshold, spike detection begins. When the absolute value of the acceleration falls back below the threshold, spike detection ends. The acceleration during the spike detection period is integrated. If it exceeds a certain energy, it means that the spike can be used as a single-foot touchdown estimation point. If it does not exceed it, it means that the spike may be caused by a glitch and cannot be used as a single-foot touchdown estimation point. The energy spike detection formula is as follows:

[0103] ,

[0104] in, is the acceleration data, and t1~t2 are the lower and upper limits of the time interval that is higher than the preset acceleration detection threshold.

[0105] Figure 2 An example diagram of energy peak detection is given. For example, the preset detection threshold can be set to 0.36.

[0106] After calculating the acceleration integral value for the current interval based on the energy peak detection formula, it is compared with a preset energy threshold. If the current acceleration integral value exceeds the preset energy threshold, the point with the maximum acceleration value within the current interval is used as the touchdown point, and the moment corresponding to this maximum acceleration value is used as the touchdown time. The preset energy threshold can be set based on experience or manually based on factors such as the height and weight of the person being tested.

[0107] After detecting the moment of contact, the current human lower limb model is used to analyze information such as the heel contact height and distance to determine the current movement pattern and the current terrain. The coordinates of the center of mass of both feet at the time of contact can be used to determine the current terrain. For example, on flat ground, the vertical coordinates of the center of mass of both feet should be roughly the same. When walking upstairs, the horizontal and vertical coordinates of the center of mass of the front foot are higher, while the horizontal and vertical coordinates of the center of mass of the back foot are lower. When walking downstairs, the horizontal coordinate of the front foot is higher and the vertical coordinate is lower, while the vertical coordinate of the back foot is lower.

[0108] Human lower limb model Figure 3 As shown, is the center of mass of the left foot, is the center of mass of the right foot, and are the angles between the left and right calves and the vertical direction, and are the angles between the left and right thighs and the vertical direction, obtained through inertial sensors. and Indicates the length of the left and right calves. and Indicates the length of the left and right thighs. and It means and Coordinates of two points. Assuming the waist is the origin of the coordinate system, the coordinates of the center of mass of the two feet are:

[0109] ,

[0110] .

[0111] S4: Based on the output data of the inertial measurement units of the left and right thighs, left and right calves and lower back, the hip joint angle, knee joint angle and forward lean angle of the human body to be measured at each moment during the current movement process are calculated, thereby constructing a hip joint angle curve, a knee joint angle curve and a forward lean angle curve, obtaining the amplitude and offset corresponding to the above curves, and comparing them with the amplitude range and offset range of the upper body forward lean angle, hip joint angle and knee joint angle corresponding to each terrain stored in the adaptive threshold knowledge base to obtain a third terrain recognition result.

[0112] The definition of output data in this step is the same as that in step S2.

[0113] It is worth noting that the terrain recognition method proposed in the present invention is a method that operates in real time in a lower-limb assisted exoskeleton mechanical device. Therefore, in this step, the constructed hip joint angle curve, knee joint angle curve, and anteversion angle curve are actually curves formed with the end time of the previous sampling period as the starting time. Furthermore, the extracted amplitude and offset are also the amplitude and offset corresponding to each curve with the end time of the previous sampling period as the starting time. In other words, the terrain type at the current moment is judged (predicted) based on the characteristics of the curve formed by the current sampling period, and a rapid response is made based on the identified terrain type. For example, assuming that the current moment is t, the end time of the previous sampling period is obtained, denoted as tn, where n is the sampling time window, generally selected to include a complete gait cycle of the left and right legs, about 1 second. Based on the hip joint angle, knee joint angle, and anteversion angle at each time point tn, t-n+1, ​​t-n+2, t-n+3, …, t-1, the hip joint angle curve, knee joint angle curve, and anteversion angle curve are constructed, and the amplitude and offset corresponding to each curve are extracted.

