JP2021083562A - Information processing device, calculation method, and program - Google Patents

Abstract

To robustly estimate external force acting on a body during body exercise on the basis of body exercise data.SOLUTION: There is provided a walking exercise analysis device 110 for estimating floor reaction force generated during walking exercise. The walking exercise analysis device 110 includes: a body exercise data acquisition unit 112 for acquiring body exercise data; a floor reaction force initial setting unit 132 for setting a provisional value of floor reaction force; an analysis execution unit 134 for executing reverse dynamic analysis using a body model 116 on the basis of the body exercise data and the provisional value of the floor reaction force; a total load evaluation unit 136 for evaluating a load generated in the body according to a dynamic quantity generated in each joint of the body model 116 on the basis of an execution result of the reverse dynamic analysis; and a floor reaction force updating unit 140 for updating the provisional value of the floor reaction force on the basis of the evaluation result.SELECTED DRAWING: Figure 2

Description

The present invention relates to an information processing device, a calculation method and a program for dynamically analyzing body movements.

In order to dynamically analyze and evaluate physical movements such as walking, it is usually necessary to measure body movements and forces acting on the body. However, it is difficult to directly measure dynamic quantities in the living body such as joint moments. Therefore, the amount of dynamics in the living body is estimated by performing inverse dynamics analysis using a musculoskeletal model that numerically expresses the mechanical characteristics of the body.

In order to perform inverse dynamics analysis using a musculoskeletal model, in addition to the three-dimensional position of each part of the body obtained from the motion capture system, external force information obtained by a force sensor such as a floor reaction force meter is required. It becomes. However, it is not easy to install a floor reaction force meter for measuring the external force generated on the foot because it is necessary to embed it in the floor or prepare a table according to the height of the equipment. In addition, the installation of the floor reaction force meter may cause the subject to become aware of the measuring device and induce unnatural walking or running.

Against the background described above, various developments have been made to eliminate the need for a floor reaction force meter when walking. As a technique that enables analysis that does not require a floor reaction force meter, a technique for estimating the floor reaction force based on the minimization of the muscle activity state of the whole body (Non-Patent Document 1 and Non-Patent Document 2) and an artificial neural network are used. A technique for estimating the floor reaction force (Non-Patent Document 3) can be mentioned.

Non-Patent Document 1 has a configuration in which 12 contact points, each having 5 artificial muscles, are provided on the foot of a musculoskeletal model constructed based on the position of a body surface marker obtained by an optical motion capture system. To disclose. Then, the unknown floor reaction force is exploratoryly obtained by minimizing the muscle activity state of each part of the body, which is originally obtained by static optimization, by the optimization calculation. Here, each artificial muscle is provided to express the frictional force, and is defined to solve the statically indeterminate problem that occurs when both legs are supported.

Further, Non-Patent Document 2 further increases the contact points of the foot, and changes the strength information of the artificial muscle, which has been determined by trial and error by function approximation or the like, according to the speed and position information of each point. Disclose the improved technology to do so. In Non-Patent Document 2, the validity of the estimated value was verified by comparing the estimated value in the walking movement of 10 male subjects with the measured value, and it was shown that the estimated value was estimated with relatively high accuracy. ing. However, since the muscle characteristics vary widely among individuals, it may be difficult to estimate with high accuracy for physical exercises of female subjects, the elderly, and persons with disabilities.

Non-Patent Document 3 discloses a technique for developing a new technique capable of estimating the floor reaction force during a flat ground walking motion with high accuracy without using a floor reaction force meter. In Non-Patent Document 3, an artificial neural network model is applied to both leg phases of walking motion in order to solve the statically indeterminate problem. The method of Non-Patent Document 3 requires learning data.

However, in the above-mentioned conventional technique, it takes time and effort to prepare for applying the technique, and there is a concern that the estimation accuracy may be significantly different due to the difference in the characteristics of the subject, which is not sufficient.

R. Fluit, et al., "Prediction of ground reaction forces and moments during various activities of daily living", Journal of Biomechanics, Vol. 47, No. 10, pp. 2321-2329 (2014). Y. Jung, et al. , “Dynamically adjustable foot ground contact model to estimate ground reaction force during walking and running”, Gait & Posture, Vol. 45, pp. 62-68 (2016). E. S. Oh, et al., “Prediction of ground reaction forces during gait based on kinematics and a neural network model”, Journal of Biomechanics, 46 (14), 2372-2380 (2013).

The present invention has been made in view of the deficiencies in the above-mentioned prior art, and the present invention is information processing capable of robustly estimating an external force acting on the body during physical exercise based on physical exercise data. It is an object of the present invention to provide an apparatus, a calculation method and a program.

In the present invention, in order to solve the above problems, an information processing device having the following features for estimating an external force generated during physical exercise is provided. This information processing device executes inverse dynamics analysis using a body model based on the acquisition means for acquiring the physical exercise data, the setting means for setting the provisional value of the external force, and the physical exercise data and the provisional value of the external force. Based on the execution means and the execution result of the inverse dynamics analysis, the evaluation means for evaluating the load generated on the body according to the dynamic amount generated on each joint of the body model, and the provisional value of the external force based on the evaluation result. Includes update means to update.

