JP7479308B2 - Walking measurement system - Google Patents

Description

The present invention relates to a gait measurement system.

As part of health promotion and nursing care prevention projects associated with the aging population, walking training is being implemented at welfare facilities and other facilities. One example of an exercise test that can evaluate functional mobility, combining walking ability, dynamic balance, agility, etc., is the Timed Up and Go (TUG) test. In the TUG test, the subject stands up from a seated position in a chair, turns around a marker, for example, 3 m away, and sits back down in the chair.

In recent years, a gait measurement system described in Patent Document 1 has been developed as a system for tracking the leg positions of a walking subject in such walking tests. The gait measurement system includes a distance information acquisition unit that emits detection waves to scan along the horizontal direction and acquires distance information related to the distance to an object over time based on the reflection state of the detection waves, and a data analysis unit that acquires the movement trajectory of the subject's legs based at least on the distance information. The data analysis unit detects observed values that are candidate leg positions from the distance information, and associates the observed values with the legs (correlation processing).

JP 2016-137226 A

In the above-mentioned technology, the predicted positions of the subject's legs are calculated from observation values at past times, and a gate is set that expands based on the predicted positions. The movement trajectory of the subject's legs is obtained based on the observation values within the set gate. In this case, however, there is a risk that the leg positions of another person, such as an attendant accompanying the subject, may be mistakenly tracked as the leg positions of the subject.

The objective of the present invention is to provide a gait measurement system that can accurately track the position of a subject's legs.

The gait measurement system according to the present invention is a gait measurement system that tracks the leg positions of a walking subject, and includes a distance information acquisition unit that emits a detection wave to scan along the horizontal direction and acquires distance information over time regarding the distance to an object that reflected the detection wave based on the reflection state of the detection wave, and a data analysis unit that acquires the movement trajectory of the subject's legs based at least on the distance information acquired by the distance information acquisition unit. The data analysis unit includes a leg detection unit that detects one or more observation values that are candidate leg positions based on the distance information, a predicted position calculation unit that calculates predicted positions of both legs of the subject based at least on observation values at past times, and a correlation processing unit that associates the observation values with the predicted positions of both legs. The correlation processing unit sets a first gate that expands based on the predicted positions of both legs, and sets a second gate that expands based on the predicted positions of both legs according to the observation values at past times, and associates the predicted positions of both legs with observation values that exist in the area where the first gate and the second gate overlap.

In this gait measurement system, even if the leg position of another person, such as an attendant, is erroneously detected as an observed value, the first gate can prevent the observed value from being associated with the predicted positions of both legs, and the second gate can also prevent this. Therefore, it is possible to prevent the leg positions of another person from being erroneously tracked as the leg positions of the subject, and it is possible to track the leg positions of the subject with high accuracy.

In the gait measurement system according to the present invention, the correlation processor may set a second gate extending from a position between the predicted positions of the right leg and the left leg when there are both observed values associated with the subject's right leg and left leg at the immediately preceding processing time. In this case, the second gate can be set centered on a position between the predicted positions of both legs with high accuracy.

In the gait measurement system according to the present invention, the correlation processor may set a second gate that extends based on the predicted position of one of the legs when there is only an observation value associated with one of the legs at the immediately preceding processing time. In this case, the second gate can be set based on the predicted position of the one of the legs with the highest degree of accuracy.

In the gait measurement system according to the present invention, the correlation processor may set a second gate extending from a position between the predicted positions of the right leg and the left leg when there is neither an observed value associated with the subject's right leg nor an observed value associated with the subject's left leg at the immediately preceding processing time. This allows the second gate to be set centered on a position between the predicted positions of both legs when it is assumed that the accuracy of the predicted positions of both legs is not high.

In the gait measurement system according to the present invention, the second gate may be an ellipse with a ratio of its major axis to its minor axis of 8:5. This makes the above-mentioned effect of tracking the subject's leg position with high accuracy particularly effective when another person approaches the subject in the minor axis direction of the second gate.

The gait measurement system according to the present invention is a system used for a walking test of a subject, in which the subject, seated in a chair, walks toward a marker placed away from the chair, turns around the marker, returns to the chair and sits down again, and the second gate may be long in the direction from the chair to the marker. In this case, the second gate may have a shape suitable for a TUG test.

In the gait measurement system according to the present invention, the first gate may include a gate extending based on the predicted position of the subject's right leg, and a gate extending based on the predicted position of the subject's left leg. In this case, the two first gates can effectively prevent erroneously detected observed values from being associated with the predicted positions of both legs.

In the gait measurement system according to the present invention, the range of the first gate may be variable, and the range of the second gate may be fixed. In this case, the variable first gate can flexibly limit the observed values, while the fixed second gate can reliably limit the observed values.

The present invention provides a gait measurement system that can accurately track the position of a subject's legs.

FIG. 1 is a block diagram showing the configuration of a gait measurement system according to an embodiment. FIG. 2 is a schematic diagram showing a TUG test to which the gait measurement system of FIG. 1 is applied. FIG. 3 is a plan view illustrating the laser range sensor of the gait measurement system of FIG. Fig. 4(a) is a diagram for explaining an example of an observation pattern used to detect an observation value, Fig. 4(b) is a diagram for explaining an example of an observation pattern used to detect an observation value, and Fig. 4(c) is a diagram for explaining an example of an observation pattern used to detect an observation value. Fig. 5A is a diagram illustrating an example of an observation pattern used to detect an observation value, and Fig. 5B is a diagram illustrating an example of an observation pattern used to detect an observation value. Fig. 6(a) is a diagram for explaining an example of an observation pattern used to detect an observation value, and Fig. 6(b) is a diagram for explaining an example of an observation pattern used to detect an observation value. 7A and 7B are diagrams illustrating an example of an observation pattern used to detect an observation value. FIG. 8 is a diagram for explaining a walking model. FIG. 9 is a diagram for explaining the predicted positions of the legs. FIG. 10 is a diagram illustrating the first gate and the second gate. FIG. 11 is another diagram illustrating the first gate and the second gate. FIG. 12 is a flowchart showing the process in the gait measurement system of FIG. Fig. 13A is a diagram for explaining matching using a particle filter, Fig. 13B is a diagram for explaining matching using a particle filter, and Fig. 13C is a diagram for explaining matching using a particle filter. Fig. 14(a) is a diagram for explaining matching using a particle filter, Fig. 14(b) is a diagram for explaining matching using a particle filter, and Fig. 14(c) is a diagram for explaining matching using a particle filter. FIG. 15 is a diagram showing tracking results of the TUG test performed by the gait measurement system according to the comparative example and the gait measurement system according to the embodiment.

Below, a preferred embodiment of the present invention will be described in detail with reference to the drawings. In the following description, the same or equivalent elements will be designated by the same reference numerals, and duplicate descriptions will be omitted.