[0114] The knee joint angle is calculated as follows:

[0115] B1: Filter the four output data to remove noise and unnecessary high-frequency interference. For example, a low-pass filter can remove high-frequency noise and retain low-frequency signals related to human motion.

[0116] B2: Convert the local coordinate systems of the four IMUs to a unified reference coordinate system. Typically, the rear IMU is used as the origin of the global reference system. Coordinate system alignment is achieved through mathematical tools such as coordinate transformation matrices or quaternions, ensuring that data from different locations are compared and calculated in the same reference system.

[0117] B3: Use a data fusion algorithm to estimate the pose of each IMU location. Common data fusion algorithms include the extended Kalman filter (EKF) and complementary filtering. These algorithms combine data from the accelerometer, gyroscope, and magnetometer (if present) to obtain more accurate and stable pose information for each body part. The estimated pose information is expressed in a suitable form, typically using quaternions or Euler angles.

[0118] B4: Using the IMU posture data of the left and right thighs and left and right calves, calculate the posture difference of the thigh relative to the calf, that is, calculate the relative rotation matrix of the thigh posture relative to the calf posture, and convert the relative rotation matrix into Euler angles (i.e., flexion and extension angles), which correspond to the movement angles of the knee joint.

[0119] Optionally, after obtaining the knee joint angle, error correction and compensation can be performed to reduce angle calculation deviations caused by factors such as sensor errors, inaccurate model assumptions, and cumulative errors in data processing, thereby improving the accuracy and reliability of hip joint angle calculation.

[0120] The forward inclination angle is calculated as follows:

[0121] C1: Filter the output data of the back waist IMU to remove noise and unnecessary high-frequency interference. For example, a low-pass filter can remove high-frequency noise and retain low-frequency signals related to human motion.

[0122] C2: Use the back IMU as the origin of the global reference system and use mathematical tools such as coordinate transformation matrix or quaternion to achieve coordinate system alignment, ensuring that data at different locations are compared and calculated in the same reference system;

[0123] C3: Use a data fusion algorithm to estimate the pose of each IMU location. Common data fusion algorithms include the Extended Kalman Filter (EKF) and the Complementary Filter. These algorithms combine data from the accelerometer, gyroscope, and magnetometer (if present) to obtain more accurate and stable pose information for each body part. The estimated pose information is expressed in a suitable form, typically using quaternions or Euler angles.

[0124] C4: Convert the IMU posture data into Euler angles and extract the rotation angle of the sagittal plane as the anteversion angle.

[0125] Optionally, after obtaining the anteversion angle, error correction and compensation can be performed to reduce angle calculation deviations caused by factors such as sensor errors, inaccurate model assumptions, and cumulative errors in data processing, thereby improving the accuracy and reliability of hip joint angle calculation.

[0126] The adaptive threshold knowledge base is used to store a first offset range of the upper body forward lean angle, a second amplitude range and a second offset range of the hip joint angle, and a third amplitude range and a third offset range of the knee joint angle corresponding to different terrains;

[0127] The first offset range, the second amplitude range and the second offset range of the hip joint angle, and the third amplitude range and the third offset range of the knee joint angle are obtained as follows:

[0128] T1: For any terrain, obtain the upper body forward lean angle, hip joint angle and knee joint angle of different people at different times, construct the upper body forward lean angle curve, hip joint angle curve and knee joint angle curve based on each moment and its corresponding upper body forward lean angle, hip joint angle and knee joint angle, and extract the amplitude and offset corresponding to the upper body forward lean angle curve, the hip joint angle curve and the knee joint angle curve.

[0129] In this step, the constructed hip joint angle curve, knee joint angle curve and anteversion angle curve are actually curves formed with the end time of the previous sampling period as the starting time, and the extracted amplitude and bias are also the amplitude and bias corresponding to each curve with the end time of the previous sampling period as the starting time.