The present invention also provides a calculation method having the following characteristics for estimating an external force generated during physical exercise. This calculation method includes a step in which the computer acquires physical exercise data and a step in which a provisional value of external force is set. In this calculation method, the computer further performs a reverse dynamics analysis using the body model based on the body movement data and the provisional value of the external force, and based on the execution result of the reverse dynamics analysis of the body model. It includes a step of evaluating the load generated on the body according to the dynamic amount generated in each joint, and a step of updating the provisional value of the external force based on the evaluation result.

The present invention further provides a program for realizing an information processing device having the above-mentioned characteristics for estimating the external force generated during the above-mentioned physical exercise.

With the above configuration, it is possible to robustly estimate the external force acting on the body during physical exercise based on the physical exercise data.

FIG. 1 is a functional block diagram showing an overall configuration of an analysis system including a walking motion analysis device according to an embodiment of the present invention. FIG. 2 is a more detailed functional block diagram of the optimization calculation unit in the walking motion analysis device according to the embodiment of the present invention. FIG. 3 is a flowchart showing a walking motion analysis process executed by the walking motion analysis device according to the embodiment of the present invention. FIG. 4 is a diagram illustrating a rigid body link model used in the walking motion analysis process according to the embodiment of the present invention. FIG. 5 is a diagram illustrating a contact point defined on the foot in the walking motion analysis process according to the embodiment of the present invention. FIG. 6 is a hardware block diagram showing a typical computer for realizing the walking motion analysis device according to the embodiment of the present invention. FIG. 7 is a diagram for explaining the environment of the measurement experiment of walking motion. FIG. 8 is a graph showing the measured value of the floor reaction force actually measured in the environment shown in FIG. 7 and the estimated value of the floor reaction force estimated by the walking motion analysis process shown in FIG. FIG. 9 is a graph showing the measured values of the floor reaction force action points actually measured in the environment shown in FIG. 7 and the estimated values of the floor reaction force action points estimated by the walking motion analysis process shown in FIG. is there. FIG. 10 shows the measured values of the floor reaction force measured in the environment shown in FIG. 7 and the estimated values of the floor reaction force estimated by the walking motion analysis process shown in FIG. 3 at each time point of the walking cycle in normal walking. It is a graph which shows the joint moment of the lower limbs.

Hereinafter, embodiments of the present invention will be described, but the embodiments of the present invention are not limited to the embodiments described below. In the embodiment described below, as an example of the information processing device for estimating the external force generated during physical exercise, the walking exercise is targeted, the floor reaction force generated during the walking exercise is estimated, and each joint of the body model is used. This will be described using the walking motion analyzer 110 for calculating the generated dynamics.

FIG. 1 is a functional block diagram showing an overall configuration of an analysis system 100 including a walking motion analysis device 110 according to the present embodiment. The analysis system 100 shown in FIG. 1 analyzes walking motion by receiving input of measurement data from a motion capture system 102 that measures the movement of a person or the like during walking exercise and the motion capture system 102. It is configured to include a walking motion analysis device 110.

The motion capture system 102 is a system that digitizes the movement of the body during operation based on any type of capture technology such as optical type, inertial sensor type, mechanical type, magnetic type or video type. Here, the optical type is a method in which a reflection marker is attached to the target body, a plurality of cameras are installed around the object, and the position of the reflection marker is optically detected, and there are an image type and an infrared type. The inertial sensor type is a method in which an inertial sensor including an angular velocity meter and an accelerometer is attached to each part of the body, and the position and posture are obtained by back calculation from the obtained acceleration and angular velocity information. The mechanical type is a method in which each joint angle is measured by using a sensor such as an encoder or a potentiometer that mechanically measures the angle of rotation or displacement. The magnetic type is a method in which a magnetic sensor is attached to a body, a magnetic field is applied from the magnetic generator, and the position and orientation from the magnetic generator are obtained based on the magnetic field lines detected by the coil in the magnetic sensor. The video method is a method of analyzing a video image to obtain position information. Hereinafter, for convenience, the case where the optical method, which is a standard method, is adopted will be mainly described.

FIG. 1 also shows a detailed functional block of the gait motion analysis device 110. As shown in FIG. 1, the walking motion analysis device 110 has a physical motion data acquisition unit 112 that acquires physical motion data from the motion capture system 102, and a predetermined floor reaction force based on the acquired physical motion data. Underneath, the inverse dynamics analysis unit 120 that performs the inverse dynamics calculation, and the optimization calculation unit that estimates the unknown floor reaction force to be obtained by the optimization calculation based on the calculation results by the inverse dynamics analysis unit 120. It is configured to include 130.

The physical exercise data acquisition unit 112 acquires physical exercise data 114 in a predetermined format that describes the physical exercise. Here, the body movement data directly obtained from the motion capture system 102 is the movement displacement data of the body in the case of the above-mentioned optical type. On the other hand, here, the body movement data 114 according to the predetermined body model 116 is acquired in order to perform the inverse dynamics calculation described later.

The body model 116 is, in certain embodiments, a rigid body link model. The rigid body link model is a model in which each body segment is assumed to be approximated by a rigid body node (link) that does not deform, and the body dynamic structure is expressed as a rigid body link model in which rigid body links are connected.