Figure 1 is a block diagram showing the configuration of a gait measurement system according to one embodiment. Figure 2 is a schematic diagram explaining a TUG test to which the gait measurement system of Figure 1 is applied. As shown in Figures 1 and 2, the gait measurement system 100 includes a chair 2, a marker 3, a laser range sensor (distance information acquisition unit) 10, a pressure sensor 15, an electronic control unit 20, a monitor 30, and a pole 40. The gait measurement system 100 is a system that tracks the leg position of a walking subject (subject) 1, and can perform quantitative gait measurement and walking ability evaluation. The gait measurement system 100 is a system used in a walking test of the subject 1. The gait measurement system 100 can be suitably applied to a TUG test, which is one of the walking tests that evaluates functional mobility by integrating walking ability, dynamic balance, agility, etc.

In the TUG test, the subject 1 stands up from a seated position on a chair 2, walks toward a marker 3 such as a cone a predetermined distance (e.g., 3 m) ahead, turns around the marker 3, and then returns to the chair 2 and sits on the chair 2 again. The TUG test is one of the manuals for physical fitness measurement by the Ministry of Health, Labor and Welfare. In the TUG test, walking at the maximum walking speed without running is required. As the marker 3, various markers can be used as long as they can serve as a mark for the turning position. In the TUG test, multiple (here, two) poles 40 are used as reference members to align the marker 3 with the laser range sensor 10. The pole 40 has a cylindrical outer shape of a predetermined height. The predetermined height is a height at which the upper end of the pole 40 is located above the height position of the laser light emitted from the laser range sensor 10 (the height of the optical window of the laser range sensor 10). The shape of the pole 40 is not particularly limited, and various poles can be used. The gait measurement system 100 measures and evaluates the movement history and gait parameters of both legs of subject 1 during such a TUG test.

By detecting the two poles 40 with the laser range sensor 10, the gait measurement system 100 can define a plane coordinate system with a straight line connecting the centers of the two poles 40 as one axis, and the floor surface on which the marker 3 is placed can be this plane coordinate system. In this case, by precisely adjusting the positional relationship between the two poles 40 and the marker 3 (specifically, by placing the marker 3 3 m from the midpoint of the line segment connecting the two poles 40 on a perpendicular line to this line segment), it is possible to precisely grasp the relative positional relationship between the laser range sensor 10 and the marker 3 even if the installation position and/or orientation of the laser range sensor 10 is roughly aligned.

The gait measurement system 100 obtains leg observation positions (hereinafter also referred to as "observation values" or "observation positions") as observation values that are candidate leg positions based on the characteristic patterns observed when the laser range sensor 10 scans both legs of the subject 1, performs tracking using a Kalman filter, and calculates gait parameters based on the obtained movement trajectories of both legs.

Specifically, in the gait measurement system 100, in order to reduce losing sight of and misidentifying the legs in a situation where both legs of the subject 1 are close to each other and in a situation where the legs cannot be observed temporarily by the laser range sensor 10, observation values are acquired based on five types of observation patterns. In addition, the observed values are associated with predicted values using an effective area (gate) that takes into account the state of the legs. In the following, "leg" is also simply referred to as "leg", "standing leg" is also simply referred to as "standing leg", and "swing leg" is also simply referred to as "swing leg". "Left leg" is also simply referred to as "left leg", and "right leg" is also simply referred to as "right leg". "One leg" is also simply referred to as "one leg", and "both legs" is also simply referred to as "both legs".

Actual human walking (the speed of both legs) changes periodically. In particular, in the TUG test, subject 1 is instructed to walk at the maximum walking speed without running, so the speed of both legs changes significantly. Therefore, the gait measurement system 100 applies an acceleration/deceleration model that takes into account the walking phase as a walking model. When walking, people perform periodic movements in which one leg acts as the pivot foot (standing leg) and the other leg (swing leg) swings. "Gait phase" is a classification of the periodic movement of both legs while a person is walking, based on the state of swing of the standing leg and the swing leg (see Figure 8, for example).

The gait measurement system 100 also uses a particle filter to match observed values with predicted values, and a Kalman filter is used to update the state quantities of both legs for each hypothesis (matching). At that time, the weighting of each particle as the likelihood is calculated, taking into account the transitions in the walking phase and the transitions in the observation pattern.

3 is a plan view illustrating the laser range sensor 10 of the gait measurement system 100 of FIG. 1. The laser range sensor 10 is a sensor for acquiring the movement trajectories of both legs. The laser range sensor 10 acquires two-dimensional plane distance information (hereinafter, simply referred to as "distance information"), which is the distance to an object around the sensor in a two-dimensional plane at a certain height. As shown in FIG. 3, the laser range sensor 10 emits a laser light (detection wave) L so as to scan along the horizontal direction, and acquires two-dimensional plane distance information over time regarding the distance to the legs F (right leg F _R and left leg F _L ) of the subject 1 that reflect the laser light L, based on the reflection state of the laser light L.

Specifically, the laser range sensor 10 emits a laser beam L and reflects the laser beam L by a rotating mirror, scanning the laser beam L horizontally in a fan shape in the measurement area. More specifically, the laser range sensor 10 scans the laser beam L horizontally in a fan shape so that the measurement area includes the subject 1 and the poles 40 installed on both sides of the subject 1. Then, the laser range sensor 10 receives the reflected light of the laser beam L reflected by, for example, the leg F of the subject 1, measures the detection angle (scanning angle) of the reflected light and the time (propagation time) from the emission of the laser beam L to its reception, and detects distance information including information related to the angle and distance to the leg F.

The laser range sensor 10 scans the subject 1 with a laser beam L, for example, before or after a TUG test trial, and detects leg information related to the width of the subject 1's leg F (the length corresponding to the diameter of the shin). The laser range sensor 10 outputs the detected distance information and leg information to the electronic control device 20.

The laser range sensor 10 is configured so that the height of the optical window that emits the laser light L can be adjusted. The laser range sensor 10 is installed so that the laser light L is emitted at a height position corresponding to the height of the shin of the subject 1 (i.e., the height from the ankle to below the knee). The height of the optical window of the laser range sensor 10 is set to a position higher than the height of the marker 3 so that the laser light L is not reflected by the marker 3. The height of the optical window of the laser range sensor 10 is set to, for example, 0.15 m to 0.27 m (here, 0.27 m) from the floor surface (ground surface) in consideration of the fact that the leg F of the subject 1 can be observed even when the leg F leaves the floor during the swing phase, and the average height at which the width of the leg F is maximum.

The laser range sensor 10 is positioned in a space formed under the seat of the chair 2 so that the emission direction of the laser light L faces the marker 3 side. The laser range sensor 10 irradiates the laser light L to the subject 1 from the rear side of the subject 1 approaching the marker 3 (moving away from the chair 2), that is, from the front side of the subject 1 moving away from the marker 3 (approaching the chair 2). The laser range sensor 10 is installed so that the emission direction of the laser light L is horizontal to the floor surface.