[0130] T2: obtaining a first offset range based on the offset corresponding to the upper body forward lean angle curve of all persons.

[0131] T3: Obtaining a second amplitude range and a second offset range based on the amplitude and offset corresponding to the hip joint angle curves of all persons.

[0132] T4: obtaining a third amplitude range and a third offset range based on the amplitudes and offsets corresponding to the knee joint angle curves of all persons.

[0133] Exemplarily, the amplitude range and the offset range are obtained based on the amplitude and offset corresponding to the angle curves of all persons, including:

[0134] Compare the amplitudes of all people and find the minimum and maximum values. The amplitude range is the interval from the minimum amplitude to the maximum amplitude. This range reflects the difference in the fluctuation range of the angle curves of different individuals. Compare the biases of all people and find the minimum and maximum values. The bias range is the interval from the minimum bias to the maximum bias. This range reflects the difference in the deviation of the angle curves of different individuals relative to the reference value. The bias range is the interval from the minimum bias to the maximum bias. For example, the data stored in the adaptive threshold knowledge base is shown in Table 2, where amp represents amplitude and bias represents bias.

[0135] Table 2

[0136] ;

[0137] The method compares the upper body forward lean angle, hip joint angle, and knee joint angle with the amplitude range and offset range stored in the adaptive threshold knowledge base to obtain a third terrain recognition result, including:

[0138] For any terrain stored in the adaptive threshold knowledge base, if the bias of the forward tilt angle is within the first bias range, and the amplitude and bias of the first hip joint angle are respectively within the second amplitude range and the second bias range, or the amplitude and bias of the second hip joint angle are respectively within the second amplitude range and the second bias range, and the amplitude and bias of the first knee joint angle are respectively within the third amplitude range and the third bias range, or the amplitude and bias of the second knee joint angle are respectively within the third amplitude range and the third bias range, then the terrain is taken as the third terrain recognition result. If no matching terrain exists, the irregular terrain is taken as the third terrain recognition result.

[0139] As can be seen from step S1, the terrain recognition method proposed in the present invention arranges inertial measurement units on the left and right thighs, left and right calves, and lower back of the human body, respectively, for a total of 5 IMUs. When calculating the hip joint angle, it is necessary to consider the hip joint angle between the left thigh and the lower back, as well as the hip joint angle between the right thigh and the lower back. Therefore, the hip joint angle includes: a first hip joint angle calculated based on the output data of the inertial measurement units of the left thigh and the lower back; and a second hip joint angle calculated based on the output data of the inertial measurement units of the right thigh and the lower back. Similarly, when calculating the knee joint angle, it is necessary to consider the knee joint angle between the left thigh and the left calf, as well as the knee joint angle between the right thigh and the right calf, to obtain the first knee joint angle and the second knee joint angle. Therefore, in step S4, two hip joint angle curves, two knee joint angle curves, and one anteversion angle curve are constructed.

[0140] During the comparison, the results corresponding to the first hip joint angle and the second hip joint angle may be different, and the results corresponding to the first knee joint angle and the second knee joint angle may be different. Therefore, if the anteversion angle is consistent, one of the first hip joint angle and the second hip joint angle is consistent, and one of the first knee joint angle and the second knee joint angle is consistent, the terrain will be used as the third terrain recognition result.

[0141] S5: Obtaining a final terrain recognition result based on the first terrain recognition result, the second terrain recognition result, and the third terrain recognition result, including:

[0142] If the first terrain recognition result, the second terrain recognition result, and the third terrain recognition result are the same, the first terrain recognition result, the second terrain recognition result, or the third terrain recognition result is used as the final terrain recognition result;

[0143] If the first terrain recognition result, the second terrain recognition result, and the third terrain recognition result are different, constructing a recognition result set based on the terrain types appearing in each terrain recognition result, and calculating the probability of each terrain type in the recognition result set based on the weights of the first terrain recognition result, the second terrain recognition result, and the third terrain recognition result, and selecting the terrain type with the highest probability as the final terrain recognition result;

[0144] The terrain recognition results include: flat land type, stair-up type, uphill type, downhill type, downhill type, and irregular terrain type.