The body motion displacement measurement data obtained from the marker measurement is three-dimensional coordinates in the absolute coordinate system (spatial coordinate system) of the position of the marker provided on the body surface. The body motion data acquisition unit 112 acquires motion displacement measurement data from the motion capture system 102, and further, from the three-dimensional coordinates in the absolute coordinate system (spatial coordinate system) of the marker, for example, the global optimization method or the point cluster method. Based on, the three-dimensional movement of each body segment, that is, the joint angle and the like are obtained. In the above-mentioned inverse dynamics calculation, in addition to the motion displacement such as the joint angle, the angular velocity and the angular acceleration are also used. Therefore, in the case of the optical type, the motion displacement is numerically differentiated from these velocity and acceleration information. Calculate by Depending on the motion capture method, it may be possible to omit a part of the above-mentioned joint angle calculation and velocity and acceleration calculation. The body movement data acquisition unit 112 uses data obtained directly from the motion capture system 102 to determine joint angles, angular velocities, and angular accelerations based on kinematic characteristics such as node length and joint freedom of each body node of the rigid body link model. The kinematics such as and the like are calculated, and the physical exercise data 114 according to the predetermined body model 116 is obtained.

The inverse dynamics analysis unit 120 performs inverse dynamics calculation of the equation of motion of the body model 116 under a predetermined floor reaction force based on the acquired body motion data (joint angle, angular velocity, angular acceleration, etc.) 114. Therefore, the dynamics of each part of the body, more specifically, the joint moment (also referred to as joint torque and muscle torque, which are synonymous with each other) and the joint reaction force generated in each joint of the body model 116 are calculated. External force information such as floor reaction force is generally used in the inverse dynamics calculation, but in the embodiment of the present invention, a provisional value 118 of the floor reaction force is assumed, and the value is lower than the provisional value 118. Then, the inverse dynamics calculation is performed. By executing the inverse dynamics calculation, the estimated value 122 of each joint moment is calculated with respect to the provisional value of the assumed floor reaction force.

Here, the floor reaction force used in the inverse dynamics calculation includes, more specifically, the floor reaction force vector representing the magnitude and direction of the floor reaction force and the point of action of the floor reaction force. In addition, walking exercise has a one-leg support period in which either the left or right foot touches the floor and a two-leg support period in which both the left and right feet touch the floor. In the one-leg support period, the floor reaction force is that of the foot that touches the ground, and in the two-leg support period, the floor reaction force is typically set separately for each of the left and right feet.

The optimization calculation unit 130, in cooperation with the inverse dynamics analysis unit 120, optimizes and calculates an unknown floor reaction force 118 (floor reaction force vector and floor reaction force action point (COP: Center Of Pressure)) to be obtained. Estimated by. The optimization calculation unit 130 gives a provisional value 118 of the predetermined floor reaction force, causes the inverse dynamics analysis unit 120 to execute the inverse dynamics analysis, evaluates the analysis result 122 obtained as a result, and evaluates the next optimization. Repeatedly updating the provisional value 118 of the floor reaction force used in the calculation to search for the floor reaction force such that the predetermined objective function is minimized (when defined as a penalty) or maximized (when defined as a reward). To do. Based on the obtained analysis results, the direction in which the objective function is improved in the search space can be obtained. By repeating this inverse dynamics analysis, evaluation, and updating of the provisional value of the floor reaction force, the estimated value of the floor reaction force is optimized and obtained as the analysis result. At the same time, the estimated value 122 of the dynamics such as the joint moment and the joint reaction force of the body model 116 when the optimized floor reaction force estimate is given is also given as the analysis result.

Once the optimal values of the dynamics such as joint moment and joint reaction force of the body model 116 are obtained, the optimization calculation method is combined with the musculoskeletal model and the musculoskeletal geometry model to further minimize the muscle load. By applying it, it is possible to calculate the in vivo load of the musculoskeletal system such as muscle strength. Here, the muscular dynamics model is a model that expresses the mechanical characteristics associated with the exertion of force of individual muscles, and the musculoskeletal geometry model is a muscle attachment position, muscle running (muscle geometric arrangement, path), etc. It is a model that represents. A musculoskeletal model is constructed by combining a rigid body link model, a musculoskeletal model, and a musculoskeletal geometry model. It should be noted that how to use the mechanical quantity of the obtained body model 116 (rigid body link model) is not particularly limited. Further, not only the mechanical quantity of the body model 116 (rigid body link model) but also the finally obtained floor reaction force vector and the point of action are useful information. In the embodiment of the present invention, the estimation of the external force such as the floor reaction force used in the inverse dynamics calculation will be mainly described. Therefore, the subsequent processing for estimating the floor reaction force vector and the point of action will be further described. Do not enter.

FIG. 2 shows a more detailed functional block of the optimization calculation unit 130 shown in FIG. As shown in FIG. 2, the optimization calculation unit 130 includes a floor reaction force initial setting unit 132, an analysis execution unit 134, a total load evaluation unit 136, an end determination unit 138, and a floor reaction force update unit 140. Consists of.

The floor reaction force initial setting unit 132 first sets a provisional value of the floor reaction force as an initial value. The initial value is not particularly limited, but an arbitrary fixed value may be set, or a random number generated within a predetermined range may be set using a random number function. Here, the floor reaction force given to the inverse kinetic analysis includes the floor reaction force vector (direction and magnitude) and the point of action of the floor reaction force, and the provisional value of the floor reaction force is the floor reaction force vector. And the provisional value of the point of action coordinates is included.