The laser range sensor 10 used has a measuring range of 0.1 to 30 m, a measuring angle of 270 deg, a measuring accuracy of ±30 mm, an angular resolution of 0.25 deg, and a sampling period of 25 ms/scan. The type and specifications of the laser range sensor 10 are not limited, and various types can be used depending on the measurement environment, for example. The laser range sensor 10 may be enclosed in a housing, for example to suppress the adverse effects of EMI (electromagnetic noise).

The pressure sensor 15 is a sensor that detects the events of the subject 1 standing up from the chair 2 and sitting down on the chair 2. The pressure sensor 15 is placed on the seat of the chair 2. The pressure sensor 15 detects pressure information related to the applied pressure and outputs the pressure information to the electronic control device 20. The pressure information of the pressure sensor 15 here is a voltage value, and the voltage value is A/D converted via the microcomputer board and input to the electronic control device 20. The electronic control device 20 associates the standing up event with its time and stores it, and associates the sitting down event with its time and stores it.

The electronic control unit 20 performs calculations based on the distance information acquired by the laser range sensor 10 to determine the position and speed of the leg F of the subject 1 over time and acquire the walking characteristics of the subject 1. The electronic control unit 20 outputs a start signal (described below) to the subject 1 in the form of at least a sound (audio) and/or a display. The start signal is output from the electronic control unit 20 in response to the operation of the measurer. The electronic control unit 20 stores the time of the start signal. The start signal may be issued by a device or a measurer other than the electronic control unit 20. In this case, the time of the start signal is input to the electronic control unit 20 by a device or a measurer other than the electronic control unit 20.

The electronic control device 20 may be, for example, a personal computer or a dedicated control computer, and is configured to include a CPU (Central Processing Unit), a ROM (Read Only Memory), a RAM (Random Access Memory), etc. The electronic control device 20 has, as functional elements, a sensor data acquisition unit 21, a storage unit 22, a seating determination unit 23, and a data analysis unit 24.

The sensor data acquisition unit 21 synchronously acquires the distance information detected by the laser range sensor 10 and the pressure information detected by the pressure sensor 15. The sensor data acquisition unit 21 may also acquire the distance information synchronously with the start signal time stored in the electronic control unit 20. The sensor data acquisition unit 21 acquires the leg information detected by the laser range sensor 10. The sensor data acquisition unit 21 outputs the acquired distance information, leg information, and pressure information to the storage unit 22. The sensor data acquisition unit 21 outputs the acquired pressure information to the seating determination unit 23.

The memory unit 22 stores the distance information and pressure information acquired by the sensor data acquisition unit 21 in association with the processing time (sampling time). The memory unit 22 stores the leg information of the subject 1 acquired by the sensor data acquisition unit 21. The seating determination unit 23 determines whether or not the subject 1 is seated based on the pressure information of the pressure sensor 15 acquired by the sensor data acquisition unit 21. For example, the seating determination unit 23 determines that the subject 1 is seated when the voltage value from the pressure sensor 15 is greater than a threshold value. The seating determination unit 23 determines that the subject is not seated when the voltage value from the pressure sensor 15 is equal to or less than the threshold value. The seating determination unit 23 outputs the determination result to the memory unit 22, and the memory unit 22 stores the determination result.

The data analysis unit 24 acquires the walking parameters of the subject 1 by analyzing the data stored in the memory unit 22 after the TUG test trial. The data analysis unit 24 acquires the movement trajectory of the leg F of the subject 1 using a particle filter based at least on the distance information acquired by the laser range sensor 10. The data analysis unit 24 has a leg detection unit (observation value detection unit) 26, a leg tracking unit 27, and a walking parameter calculation unit 28. The walking parameters include general walking parameters and walking parameters related to the TUG test.

The general walking parameters are used to evaluate the risk of falling in a general walking test, and include landing position, walking speed [m/sec], number of steps [steps], walking rate [steps/sec], step length [m], stride length [m], step width [m], and stance time (at least one of single stance time [s] and double stance time [s]). The walking parameters related to the TUG test are used to evaluate the risk of falling in a TUG test, and include reaction time (time to stand up) [sec], TUG execution time [sec], turn direction, first step time (time to leave the bed for the first step) [sec], and shortest distance to marker 3 [m].

The walking parameters can be calculated by dividing them into the following five phases. In other words, the walking parameters are calculated for each of the following five phases. At this time, for the step length, stride length, step width, single leg stance time, and double leg stance time, the average, standard deviation, and coefficient of variation are calculated for each phase. Note that the walking parameters are not limited to being calculated by dividing them into these five phases, but may be calculated overall (without dividing them), or may be calculated by dividing them into two parts, the first half and the second half (left and right). The division into each phase may be performed based on, for example, the position of the leg F of the subject 1, etc.
(1) Starting phase
(2) Forward phase
(3) Turning phase
(4) Return phase
(5) Sitting down phase

The starting phase is the phase from the seated state to the start of walking. For example, the starting phase is the phase from the time when the subject 1 leaves the chair 2 within a certain range.
The forward phase is a phase from the start of walking to just before the turning motion. For example, the forward phase is a phase from when the subject 1 starts walking to before turning around the marker 3. The turning phase is a phase during the turning motion. For example, the turning phase is a phase during which the subject 1 turns around the marker 3. The return phase is a phase from after the turning motion to before sitting down. For example, the return phase is a phase from after the subject 1 turns around the marker 3 to before sitting down. The sitting down phase is a phase when the subject 1 returns to the chair, turns around, and sits down. For example, the sitting down phase is a phase when the subject 1 leaves a certain range from the chair 2 once and then re-enters the certain range. The certain range is, for example, a range of 0.25 m from the chair 2 to the marker 3 side.

The landing position is the position where leg F lands on the floor. The walking rate is the number of steps per unit time. The single stance time is the time from landing (on the floor) to lifting off the floor for the same leg. The double stance time is the time from landing on the floor for one leg to lifting off the other leg. The step length is the distance from the landing position of one leg (e.g., heel position) to the landing position of the other leg. The stride length is the distance between adjacent landing positions for the same leg. The step width is the distance between the heels of both legs in the frontal plane (when the subject 1 is viewed from the front).

The reaction time is the time from the start of the walking test (the time of the start signal) to when subject 1 stands up (the buttocks leave the seat of chair 2). The TUG performance time is the time it takes to perform the TUG test, and is the time from the start of the walking test to its end (the time when subject 1, seated on chair 2, starts moving and sits down again). The first step time is the time from the start of the walking test to when subject 1 lifts leg F for the first step (leaves the bed). The shortest distance to marker 3 is the shortest distance between the center of marker 3 and the movement trajectory of leg F.

In this embodiment, the data analysis unit 24 detects one or more observation values based on the distance information acquired by the laser range sensor 10. The data analysis unit 24 randomly associates the predicted positions of both legs of the subject 1 with either the observation value or the observation result of no observation value for each of the multiple particles, and acquires the positions of both legs in association with time. At the same time, the data analysis unit 24 calculates weights from the transition of the walking phase and the transition of the observation pattern to which both legs belong. The data analysis unit 24 selects one of the associated particles based on the weights from the multiple particles, and acquires the movement trajectory of the leg F of the subject 1 based on the predicted positions of both legs for the selected particle. The processing of the data analysis unit 24 will be described in detail below.