[0145] The calculating the probability of each terrain type in the recognition result set includes:

[0146] ,

[0147] Wherein, c is the terrain type in the recognition result set, is the probability value when the terrain type is c, 、 、 is the result value of terrain type c; when the terrain type c is consistent with the first terrain recognition result, ,otherwise, ; When the terrain type c is consistent with the second terrain recognition result, ,otherwise, ; When the terrain type c is consistent with the third terrain recognition result, ,otherwise, ; 、 、 are the weights corresponding to the first terrain recognition result, the second terrain recognition result, and the third terrain recognition result, and the sum of the three weights is 1.

[0148] For example, if the first terrain recognition result is flat ground, the second terrain recognition result is upstairs, and the third terrain recognition result is upstairs, then the first terrain recognition result, the second terrain recognition result, and the third terrain recognition result are different, and a recognition result set {flat ground, upstairs} is constructed. Based on the weights of the first terrain recognition result, the second terrain recognition result, and the third terrain recognition result, the probability of each terrain type in the recognition result set is calculated, and the probability of the final terrain recognition result being flat ground is for:

[0149] ,

[0150] The final terrain recognition result is the probability of going upstairs for:

[0151] ,

[0152] Compare and The larger one is taken as the final terrain recognition result.

[0153] The weights of the various terrain recognition results can be set to be equal or based on experience.

[0154] The method proposed in the present invention also supports adaptive adjustment of weights to improve the reliability of the decision-making classification system. The main adjustment principles are: 1. Reduce the weight of the method when there are errors or large fluctuations in the data used by the method, such as when the adaptive oscillator does not converge. 2. For methods based on data feature judgment, reduce the weight of the method when there are specific terrain features that are similar. For example, in a terrain recognition algorithm based on joint angle feature information, the characteristics of the knee joint when going up and down stairs are similar, and when there may be confusion, reduce the weight of the method.

[0155] Compared to the prior art, this embodiment provides a terrain recognition method based on lower-limb inertial information. The final terrain recognition result is determined by combining the recognition results of an adaptive oscillator-based terrain recognition algorithm, a ground contact detection-based terrain recognition algorithm, and a joint angle feature information-based terrain recognition algorithm. This reduces the error associated with obtaining the recognition result using only a single recognition method and improves the accuracy and reliability of the terrain recognition result. Furthermore, the method proposed in this invention only requires data acquired by five inertial measurement units to achieve terrain recognition. Compared to the prior art method of terrain recognition using a lower-limb assisted exoskeleton based on visual sensors, its hardware cost is lower. Furthermore, since all inertial measurement units are placed on the human body, this method is unaffected by the external environment. Its high-precision motion capture capabilities and efficient dynamic response capabilities make it better suited for complex terrain recognition.

[0156] Those skilled in the art will appreciate that all or part of the process steps of the above-described embodiments can be implemented by instructing related hardware through a computer program, and the program can be stored in a computer-readable storage medium, such as a magnetic disk, an optical disk, a read-only memory, or a random access memory.

[0157] The above description is only a preferred specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any changes or substitutions that can be easily thought of by any technician familiar with this technical field within the technical scope disclosed in the present invention should be covered by the scope of protection of the present invention.