The analysis execution unit 134 gives the acquired body movement data and the provisional value of the floor reaction force currently set to the inverse dynamics analysis unit 120, executes the inverse dynamics analysis using the body model 116, and reverses. Obtain the execution result of the dynamic analysis. The results of the inverse dynamics analysis include estimates of the dynamics that occur in each joint. The total load evaluation unit 136 calculates the total amount of dynamics generated in each joint of the body model based on the execution result of the inverse dynamics analysis by the inverse dynamics analysis unit 120, that is, based on the estimated value of this dynamics. , Evaluate the load on the body. The floor reaction force updating unit 140 updates the provisional value of the floor reaction force based on the evaluation result by the total load evaluation unit 136.

Under predetermined conditions, the inverse dynamics analysis unit 120 executes the inverse dynamics analysis, the total load evaluation unit 136 evaluates, and the floor reaction force updating unit 140 updates the provisional value of the floor reaction force. The objective function is minimized (when defined as a penalty) or maximized (when defined as a reward) and the floor reaction force is optimized.

The end determination unit 138 determines whether or not the end condition of the inverse dynamics analysis by the inverse dynamics analysis unit 120 is satisfied. For example, conditions for detecting a predetermined number of censoring times and convergence of the objective function are used. When the end determination unit 138 determines that the end condition is satisfied, the analysis is terminated to obtain the final value of the mechanical quantity generated in each joint of the body model 116 and the final value of the floor reaction force.

FIG. 2 also shows a preferred embodiment, and as shown in FIG. 2, in the preferred embodiment, the optimization calculation unit 130 further includes a grounding determination unit 142 and an inside / outside determination unit 144. .. In a preferred embodiment, the predetermined conditions for the above-mentioned optimization include a constraint for keeping the point of action of the floor reaction force within the contact patch of the foot. By introducing this constraint, it is possible to reduce the possibility that the position of the action point of the floor reaction force is estimated outside the sole of the foot.

As described above, the motion displacement measurement data holds the three-dimensional coordinate positions of the markers provided on the body surface, but in a preferred embodiment, a plurality of markers are further interpolated so as to interpolate the plurality of markers constituting the foot. Contact point is defined. The ground contact determination unit 142 makes a ground contact determination for each contact point defined on the foot based on the height of the contact point and the speed of the contact point. Then, a contact surface is provided as a convex hull composed of contact points determined to be in contact with the ground.

The inside / outside determination unit 144 determines whether or not the provisional value of the action point is located inside the ground contact surface obtained based on the ground contact determination of the contact point. The constraint for keeping the point of action of the floor reaction force in the ground plane of the foot may be incorporated as a constraint equation (constraint condition) of the optimization calculation, or may be incorporated into the objective function of the optimization calculation. Good. In certain embodiments, as a penalty term given when the provisional value of the point of action is located outside the tread, or as a reward term given when the provisional value of the point of action is located inside the tread. It is incorporated into the objective function of the optimization calculation.

Hereinafter, the walking motion analysis process including the estimation of the floor reaction force during the walking motion according to the present embodiment will be described in more detail with reference to FIG. FIG. 3 is a flowchart showing a walking motion analysis process executed by the walking motion analysis device 110 according to the embodiment of the present invention. The walking motion analysis process shown in FIG. 3 is executed by a processor such as a CPU included in the walking motion analysis device 110. Further, the process shown in FIG. 3 is based on the physical exercise data at a predetermined time point during the walking exercise, and in order to analyze one walking cycle, the heel touches the heel of one foot, the heel separates, and the toe separates. The same process will be performed at each point in time until the heel touches down again. In that case, the initial value of the calculation at the next time point may be determined based on the result obtained at the previous time point.

The walking motion analysis process shown in FIG. 3 is started from step S100, for example, in response to an instruction from the operator. In step S101, the processor reads the body model 116.

FIG. 4 is a diagram illustrating a rigid body link model 200 as a body model 116 used in the walking motion analysis process according to the embodiment of the present invention. As shown in FIG. 4, the rigid body link model 200 includes a plurality of links 202 and joint portions 204 that connect the links to each other, and each joint portion 204 is predetermined in advance according to the joint movable characteristics. The degree of freedom is set. For example, the hip joint has three degrees of freedom of flexion / extension, adduction / abduction, and rotation, and the knee joint and ankle joint have three degrees of freedom of rotation, and basic motor restraints are set.

Referring again to FIG. 3, in step S102, the processor acquires measurement data from the motion capture system 102 and acquires physical exercise data at a predetermined time point during walking exercise. The body movement data is converted into a format including information such as joint angle, angular velocity and angular acceleration of each joint of the body model 116. In step S103, the processor sets the floor reaction force vector and the provisional value of the action point of the floor reaction force as initial values. The initial value is set to a fixed value, a random value, or an optimum value at a time in the past. After that, in steps S104 to S107, the optimization calculation is executed using a predetermined objective function and constraint expression.

In step S104, the processor performs an inverse kinetic analysis of the body model 116, given the acquired body movement data and the current floor reaction force vector and provisional values of the floor reaction force action points. In step S105, the processor evaluates the load generated on the body according to the mechanical quantity generated on each joint of the body model based on the execution result of the inverse dynamic analysis. More specifically, in step S105, the following objective function I is evaluated. The objective function I is configured with the load as a penalty.