[Leg detection]
4 to 7 are diagrams for explaining examples of observation patterns used to detect observed values. FIG. 4(a) and FIG. 6(a) are diagrams for explaining an example of a Single Leg (SL) pattern. FIG. 4(b) and FIG. 6(b) are diagrams for explaining an example of a two Legs Together (LT) pattern. FIG. 4(c) and FIG. 7(a) are diagrams for explaining an example of a Forward Straddle Observable (FS_O) pattern. FIG. 5(a) and FIG. 7(b) are diagrams for explaining an example of a Forward Straddle Unobservable (FS_U) pattern. FIG. 5(b) is a diagram for explaining an example of an Unobservable (UO) pattern.

The leg detection unit 26 detects an observation value (observation value vector ykj:=(xkj, ykj), (j=1...J)) which is a position candidate of the leg F of the _subject 1 based on the two-dimensional plane distance information acquired _by the ^laser range sensor 10 using a preset observation pattern ^. Specifically, in consideration of the fact that the two-dimensional plane distance information of the leg F has a characteristic shape, ^the _leg detection unit 26 performs an edge detection process on the two-dimensional plane distance information to detect the position of the edge E of the leg F. The leg detection unit 26 detects an observation value from the position of the edge E by pattern recognition using a plurality of observation patterns. At this time, the leg detection unit 26 uses leg information related to the width _wl of the leg F of the subject 1 and assumes that the width _wl of the leg F is known. The leg detection by the leg detection unit 26 will be described in detail below. In the following, the "observation value vector" is also simply referred to as the "observation value".

When the subject 1 is within the measurement range of the laser range sensor 10 and there is no obstacle between the laser range sensor 10 and the subject 1, the two-dimensional planar distance information acquired by the laser range sensor 10 will have a characteristic shape as shown in Figures 4(a) to 5(b). The leg detection unit 26 calculates the observation value using a leg detection method that takes into account five types of observation patterns.

First, the leg detector 26 detects an edge e h m (h=B, F) when the i-th distance information from the right acquired by the laser range sensor 10 is l _i and the two-dimensional plane distance information of the adjacent laser range sensor 10 satisfies the following formula 1 with respect to the width w _l of the leg F of the subject 1. In this case, if l _i >l _i+1 , the edge ^{e B m} _{= i} and the edge e ^F _m+1 =i+1. If l _i <l _i+1 , the edge e ^F _m =i and the edge e ^B _m+1 =i+1. The superscript B indicates that the edge E is on the far side of the laser range sensor 10, and the superscript F indicates that the edge E is on the near side. The subscript m (m=1, ..., M _k ) indicates that the edge E is the m-th edge detected. M _k is the total number of edges E detected at time (hereinafter also referred to as "processing time") k.

Then, based on the positional relationship of the four consecutive edges e ^h _n , e ^h _n+1 , e ^h _n+2 , and e ^h _n+3 and the width w e between the edges e ^h _{n+1 and} e ^h _n+2 , the observation patterns are classified into five types, and the observation value y _k is calculated, where n = 1, ..., M _k -3, and the subscript k indicates the observation position at time k.

As shown in Fig. 4(a) and Fig. 6(a), the observation pattern O1 as the SL pattern is an observation pattern when the leg F can be completely observed by itself from the laser range sensor 10. In the observation pattern O1, the combination of edges E satisfies ( ^eBn _, ^eFn ₊₁ , ^eFn ₊₂ , ^eBn ₊₃ ), and the width between the edges E satisfies _0.5wl < _we ≦ _1.5wl . In this case, the observation value _yk is calculated as shown in Fig. 4(a) taking into account the width _wl of the leg F of the subject 1. For example, the observation value _ykj ^is calculated from a line passing through ( ^eFn ₊₁ + ^eFn ₊₂ )/2 and 1/2 of the width _wl of the leg F, as shown in the figure.

As shown in Fig. 4(b) and Fig. 6(b), the observation pattern O2 as the LT pattern is an observation pattern when the legs F are aligned and completely observable by the laser range sensor 10. In the observation pattern O2, the combination of edges E satisfies ( ^eBn , ^eFn ₊₁ , ^eFn ₊₂ ^{, eBn} ₊₃ ), ( _{eFn ,} eBn _{+1, eFn+2} ^, _eBn ⁺ _{3 )} or ( ^eBn , ^eFn ₊₁ , ^eBn ₊₂ ^{, eFn +} ₃ ), and the width between edges E satisfies _1.5wl < w _e < _3.0wl . In this case, the observation value _yk is calculated as shown in Fig. 4(b), taking into consideration that both legs of the subject 1 are aligned. For example, the observed value _ykj ⁺¹ is obtained from a line passing through ( ^eFn ₊₁ + ^3eFn ₊₂ )/4 and 1/2 the width _wl of the leg F, as shown in the figure. For example, the observed value _ykj is obtained from a line passing through ( ^3eFn ₊₁ + ^eFn ₊₂ )/4 and 1/2 the width _wl of the leg F ^, as shown in the figure.

4(c) and 7(a), the observation pattern O3 as the FS_O pattern is an observation pattern in which the pattern is stepped due to the influence of one leg F or a walking stick, the width between the positions of the edges E is 1/2 or more of the width _wl of the legs F of the subject 1, and the center position of the legs F can be observed by the laser range sensor 10. In the observation pattern O3, the combination of the edges E satisfies ( ^eFn _, ^eBn ₊₁ , ^eFn ₊₂ , ^eBn ₊₃ ) or ( ^eBn _, ^eFn ₊₁ , ^eBn ₊₂ , ^eFn ₊₃ ), and the width between the edges E satisfies _0.5wl ≦ _we < _1.5wl . In this case, the observation value _yk is calculated as shown in FIG. 4(c), taking into consideration that the center position of the legs F of the subject 1 can be observed by the laser range sensor 10. For example, the observed value y _k ^j is determined from a line passing through (e ^B _n+1 +e ^F _n+2 )/2 and 1/2 of the width w _l of the leg F, as shown in the figure.

5(a) and 7(b), observation pattern O4 as an FS_U pattern is an observation pattern in which the width between edges E is smaller than 1/2 the width wl of leg F of subject 1 and the center position of leg F cannot _be directly observed by laser range sensor 10, as compared with observation pattern O3 which is an FS_O pattern. Observation pattern O3 has a combination of edges E that satisfies ( ^eFn _, ^eBn ₊₁ , ^eFn ₊₂ , ^eBn ₊₃ ) or ( _eBn ^{, eFn} + ₁ , ^eBn ₊₂ , ^eFn ₊₃ ), and the width between edges E satisfies _0.3wl < _we < _0.5wl . At this time, unlike observation patterns O1 to O3, the center position of leg F cannot be directly observed by laser range sensor 10, so the observation value _yk is virtually calculated as shown in FIG. 5(a). For example, the observed value y _k ^j can be determined from a line passing through e ^B _n+1 and a line passing through e ^F _n+2 , and 1/2 the width w _l of the leg F. The hollow squares in Fig. 7(b) indicate imaginary edges.