Claims

1. A terrain recognition method based on lower limb inertial information, characterized in that: include: Inertial measurement units are arranged on the left and right thighs, left and right calves, and lower back of the human body to be measured; calculating a hip joint angle based on output data of an inertial measurement unit of the left and right thighs and the lower back, constructing a self-oscillating sine curve based on the hip joint angle, and obtaining a first terrain recognition result based on the self-oscillating sine curve; performing energy peak detection based on acceleration data output by the inertial measurement unit of the rear waist, and obtaining a second terrain recognition result based on the detection result; Calculating the hip joint angle, knee joint angle, and forward lean angle of the human body at each moment during the current motion process based on output data from the inertial measurement units of the left and right thighs, left and right calves, and lower back, thereby constructing a hip joint angle curve, a knee joint angle curve, and a forward lean angle curve, obtaining amplitudes and offsets corresponding to the above curves, and comparing them with amplitude ranges and offset ranges of the upper body forward lean angle, hip joint angle, and knee joint angle corresponding to each terrain stored in an adaptive threshold knowledge base to obtain a third terrain recognition result; A final terrain recognition result is obtained based on the first terrain recognition result, the second terrain recognition result, and the third terrain recognition result.

2. The terrain recognition method based on lower limb inertial information according to claim 1, characterized in that: The hip joint angles include: a first hip joint angle calculated based on output data of an inertial measurement unit of the left thigh and the lower back; and a second hip joint angle calculated based on output data of an inertial measurement unit of the right thigh and the lower back. The self-oscillating sinusoidal curve includes: a first self-oscillating sinusoidal curve constructed based on the first hip joint angle, and a second self-oscillating sinusoidal curve constructed based on the second hip joint angle.

3. The terrain recognition method based on lower limb inertial information according to claim 2, characterized in that: The constructing of a self-oscillating sinusoidal curve based on the hip joint angle comprises: S1: Initialize the amplitude, frequency, phase and bias of the self-oscillating sinusoid; S2: obtaining the actual hip joint angle at the current moment, and calculating the tracking error based on the actual hip joint angle and the simulated hip joint angle output by the self-oscillating sine curve; S3: If the tracking error is less than the first target threshold, the current self-oscillating sinusoidal curve is used as the self-oscillating sinusoidal curve corresponding to the human body to be measured; if the tracking error is greater than or equal to the first target threshold, the amplitude, frequency, phase and offset of the self-oscillating sinusoidal curve at the current moment are updated based on the tracking error, and the updated values ​​are used as the amplitude, frequency, phase and offset of the self-oscillating sinusoidal curve at the next moment, and the process returns to step S2.

4. The terrain recognition method based on lower limb inertial information according to claim 3, characterized in that: The obtaining of a first terrain recognition result based on the self-oscillating sinusoidal curve includes: Comparing the amplitude, frequency, phase, and offset of the first self-oscillating sinusoidal curve with a preset first threshold classification table to obtain a first initial terrain recognition result, and comparing the amplitude, frequency, phase, and offset of the second self-oscillating sinusoidal curve with the preset first threshold classification table to obtain a second initial terrain recognition result; If the first terrain initial recognition result is consistent with the second terrain initial recognition result, the first terrain initial recognition result or the second terrain initial recognition result is used as the first terrain recognition result; if they are inconsistent, the irregular terrain is used as the first terrain recognition result; The first threshold classification table is used to store the amplitude range, frequency range, phase range and offset range corresponding to each type of terrain.

5. The terrain recognition method based on lower limb inertial information according to claim 3, characterized in that: The updating of the amplitude, frequency, phase and offset of the self-oscillating sinusoidal curve at the current moment based on the tracking error includes: ； in, is the updated phase, is the updated frequency, is the updated amplitude, is the updated bias, is the horizontal learning parameter, is the longitudinal learning parameter, is the tracking error.

6. The terrain recognition method based on lower limb inertial information according to claim 1, characterized in that: Obtaining a final terrain recognition result based on the first terrain recognition result, the second terrain recognition result, and the third terrain recognition result includes: If the first terrain recognition result, the second terrain recognition result, and the third terrain recognition result are the same, the first terrain recognition result, the second terrain recognition result, or the third terrain recognition result is used as the final terrain recognition result; If the first terrain recognition result, the second terrain recognition result, and the third terrain recognition result are different, constructing a recognition result set based on the terrain types appearing in each terrain recognition result, and calculating the probability of each terrain type in the recognition result set based on the weights of the first terrain recognition result, the second terrain recognition result, and the third terrain recognition result, and selecting the terrain type with the highest probability as the final terrain recognition result; The terrain recognition results include: flat land type, stair-up type, uphill type, downhill type, downhill type, and irregular terrain type.