Here, f _pelvis showed joint reaction forces occurring pelvic nodes corresponding to the degree of freedom with respect to the spatial coordinates in the inverse dynamics calculation, n _i denotes the joint moment occurring on the joint i, f _grf is the body Represents the floor reaction force, which is an external force acting on. The hat attached to each symbol indicates that the hat is normalized based on the Hof method. In this normalization, the force is normalized by the weight given as the physique condition, and the moment may be the weight given as the physique condition and the length of each part such as the leg length (the length of each part such as the foot length, which is also the physique condition). It is normalized by the product of). Moreover, in a specific embodiment, the normalization process for the speed change may be performed. In the case of walking exercise, the number of external forces is 2 on the left and right feet. In the above equation, a ₁ , a ₂ , and a ₃ are optimization weighting coefficients. The parameters a ₁ , a ₂ , and a ₃ are typically set to arbitrary values that satisfy a predetermined condition.

The objective function I is based on the assumption that "walking is performed so that the load generated on each part of the human body is minimized", and evaluates the sum of the mechanical quantities obtained by the inverse dynamics calculation. is there. The optimization calculation is formulated as the following equations (1) and (2).

Regarding the first condition of the above equation (2), the reaction force f _pervis of the pelvic node is ideally zero, but it does not actually become zero due to the influence of modeling and measurement error, so the weight is larger than other terms. By doing so (a ₁ > a ₂ > a ₃ ), the effect is minimized. Further, each term of the above equation (1) is a sum of cubes of absolute values in order to show the characteristics of each term prominently. Although not limited to the sum of cubes, it is preferable to use a non-linear monotonically increasing function.

The second condition of the above equation (2) is _{a constraint for keeping} the floor reaction force action point position pgrf (x, y) in the ground contact region C of the foot. The foot contact area C is obtained by defining contact points based on the position of the marker on the foot and making a ground contact determination for each contact point.

FIG. 5 is a diagram illustrating a contact point defined on the foot in the walking motion analysis process according to the embodiment of the present invention. As shown in FIG. 5, a plurality of contact points 214 are defined so as to interpolate a plurality of markers 212 constituting the foot portion 210. Then, the ground contact discriminant represented by the following equation (3) is applied to each of the defined contact points 214, and the foot ground contact region C is obtained as a convex hull composed of the contact points determined to be grounded. Be done.

In the above equations (3) and (4), the subscript grf indicates the external force number, j indicates the contact point number, p indicates the position of the contact point, v indicates the velocity of the contact point, and is attached to p and v. The letter thres indicates that it is a threshold value for the contact point position and the contact point velocity, and x, y and z attached to p and v indicate that they are components in the traveling direction, the left-right direction and the vertical direction, respectively. f represents the estimated floor reaction force vector. In (3) above, it is determined that the contact point where the height is equal to or less than the predetermined reference and the speeds in the horizontal direction and the vertical direction are smaller than the reference is grounded. The contact area can be defined when the number of contact points is 3 or more per foot. Therefore, according to the above equation (4), the number of contact points is counted and the floor reaction force is applied when the number of contact points becomes 3 or more.

_{In a particular embodiment, the constraint for keeping} the floor reaction force action point position p frg (x, y) in the foot ground contact area C is that the action point position p _frg (x, y) is outside the ground contact area C. In some cases, it can be incorporated into the optimization calculation by adding the penalty shown in the following equation (5) to the objective function of the above equation (1) as the fourth term.

In the above equation (5), a is a weighting coefficient, and d _{represents the distance from the center of the foot contact area C to the action point position pgrf} (x, y). By adding a penalty according to the distance of the position of the point of action from the center, it is possible to prevent the value from diverging.

The predetermined conditions in the optimization calculation described above may further include a constraint for minimizing the distance between the point of action of the external force and the zero moment point (ZMP). With reference to the above equation (2) again, the third condition of the above equation (2) represents a constraint that minimizes the distance between the estimated floor reaction force action point position and the zero moment point. ZMP is a point on the floor where the moment generated by inertial force and gravity at the center of gravity of the body becomes zero, and it is said that this coincides with the position of the action point of the floor reaction force when walking on a plane. In the one-leg support period, the distance between the ZMP and the action point of the floor reaction force on the one-leg side to be supported is evaluated, and in the two-leg support period, the action point on which the resultant force of the floor reaction force of the left and right feet acts and the ZMP. The distance to and can be evaluated. The zero moment point is obtained by recursively finding the position of the center of gravity of each body segment based on the measurement data from the motion capture system 102, and finding the acceleration of the center of gravity of each body segment from the position of the center of gravity. It is derived from the acceleration of the center of gravity of each body segment.

With reference to FIG. 3 again, in step S106, the processor determines whether the termination condition is satisfied. When the convergence condition such as reaching the predetermined upper limit of censoring or the improvement of the objective function cannot be expected any more is satisfied, it is determined that the end condition is satisfied. If it is determined in step S106 that the optimization has not been completed yet, the process is branched to step S107. In step S107, the processor updates the provisional value of the floor reaction force based on the result of the evaluation described above, preferably in the direction of improvement in which the objective function decreases.