As shown in FIG. 5(b), observation pattern O5 as a UO pattern is an observation pattern when one leg F cannot be observed, such as when it is hidden by the other leg F or when the leg F is raised higher than the laser range sensor 10.

Even in a situation where the leg F cannot be directly observed by the laser range sensor 10, it is expected that the measurement accuracy of the position of the leg F can be improved by defining the observation pattern _O4 , which is an FS_U pattern, and virtually obtaining the observation value yk. In detecting the observation value _yk , it is possible to determine that the position of the edge E that does not satisfy the observation patterns O1 to O4 is not the leg F. Furthermore, in order to avoid unnecessary correlation processing, an area that is wider than the walkway area by a certain length is set as the measurement area, and the observation value _yk outside the measurement area and the corresponding position of the edge E are excluded.

In detecting the observation value _yk , the position detected is not the center between the positions of the edges E, but, for example, a position that is 1/2 the width _wl of the leg F inward from the position of the edge E. By detecting the observation _{value yk} of the leg F according to the multiple observation patterns O1 to O5 as described above, it becomes possible to appropriately detect the leg F even when one of the legs F is overlapped and hidden when viewed from the laser range sensor 10.

[Leg Tracking]
The leg tracking unit 27 performs a prediction process using a walking model that takes into account the walking phase, and estimates the positions and velocities of both legs. The prediction process uses a Kalman filter.

FIG. 8 is a diagram for explaining a walking model. When a person walks, one leg is used as a pivot leg (standing leg) while swinging the other leg (free leg) to proceed with walking. Both legs move periodically with acceleration and deceleration, switching between free and standing legs depending on the landing. As shown in FIG. 8, the leg tracking unit 27 applies a walking model including the following six walking phases, including when stationary, focusing on changes in the moving speed v and acceleration a of the leg F.
・Phase 0: At rest, both legs are standing ・Phase 1: Left leg is swinging (acceleration, swinging up), right leg is standing ・Phase 2: Left leg is swinging (deceleration, movement towards landing), right leg is standing ・Phase 3: Left leg is standing, right leg is swinging (acceleration, swinging up)
Phase 4: Left leg stance, right leg swing (deceleration, movement toward landing)
Phase 5: Both legs swing (not seen during walking)

In the TUG test, subject 1 is instructed to walk at the maximum speed without running, so there is a large change in the speed of both legs. For this reason, a model that accelerates and decelerates is applied as the walking model. Furthermore, when tracking a person's center of gravity, a nonlinear model that includes angles and angular velocity as state variables is often used, but in the TUG test, since the direction of movement of both legs changes suddenly when turning, it is desirable to apply a linear movement model that has no directionality with respect to the direction of movement. Therefore, as the walking model, a linear movement model that has no directionality with respect to the direction of movement is applied.

The leg tracking unit 27 sets a state equation for each of the left and right legs F with the position and velocity as state quantity vectors, and performs tracking processing for each leg F. The leg tracking unit 27 defines three motion models: stance leg, swing leg (acceleration), and swing leg (deceleration). A discrete-time system equation for each leg F considering such a walking model involving acceleration and deceleration is given by the following Equation 2. The superscript f indicates the left leg and the right leg, where f=L and R, respectively.

The state quantity vector is shown in the following formula 3. The following formula 4 respectively represent the position and velocity of each leg F of the subject 1 at time k.

Vectors u ^f,m _k are unknown external inputs (acceleration inputs) given according to the walking phase, and are shown in Equation 5 below. The superscript m indicates the m-th motion model. Model 1 is the motion model of the standing leg (stationary), model 2 is the motion model of the swing leg (acceleration), and model 3 is the motion model of the swing leg (deceleration). In motion model 1 at the time of stance, 0 is given, in motion model 2 in which the swing leg accelerates, a positive value is given in the forward direction, and in motion model 3 in which the swing leg decelerates, a negative value is given in the forward direction. The magnitude of the acceleration input is calculated as the average value of the norm of the acceleration vector (see Equation 6 below) during the swing leg for the past N _ac time steps.

In addition, since the movement of the leg F fluctuates including the movement direction, the fluctuation is included in the system noise (see Equation 7 below) as an acceleration disturbance from the motion model, and is given as normal white noise with mean 0 and covariance matrix ^Qf . The state transition matrix ^Af , input matrix ^Bfu _, and drive matrix ^Bf of the state equation are respectively expressed by Equation 8 below.

Furthermore, when the observed value y ^f _k of each leg F acquired using the laser range sensor 10 is expressed by the following formula 9, the observation equation is given by the following formula 10. Here, Δy ^f _k is assumed to be normal white noise with mean 0 and covariance R ^f . In this case, the matrix C ^f is expressed by the following formula 11.

As shown in FIG. 1, the leg tracking unit 27 includes a stance leg swing determination unit 27a, a walking phase determination unit 27b, a predicted position calculation unit 27c, a correlation processing unit 27d, an update processing unit 27e, a weight calculation unit 27f, and a resampling unit 27g.

The stance leg swing leg determiner 27a performs stance leg swing leg determination to determine whether the leg F is a stance leg (supporting leg) or a swing leg based on the speed of the leg F. The stance leg swing leg determiner 27a determines the right leg as a stance leg when the following formula 12 is satisfied, and determines the right leg as a swing leg when the following formula 13 is satisfied. v _{st_th} indicates the speed threshold of the stance leg, and v _{sw_th} indicates the speed threshold of the swing leg. The stance leg swing leg determiner 27a performs stance leg swing leg determination for the left leg in the same manner, that is, using the following formulas 12 and 13 with the right leg and the left leg swapped.

The walking phase determination unit 27b determines the walking phase based on the relative positional relationship and speed of both legs, and specifies whether the walking phase is one of phases 0 to 5. Specifically, first, when both legs are standing, the phase is specified as 0. Next, when the left leg is a swing leg and the right leg is a standing leg, if the inner product of the position vector of the left leg relative to the right leg and the velocity vector of the left leg (see Equation 14 below) is a positive value, the phase is specified as 1, and if the inner product is a negative value, the phase is specified as 2. Similarly, when the left leg is a standing leg and the right leg is a swing leg, similarly, if the inner product of the position vector of the right leg relative to the left leg and the velocity vector of the right leg is a positive value, the phase is specified as 3, and if the inner product is a negative value, the phase is specified as 4. When both legs are swing legs, the phase is specified as 5. When the walking phase determination unit 27b determines that the walking phase is 5, the data analysis unit 24 determines that the subject 1 is running.