7. The terrain recognition method based on lower limb inertial information according to claim 6, characterized in that: The calculating the probability of each terrain type in the recognition result set includes: ， Wherein, c is the terrain type in the recognition result set, is the probability value when the terrain type is c, 、 、 is the result value of terrain type c; when the terrain type c is consistent with the first terrain recognition result, ,otherwise, ; When the terrain type c is consistent with the second terrain recognition result, ,otherwise, ; When the terrain type c is consistent with the third terrain recognition result, ,otherwise, ; 、 、 are the weights corresponding to the first terrain recognition result, the second terrain recognition result, and the third terrain recognition result.

8. The terrain recognition method based on lower limb inertial information according to claim 1, characterized in that: The step of performing energy peak detection based on the acceleration data output by the inertial measurement unit of the lower back, and obtaining a second terrain recognition result based on the detection result, includes: Acquiring the acceleration data, taking a period in which the acceleration value in the acceleration data exceeds a preset detection threshold as a peak detection interval, and performing energy peak detection based on the peak detection interval to determine a touchdown point moment; Based on the output data of the inertial measurement unit at the time of contact and the limb lengths of the left and right thighs and left and right calves, a human lower limb model at the time of contact is constructed, the coordinates of the center of mass of the left and right feet are determined based on the human lower limb model, and a second terrain recognition result is obtained based on the coordinates of the center of mass of the left and right feet.

9. The terrain recognition method based on lower limb inertial information according to claim 1, characterized in that: The adaptive threshold knowledge base is used to store a first offset range of the upper body forward lean angle, a second amplitude range and a second offset range of the hip joint angle, and a third amplitude range and a third offset range of the knee joint angle corresponding to different terrains; The first offset range, the second amplitude range and the second offset range of the hip joint angle, and the third amplitude range and the third offset range of the knee joint angle are obtained as follows: For any terrain, obtain the upper body forward lean angle, hip joint angle, and knee joint angle of different people at different times, construct upper body forward lean angle curves, hip joint angle curves, and knee joint angle curves based on the upper body forward lean angle, hip joint angle, and knee joint angle curves at each time and their corresponding upper body forward lean angle, hip joint angle, and knee joint angle curves, and extract the amplitudes and offsets corresponding to the upper body forward lean angle curves, the hip joint angle curves, and the knee joint angle curves; Obtaining a first offset range based on the offset corresponding to the upper body forward lean angle curve of all persons; Obtaining a second amplitude range and a second offset range based on the amplitudes and offsets corresponding to the hip joint angle curves of all persons; A third amplitude range and a third offset range are obtained based on the amplitudes and offsets corresponding to the knee joint angle curves of all persons.

10. The terrain recognition method based on lower limb inertial information according to claim 9, characterized in that: The method compares the upper body forward lean angle, hip joint angle, and knee joint angle with the amplitude range and offset range stored in the adaptive threshold knowledge base to obtain a third terrain recognition result, including: For any terrain stored in the adaptive threshold knowledge base, if the bias of the forward tilt angle is within the first bias range, and the amplitude and bias of the first hip joint angle are respectively within the second amplitude range and the second bias range, or the amplitude and bias of the second hip joint angle are respectively within the second amplitude range and the second bias range, and the amplitude and bias of the first knee joint angle are respectively within the third amplitude range and the third bias range, or the amplitude and bias of the second knee joint angle are respectively within the third amplitude range and the third bias range, then the terrain is taken as the third terrain recognition result. If no matching terrain exists, the irregular terrain is taken as the third terrain recognition result.

Images