If it is determined in step S106 that the end condition is satisfied, the process is branched to step S108. The floor reaction force (floor reaction) (floor reaction) is repeated by repeating the execution of the inverse dynamics calculation in step S104, the evaluation of the objective function in step S105, and the update of the provisional value of the floor reaction force in step S107 until the end condition is satisfied in step S106. The value of the force vector and the point of action) is optimized. In step S108, the processor takes the provisional value of the current floor reaction force as the final value, and the dynamics such as the joint moment and the joint reaction force as a result of the inverse dynamics calculation when the floor reaction force is given. Output the value as the final value. In step S109, this process ends.

Hereinafter, the hardware configuration of the walking motion analysis device 110 will be described. The gait motion analysis device 110 is implemented by a general-purpose computer in a specific embodiment. FIG. 6 is a hardware configuration diagram of the walking motion analysis device 110 according to the present embodiment. The gait motion analyzer 110 enables program processing by a single-core or multi-core microprocessor unit (MPU) 12, a non-volatile memory 14 that stores a BIOS (Basic Input Output System), and an MPU 12 on a board 10. Includes a memory 16 that provides an execution storage space.

The MPU 12 is connected to the storage control interface 18 via the internal bus 22, and the hard disk 20 executes data writing or reading in response to an input / output request from the MPU 12. The MPU 12 controls a serial or parallel interface 24 such as USB via the internal bus 22 to communicate with an input / output device 26 such as a keyboard, a mouse, and a printer, and receives an input from a user.

The gait motion analysis device 110 can further include a VRAM 28 and a graphic chip 30. The graphic chip 30 processes the video signal in response to the command from the MPU 12 and displays it on the display device 32. The MPU 12 also connects to the network I / F (NIC; Network Interface Card) 34 via the internal bus 22. As a result, the walking motion analysis device 110 is communicated with an external device such as a motion capture system through the network.

The walking motion analysis device 110 is stored in a storage device such as a non-volatile memory 14, a hard disk 20 (or SSD (solid state drive)), NV-RAM (not shown), or an SD card (not shown). A program (not shown) is read and expanded in the memory area of the memory 16. As a result, each of the above-mentioned functional means and each process is realized under an appropriate operating system (OS). The OS includes Windows (registered trademark), UNIX (registered trademark) or LINUX (registered trademark), android (registered trademark), iOS (registered trademark), MacOS (registered trademark), iPadOS (registered trademark), and the like. An OS having an architecture can be adopted.

The above functional unit can be realized by a computer-executable program described in a legacy programming language such as an assembler, C, C ++, C #, Java (registered trademark), an object-oriented programming language, or the like. Stored in device-readable recording media such as EPROM, flash memory, flexible disc, CD-ROM, CD-RW, DVD-ROM, DVD-RAM, DVD-RW, Blu-ray disc, SD card, MO, or telecommunications line Can be distributed through.

As described above, according to the embodiment of the present invention, there is provided an information processing device, a calculation method and a program capable of robustly estimating the floor reaction force acting on the body during walking exercise based on the body exercise data. can do.

The algorithm for estimating the floor reaction force described above is based on the assumption that "humans are walking so as to minimize the load generated on the body", and the body model 116 expressing the kinematic characteristics and inertial characteristics of the body is used. By minimizing the sum of the dynamics obtained from the inverse kinematics calculation based on it, the variable and unknown floor reaction force and the position of its point of action are searched for.

Since the motion information required for mechanical analysis is only physical motion data and no external force information is required, a force sensor such as a floor reaction force meter is provided for exercise in an environment where a camera for motion capture can be installed. It can be analyzed without it. In addition, since the estimation method uses the inverse dynamics calculation of the body model constructed based on the marker position, uniform estimation is possible regardless of the subject. As a result, uniform estimation can be expected regardless of the measuring device and the subject.

In a particular embodiment, a large number of contact points are provided in the sole of the foot from the position of the foot marker, to which a ground contact discriminant consisting of height and velocity is applied. Then, the contact surface is formed only by the contact points that come into contact with the ground, and the inside / outside determination of the estimated action points is performed on this surface. When the point of action is determined to be outside, the position of the point of action with the minimum in-vivo load can be estimated in the sole of the foot by adding the constraint amount according to the amount to the optimization calculation formula.

In the above description, human walking has been the subject of analysis, but the subject of analysis is not limited to humans. Even objects such as animals and robots can be analyzed.

Verification experiment Below, the above-mentioned floor reaction force estimation algorithm was implemented and applied to the actually measured physical exercise data during walking, and a verification experiment was conducted. The verification experiment will be described below.

In order to verify the validity of the floor reaction force estimated by the above-mentioned floor reaction force estimation algorithm and the position of the point of action, an actual measurement experiment of walking motion was performed. The subjects were 6 healthy adult males, and the measuring device was an optical motion capture system (Optitrac, Natural Point) and a floor reaction force meter (TF-406-D, Tech Gihan). It was installed as shown in 7. The floor reaction force meter was used to obtain the measured value to be compared with the estimated value. In addition, since it is difficult to measure the end point of one walking cycle with one floor reaction force meter, a mat switch (OM-1074, Osaka Automatic Electric Co., Ltd.) is used to determine the point of heel contact, which is the end point of one walking cycle. I measured it. In FIG. 7, the camera of the optical motion capture system is indicated by 308 and the floor platform is indicated by 304. Reference numeral 306 is a mat switch, and reference numeral 302 denotes a table for forming a surface flush with the location where the floor reaction force meter is provided.