Here, the leg tracking unit 27 of this embodiment employs a method of tracking both legs based on a hypothesis of correspondence over multiple time periods, taking into account the transition of the walking phase and the observation pattern of the leg F, using a Rao-Blackwellized particle filter (hereinafter, simply referred to as a "particle filter"). In tracking multiple objects, it is necessary to estimate the correspondence c _k (c _k indicates the index of correspondence) between the tracked object and the observation value y _k acquired by the sensor at time k, and the state quantity x _k of the tracked object. The correspondence and the state of the tracked object can be obtained by the particle filter alone, but it is necessary to generate a large number of particles for the state quantity for each combination of correspondence, which results in a large amount of calculation. Therefore, in the particle filter, the state quantity and correspondence of the tracked object are obtained by the following formula 15. Note that, for example, the number of particles may be 100.

In the particle filter, when an observation is obtained in p(c1 _:k |vector y1 _:k ), only the correspondence is calculated by the particle filter. Next, in p(vector _xk |c1 _:k , vector y1 _:k ), the state quantity is updated by the Kalman filter according to the associated observation value _yk . Since the state quantity update process is performed by the Kalman filter, it is not necessary to generate a large number of particles for each combination of correspondence. In the leg tracking unit 27, the weight is calculated as the likelihood of each particle based on the state transition probability of the walking phase and the observation pattern, thereby improving the performance of tracking both legs of the subject 1, for example, an elderly person, during walking including a turning motion.

The predicted position calculation unit 27c calculates the predicted positions of both legs by predicting the positions of both legs using a Kalman filter. The predicted position calculation unit 27c calculates a leg predicted position (hereinafter also simply referred to as a "predicted position", see "x" in FIG. 9) which is the predicted position of each leg F at time k based on the positions of both legs at past times for each of a plurality of particles. The superscripts (n), (n=1, ..., N) indicate the nth particle. The predicted position vector y^ ^f,(n) _k/k-1 of each leg F is given by the following formula 16 according to the state equation shown in the above formula (2). The vector x^ ^f,(n) _k/k-1 indicates a prior state estimate value at time k of the nth particle. The vector x^ ^f,(n) _k-1/k-1 is a posterior state estimate value at time k-1 of the nth particle. 9, "◇" indicates the observed value _yk , the line indicates the movement trajectory of the leg F, and "□" and "■" on the line indicate the positions of the left leg and the right leg, respectively (the same applies to the subsequent figures). In the following, the "predicted position vector" will also be simply referred to as the "predicted position".

The correlation processing unit 27d associates the predicted positions of both legs with the observed value _yk . The correlation processing unit 27d randomly associates the predicted positions of both legs with the observed value _yk for each particle. However, in order to limit the observed value _yk to be associated, the correlation processing unit 27d sets a gate GA (see FIG. 10) at the predicted position, and randomly associates each of the predicted positions of both legs with either "(i) one or more observed values _yk included in the gate GA" or "(ii) an observation result in _which the observed value ^yk is not _present (when the observed value _yk is not present)" in consideration of the case where the observed value yk includes a false positive.

In this embodiment, the gate GA includes a first gate GA1 and a second gate GA2. That is, the correlation processor 27d sets the first gate GA1 that spreads based on the predicted positions of both legs, and sets the second gate GA2 that spreads based on the predicted positions of both legs according to the observation value at the past time. In the case of (i) above, the correlation processor 27d associates the predicted positions of both legs with the observation value _yk that exists in the area where the first gate GA1 and the second gate GA2 overlap.

The first gate GA1 includes a gate that expands based on the predicted position of the right leg F _R of the subject 1 and a gate that expands based on the predicted position of the left leg F _L of the subject 1. The first gate GA1 here is two circular areas that expand with the predicted positions of both legs (see the solid line "x" in FIG. 10) as centers. The range of the first gate GA1 is variable. For example, the range of the first gate GA1 when the leg F is determined to be in a swing state may be changed so as to be wider than the range of the first gate GA1 when the leg F is determined to be in a stance state. For example, the range of the first gate GA1 may be changed so as to be wider as the velocity v of the leg F is higher.

The second gate GA2 has an elongated shape extending from the chair 2 toward the marker 3. The second gate GA2 is an ellipse with the major axis extending from the chair 2 toward the marker 3. The ellipse of the second gate GA2 has a ratio of the major axis to the minor axis of 8:5, taking into consideration that an attendant is often present to the side of the subject 1. The range of the second gate GA2 is fixed. The range of the second gate GA2 is a predetermined range determined in advance based on the stride length of an average person.

For example, if the horizontal direction from chair 2 to marker 3 is the x coordinate, and the horizontal direction perpendicular to the horizontal direction from chair 2 to marker 3 is the y coordinate, then the ellipse of second gate GA2 can be expressed as ((x- _xc )/a) ² + ((y- _yc )/b) ² < 1. In this case, the center coordinates are (xc, yc). a is a constant, for example, 0.8. b is a constant, for example, 0.5.

The setting of the second gate GA2 will be specifically described. The correlation processing unit 27d sets the second gate GA2 that spreads based on the predicted positions of both legs depending on the presence or absence of the observed values y _k of both legs at past times. As shown in FIG. 10, when both the observed value y _k associated with the right leg F _R of the subject 1 and the observed value y _k associated with the left leg F _L of the subject 1 exist at the immediately preceding processing time, the correlation processing unit 27d sets the second gate GA2 that spreads from a position between the predicted position of the right leg F _R and the predicted position of the left leg F _L at the current processing time. As shown in the figure, the center of the second gate GA2 here is the midpoint between the predicted position of the right leg F _R and the predicted position of the left leg F _L (see the dotted line "x" in the figure). The immediately preceding processing time is the processing time immediately before the current processing time. For example, the current processing time is time k, and the immediately preceding processing time is time k-1.

As shown in FIG. 11, when there exists only an observation value y _k corresponding to one of the two legs at the immediately previous processing time, the correlation processing unit 27 d sets a second gate GA2 extending from the predicted position of the one of the two legs at the current processing time.

When there is neither an observed value _yk associated with the right leg F _R of the subject 1 nor an observed value _yk associated with the left leg F _L of the subject 1 at the immediately preceding processing time, the correlation processing unit 27d sets a second gate GA2 that extends at the current processing time centered on a position between the predicted positions of the right leg F _R and the left leg F _L. The center of the second gate GA2 here is the midpoint between the predicted positions of the right leg F _R and the left leg F _L.

Whether the observed value y _k is included in the first gate GA1 and whether the observed value y _k is included in the second gate GA2 are determined using the following formula 17. ^{d (n) 2} _f,j is the Mahalanobis distance and is given by the following formula 18. ^{S f,(n)} _k is the covariance matrix of the observed prediction error (y ^j _k -y^ ^f,(n) _k/k-1 ). G corresponds to each of the first gate GA1 and the second gate GA2. In the present embodiment in which the observed value y _k is a two-dimensional vector and a Kalman filter is used, G is determined based on a chi-squared (χ ² ) distribution with two degrees of freedom.

The update processing unit 27e updates the state quantity of the leg F for each of the plurality of particles after the correlation processing unit 27d performs the correspondence process for each of the plurality of particles, using the observation value y _k for which the correlation processing unit 27d performs the update process for the Kalman filter (for example, updating the state quantity such as the covariance matrix). However, when there is no corresponding observation value y _k (for example, in the case of the observation pattern O5 as the UO pattern), the update processing unit 27e sets the prior state estimate value and the prior covariance matrix as the posterior state estimate value and the posterior covariance matrix, respectively. As a result, the position and speed of both legs are estimated for each of the plurality of particles. In addition, the update processing unit 27e determines the walking phase by the walking phase determination unit 27b based on the information on the estimated position and speed of both legs for each of the plurality of particles.

After updating by the update processing unit 27e, the weight calculation unit 27f calculates a weight w ⁽ⁿ⁾ _k for each of the plurality of particles from the transition of the walking state of the subject 1. Specifically, the weight calculation unit 27f calculates (updates) the weight w ⁽ⁿ⁾ _k for each particle from the transition probability of the walking phase (phases 0 to 5) of each leg from one time point before and the observation pattern (observation patterns O1 to O5) of the leg F, as shown in the following formula 19.

The second term in the above formula 19 is calculated as the product of the likelihood of the observation value y _k of the Kalman filter of each leg F and the state transition probability of the observation patterns O1 to O5. In this embodiment, the transition probability matrix from the observation patterns O1 to O5 one time before is defined as in the following formula 20. The horizontal arrangement in the following formula 20 corresponds, from the left, to the observation patterns O1 to O5 (SL pattern, LT pattern, FS_O pattern, FS_U pattern, UO pattern) before the transition (one time before). The values in the vertical columns in the following formula 20 indicate, from the top, the probability of transition from the observation pattern before the transition to the observation patterns O1 to O5. As expressed in the following formula 20, the transition of the observation patterns O1 to O5 for calculating the weight w ⁽ⁿ⁾ _k has the following characteristics. In the case of the observation pattern O1 in which the leg is observed alone one time before, it is set so that the probability of the observation pattern being O1 is high even at the next time, and conversely, it is set so that there is no transition to the observation pattern O4 in which the leg is almost hidden by the other leg. In the case of observation pattern O2, in which both legs are aligned one time before, the probability of the observation pattern being O2 at the next time and the probability of transitioning to observation pattern O1, in which the legs are observed alone, are designed to be high. In the case of observation pattern O3, in which both legs are shifted and observed in a stepped manner one time before, the probability of the observation pattern being O3 at the next time and the probability of transitioning to observation pattern O1, in which the legs are observed alone, or observation pattern O4, in which the legs are almost hidden by the other leg, are designed to be high. In the case of observation pattern O4, in which the legs are almost hidden by the other leg one time before, the probability of the same observation pattern O4 at the next time and the probability of the observation pattern being O3, in which the legs are out of sight behind the other leg and can be observed, or observation pattern O5, in which the legs are completely hidden by one leg, are set to be high. In the case of observation pattern O5, in which the legs were not observed one time before, the probability of the observation pattern being O5 at the next time is high, and the transition to other observation patterns is set to be approximately equal in probability.

The third term in the above formula 19 relates to the state transition probability of the walking phase. In this embodiment, the transition probability matrix from phases 0 to 5 one time before is defined as in the following formula 21. The horizontal arrangement in the following formula 21 corresponds to phases 0 to 5 before the transition (one time before), from the left. The values in the vertical columns in the following formula 21 indicate the probability of transition from the phase before the transition to phases 0 to 5 after the transition, from the top. As expressed in the following formula 21, the transition of phases 0 to 5 for calculating the weight w ⁽ⁿ⁾ _k has the following characteristics. In the case of phase 0, which is a stationary state one time before, the probability of being phase 0 at the next time and the probability of transitioning to phases 1 and 3 when starting to walk with either the left or right leg are set to be high. In the case of phase 1, which is accelerating with the left leg as a swing leg one time before, the probability of being phase 1 at the next time and the probability of transitioning to phase 2, which is decelerating, are set to be high. In the case of phase 2, in which the left leg is swinging and decelerating one time before, the probability that it will remain in phase 2 at the next time and the probability of it transitioning to phase 3, in which the right leg starts to accelerate as a swing leg, or to phase 0, in which it remains stationary, are set to be high. For phases 3 and 4, in which the right leg is swinging, the design takes into consideration transitions similar to those of phases 1 and 2. In addition, in the case of phases 0 to 4, the design is such that the probability of transitioning to phase 5, which does not occur during walking, at the next time is very small. In addition, in the case of phase 5 one time before, the design is such that there is an equal probability of transitioning to phases 0 to 4 at the next time.

In this embodiment, when the walking test of the subject 1 is in the starting phase, after the update by the update processing unit 27e, the weight calculation unit 27f selects a part of the particles in which the estimated position of the right leg F _R is to the left of the estimated position of the left leg F _L (in other words, the estimated position of the left leg F _L is to the right of the estimated position of the right leg F _R ) from among the particles. That is, the weight calculation unit 27f selects a part of the particles that would have been erroneously estimated due to the left and right legs F being swapped in the starting phase. Then, the weight calculation unit 27f reduces the weight w ⁽ⁿ⁾ _k for the selected part of the particles (for example, to 1/10). Specifically, the weight calculation unit 27f calculates (updates) the weight w ⁽ⁿ⁾ _k for the selected part of the particles by multiplying the right side of the above formula 19 by a coefficient (for example, 1/10) that is greater than 0 and less than 1.

Whether the estimated position of the right leg F _-R is to the left of the estimated position of the left leg F- _L can be determined based on which estimated position has a larger y coordinate, where the right end is 0 in the y coordinate system and the left end is positive.

The resampling unit 27g resamples the particles so that they are selected and replicated based on the weight w ⁽ⁿ⁾ _k . For example, the resampling unit 27g resamples the particles so that the smaller the weight w ⁽ⁿ⁾ _k , the more the particles are selected and the larger the weight w ⁽ⁿ⁾ _k , the more the particles are replicated. Specifically, the resampling unit 27g normalizes the weight w ⁽ⁿ⁾ _k of each particle and calculates the effective sample size by the following formula 22. The resampling unit 27g performs resampling when N ^eff _k is smaller than a threshold value. For example, the resampling unit 27g performs equal-interval resampling. Note that the specific method of resampling performed by the resampling unit 27g is not particularly limited, and various known resampling methods may be used.

After estimating the positions and velocities of both legs at all times for the plurality of particles and updating and resampling the final weight w ⁽ⁿ⁾ _k , the leg tracking unit 27 selects a corresponding particle from the plurality of particles based on the weight w ⁽ⁿ⁾ _k . Here, the leg tracking unit 27 selects one of the particles with the largest weight w ⁽ⁿ⁾ _k . The leg tracking unit 27 obtains the movement trajectory of the leg F of the subject 1 based on the positions and velocities of both legs for the selected particle.

[Calculation of walking parameters]
The walking parameter calculation unit 28 calculates walking parameters from the acquired movement trajectories of both legs. The walking parameter calculation unit 28 includes a stance time calculation unit 28a, a landing position calculation unit 28b, a step length, step width and stride length calculation unit 28c, a first step time calculation unit 28d, a reaction time calculation unit 28e, and a TUG execution time calculation unit 28f.