The measurement frequency of each device was unified to 100 Hz, the measurement signal was passed through a low-pass filter for smoothing, and the cutoff frequency was set to 10 Hz. In the verification experiment, normal walking, fast walking, and slow walking on the walking path of each subject were measured in three trials. The degree of slowness with respect to normal walking at this time was determined by each subject. As an index for comparing the estimated value and the measured value, the coefficient of determination indicating the similarity and the mean square error indicating the magnitude of the error were used.

In addition, based on the three-dimensional positions of each part of the body obtained by the optical motion capture system, the positions and coordinate systems of each joint of the body were defined, and a body model with 18 segments and 47 degrees of freedom of rotation was constructed. .. In addition to the arrangement of VICON (registered trademark) Plug-in-Gait, the markers to be measured were attached to a total of 51 parts of the body with reference to the VICON (registered trademark) Oxford Foot Model specialized for the foot. Further, as shown in FIG. 5, the positions of the contact points were arranged so that each contact point was located at the midpoint of the marker and other contact points, and a total of 31 contact points were provided.

Results of verification experiment Compare the estimated value of the floor reaction force and its position of action with the measured value. FIG. 8 is a graph showing the measured value of the floor reaction force actually measured in the environment shown in FIG. 7 and the estimated value of the floor reaction force estimated by the walking motion analysis process shown in FIG. FIG. 9 is a graph showing the measured values of the floor reaction force action points actually measured in the environment shown in FIG. 7 and the estimated values of the floor reaction force action points estimated by the walking motion analysis process shown in FIG. is there. 8 and 9 show the estimated value and the measured value of the floor reaction force and the position of the point of action at each walking speed in one subject. On the horizontal axis, one walking cycle was taken from the time when the right foot touched the floor reaction force meter until the foot on the same side touched the mat switch. In addition, the position of the point of action shows the locus during the stance period. Table 1 below summarizes the comparison results between the estimated values and the measured values using the evaluation index.

The upper part of Table 1 above shows the evaluation of this experiment. If the coefficient of determination is close to 1 and the mean square error is close to 0, it can be said that the estimated value is close to the measured value.

From the above results, first, regarding the floor reaction force, the anteroposterior and vertical components were estimated to be close to the measured values in the similar behavior. Regarding the magnitude of the error, the error was relatively small in the front-back direction and the vertical direction, but the error was large in the vertical direction as compared with other components, and the error increased as the walking speed increased. Regarding the position of the point of action, the error in the front-back direction was larger than that in the left-right direction where there was little movement. Regarding the difference in estimation accuracy between the subjects, there was no significant difference in the floor reaction force. Regarding the position of the point of action, there were variations among the subjects, but in the subjects shown in FIGS. 8 and 9, the value of the coefficient of determination in the anteroposterior direction was 0.7 or more, showing relatively high estimation accuracy. The magnitude of the error also varied among the subjects, but in the examples of FIGS. 8 and 9, the error was about 0.03 m.

For both the floor reaction force and the position of the point of action, the reason why the estimation accuracy decreases as the walking speed increases is that each term of the objective function of the optimization calculation is not normalized for the speed change. Conceivable. Since each dynamic quantity of the objective function is normalized based on Hof's method, the optimization calculation is less likely to be affected by the difference in the physique information of the subject. However, when the same subject walks at a speed higher than the normal walking speed, the burden on each joint of the body increases, so that the value of the objective function increases as compared with the low speed. Therefore, it is considered that the higher the walking speed, the lower the estimation accuracy.

Regarding the position of the action point, the value of the error variation in the left-right direction of the action point position is very close to the value of the average miscalculation compared to the floor reaction force and the action point position in the front-back direction, which is ZMP. It is considered that the estimation error of is affected. The derivation of ZMP is calculated from the position of the center of gravity of the body segment obtained recursively based on the motion capture data. Therefore, it is considered that the error at that time has an influence. In addition, since there is less change in the position of the point of action in the left-right direction than in the front-back direction, it is considered that the error in the acceleration of the center of gravity may affect the result.

Using the floor reaction force estimated here and the position of the point of action, the joint moment in the flexion / extension direction at each joint of the lower limbs is calculated and compared with the joint moment calculated from the actually measured floor reaction force.

FIG. 10 shows the joint moment based on the estimated value and the measured value of the floor reaction force in the normal walking of the same subject as in FIGS. 8 and 9, which showed the highest estimation accuracy. From FIG. 10, it can be seen that the error occurring in the latter half of the stance phase in the ankle joint affects the knee joint and the hip joint. It is expected that the estimation accuracy will be improved by normalizing the joint moment of the objective function according to each joint and body segment. In the current normalization method, each joint moment is divided by a formula using body weight and leg length, so by normalizing by the length and weight of the segment connected to the joint, the priority of minimization can be set. It is possible to unify each joint, which may lead to improvement in estimation accuracy.

If there is measurement data from the motion capture system, it is possible to perform mechanical evaluation of physical exercise without using the information of the force sensor, so it is possible to live in facilities where elderly people and people with disabilities who have difficulty moving live. It enables mechanical analysis of body movements in places where it was difficult to install floor reaction force meters, such as factories where many machine and production machines are installed.

Although the embodiments of the present invention have been described so far, the embodiments of the present invention are not limited to the above-described embodiments, and those skilled in the art may think of other embodiments, additions, changes, deletions, and the like. It can be changed within the range possible, and is included in the scope of the present invention as long as the action / effect of the present invention is exhibited in any of the embodiments.

100 ... analysis system, 102 ... motion capture system, 110 ... walking motion analysis device, 112 ... body movement data acquisition unit, 114 ... body movement data, 116 ... body model, 118 ... provisional value of floor reaction force, 120 ... Reverse dynamics analysis unit, 122 ... Estimated value of dynamics, 130 ... Optimization calculation unit, 132 ... Floor reaction force initial setting unit, 134 ... Analysis execution unit, 136 ... Total load evaluation unit, 138 ... Convergence judgment unit, 140 ... Floor reaction force update part, 142 ... Grounding judgment part, 144 ... Inside / outside judgment part, 200 ... Rigid body link model, 202 ... Link, 204 ... Joint part, 210 ... Foot part, 212 ... Foot marker, 214 ... Contact point, 302 ... stand, 304 ... floor reaction force meter, 306 ... mat switch, 308 ... camera, 310 ... body, 10 ... board, 12 ... MPU, 14 ... BIOS, 16 ... memory, 18 ... storage control interface, 20 ... hard disk Drive (HDD), 22 ... Internal bus, 24 ... Interface, 26 ... Input / output device, 28 ... VRAM, 30 ... Graphic chip, 32 ... Display device, 34 ... NIC

Claims (12)

An information processing device for estimating the external force generated during physical exercise.
Acquisition method for acquiring physical exercise data,
Setting means for setting the provisional value of the external force and
An execution means for performing a reverse dynamics analysis using a body model based on the body movement data and the provisional value of the external force.
Based on the execution result of the inverse dynamic analysis, an evaluation means for evaluating the load generated on the body according to the mechanical quantity generated on each joint of the body model, and an evaluation means.
An update means for updating the provisional value of the external force based on the result of the evaluation, and
Information processing equipment, including. The information processing according to claim 1, wherein the execution result of the inverse dynamics analysis includes an estimated value of the dynamics generated in each joint, and the evaluation by the evaluation means is performed based on the estimated value of the dynamics. apparatus. The value of the external force is optimized by repeating the execution of the inverse dynamic analysis by the executing means, the evaluation by the evaluation means, and the update of the provisional value of the external force by the updating means under a predetermined condition. Item 2. The information processing apparatus according to item 1 or 2. The third aspect of claim 3, wherein the external force includes a direction and magnitude of the external force and a point of action on which the external force acts, and the predetermined condition includes a constraint for keeping the point of action within the contact patch of the foot. Information processing device. The information processing device includes a grounding determination means for determining the grounding of a contact point defined on the foot.
The provisional value of the action point is located inside, including an inside / outside determination means for determining whether or not the provisional value of the action point is located inside the ground contact surface obtained based on the ground contact determination of the contact point. The information processing apparatus according to claim 4, wherein whether or not the information processing device is reflected in the evaluation by the evaluation means. The contact point is defined to interpolate a plurality of markers constituting the foot, and the ground contact determination is performed for each contact point based on the height of the contact point and the velocity of the contact point. The information processing apparatus according to 5. The information processing apparatus according to any one of claims 4 to 6, wherein the predetermined condition further includes a constraint for minimizing the distance between the action point of the external force and the zero moment point. The information processing device
The end determination means for determining the end of the inverse dynamics analysis is further included, and when the end determination means determines that the analysis is completed, the final value of the mechanical quantity generated in each joint of the body model and the said. The information processing apparatus according to any one of claims 1 to 7, wherein the final value of the external force can be obtained. The body movement is a walking movement, the external force is a floor reaction force, the body model includes a rigid body link model representing a physical dynamic structure as a model in which rigid body links are connected, and the mechanical quantity is a joint moment. The information processing apparatus according to any one of claims 1 to 8, further comprising. It is a calculation method for estimating the external force generated during physical exercise, and the computer
Steps to get physical exercise data and
The step of setting the provisional value of the external force and
A step of performing a reverse dynamics analysis using a body model based on the body movement data and the provisional value of the external force, and
Based on the execution result of the inverse dynamics analysis, a step of evaluating the load generated on the body according to the mechanical quantity generated on each joint of the body model, and a step of evaluating the load generated on the body.
A calculation method including a step of updating the provisional value of the external force based on the result of the evaluation. The calculation method according to claim 10, wherein the calculation method includes a step of optimizing the value of the external force by repeating the step of executing, the step of evaluating, and the step of updating under predetermined conditions. Calculation method. A program for realizing an information processing device for estimating the external force generated during physical exercise, using a computer.
Acquisition method for acquiring physical exercise data,
Setting means for setting the provisional value of the external force,
An execution means for performing inverse dynamics analysis using a body model based on the body movement data and the provisional value of the external force.
An evaluation means for evaluating the load generated on the body according to the mechanical quantity generated in each joint of the body model based on the execution result of the inverse dynamic analysis, and the provisional value of the external force based on the evaluation result. Update means to update,
A program to function as.

Images