The stance time calculation unit 28a calculates the single-leg stance time and the double-leg stance time. In this embodiment, to calculate the single-leg stance time and the double-leg stance time, the floor contact time and floor departure time are precisely calculated after the movement trajectory is acquired, based on the analysis results of a subject experiment in which pressure sensors were attached to the heels and metatarsals of both legs.

The landing position calculation unit 28b first determines the landing position in order to calculate the step length, stride length, and step width. Because the laser range sensor 10 acquires the movement trajectory of the leg F at shin height, it is difficult for the landing position calculation unit 28b to directly measure the position of the leg F. Therefore, the landing position calculation unit 28b takes into consideration that the leg F becomes perpendicular to the floor surface at the time when the speed of the leg F is at a minimum during stance, and calculates the position of the leg F at that time as the landing position.

The step length, step width and stride length calculation unit 28c calculates the step length, step width and stride length based on the calculated landing position. The first step time calculation unit 28d calculates the time difference from the time of the start signal to the time when the leg F leaves the floor for the first step as the first step time. The reaction time calculation unit 28e calculates the time when the voltage value from the pressure sensor 15 installed on the chair 2 changes from the sitting state (the time when it drops below the threshold value) as the reaction time. The reaction time calculation unit 28e calculates the time difference from the time of the start signal to the reaction time as the reaction time.

The TUG execution time calculation unit 28f uses the fact that the voltage value of the pressure sensor 15 changes when the subject 1 sits back down on the chair 2, and sets the rising time of this voltage value as the end time. In other words, it detects when the subject 1 sits down on the chair 2 based on the pressure information detected by the pressure sensor 15. The TUG execution time calculation unit 28f calculates the time difference from the start signal to the end time as the test execution time of the TUG test.

The walking parameter calculation unit 28 also calculates the shortest distance to the marker 3 based on the movement trajectories of both legs and the center position of the marker 3. The walking parameter calculation unit 28 calculates other general walking parameters from the acquired movement trajectories of both legs.

The monitor 30 is an output unit that outputs the analysis results analyzed by the data analysis unit 24. The monitor 30 displays the analysis results analyzed by the data analysis unit 24 and provides feedback to the subject 1. The monitor 30 can also output distance information, leg information, and pressure information acquired by the sensor data acquisition unit 21 or stored in the memory unit 22. The monitor 30 may be configured to output voice or the like instead of or in addition to a display. Also, instead of or in addition to the monitor 30, a printer that prints out the calculation results, etc. on a paper medium may be provided. The monitor 30 may be provided integrally with the electronic control unit 20, and may be, for example, a display unit of a personal computer, a dedicated control computer, etc.

[Gait measurement using gait measurement system 100]
Next, a case where gait characteristics are measured in a TUG test using the gait measurement system 100 of this embodiment will be described with reference to the flowchart shown in FIG.

As shown in FIG. 12, the measurement process in the TUG test is divided into three phases: a calibration phase (calibration before a trial), a measurement phase (measurement during a TUG test trial), and an analysis phase (data analysis after a trial). In the calibration phase, the laser range sensor 10 is aligned using two poles 40, and the width _w1 of the leg F of the subject 1 is measured. After the calibration is completed, the electronic control device 20 notifies the subject 1 of a start signal (instruction to start the TUG test) by sound, for example, and the measurement phase begins. In the measurement phase, the distance information of the laser range sensor 10 and the pressure information of the pressure sensor 15 are acquired and stored in a synchronized manner. Then, after the subject 1 is detected as being seated from the pressure information, the analysis phase begins. In the analysis phase, the data analysis unit 24 tracks the positions of both legs based on the stored data to acquire a movement trajectory, and the walking parameters are calculated based on the acquired movement trajectory. This will be described in detail below.

First, before the subject 1 starts moving, calibration is performed. In the calibration, a scan is performed by the laser range sensor 10. That is, the laser range sensor 10 emits laser light L so as to scan along the horizontal direction, and distance information is obtained based on the reflection state of the laser light L (step S1). This allows the positions of the legs F of the subject 1 and the two poles 40 to be observed.

The width (leg information) _wl of the leg F of the subject 1 is detected from the width between the observed edge positions of the leg F (step S2). The detected leg information is stored in the memory unit 22. By determining the positional relationship between the two poles 40 using the laser range sensor 10, a measurer or the like can check whether the laser range sensor 10 is installed in an appropriate position and direction relative to the marker 3, and align the laser range sensor 10 as necessary (step S3). As described above, the installation precision of the laser range sensor 10 can be coarse as long as the relative positional relationship between the two poles 40 and the marker 3 is strictly maintained.

The series of processes in steps S1 to S3 above are repeated until the calibration is completed (step S4). A few seconds after the calibration is completed, a start signal is automatically given to the subject 1 (or manually by the examiner), and the TUG test begins (step S5).

During the TUG test, a scan is performed by the laser range sensor 10, and the sensor data acquisition unit 21 acquires distance information regarding the distance of the leg F of the subject 1 over time (step S6). The sensor data acquisition unit 21 acquires pressure information from the pressure sensor 15 (step S7). The seating determination unit 23 determines whether the subject 1 has stood up or not based on the pressure information (step S8). If it is determined that the subject 1 has not stood up, the process returns to step S6. If it is determined that the subject 1 has stood up, the seating determination unit 23 determines whether the subject 1 has sat down or not based on the pressure information (step S9). If it is determined that the subject 1 has not sat down, the TUG test is still in progress, and the process returns to step S6. If it is determined that the subject 1 has sat down, the TUG test is deemed to have ended, and the following data analysis is performed.

In the data analysis, a leg position tracking process is performed, and a plurality of particles in the particle filter are initialized (step S10). At time k, the leg detection unit 26 detects an observation value y _k based on the stored distance information (step S11). At time k, the predicted position calculation unit 27c calculates a predicted position for each particle (step S12). At time k, the correlation processing unit 27d randomly associates each particle (step S13). At time k, the update processing unit 27e updates the state quantity for each particle (step S14). This allows the positions and velocities of both legs at time k to be estimated. At time k, the weight calculation unit 27f calculates a weight w ⁽ⁿ⁾ _k for each particle (step S15). At time k, the resampling unit 27g performs resampling (step S16).

Next, a data end determination is performed to determine whether each process related to the data analysis has been completed for all of the saved data (distance information saved in association with time) stored in the memory unit 22 (step S17). If the answer is No in step S17, the process proceeds to step S11 at the next time. On the other hand, if the answer is Yes in step S17, it is determined that the leg position tracking process has been completed.

Then, the leg tracking unit 27 selects one of the particles with the largest weight w ⁽ⁿ⁾ _k . The leg tracking unit 27 acquires the movement trajectories of both legs of the subject 1 in the TUG test based on the positions and velocities of both legs in the selected particle. Then, the walking parameter calculation unit 28 calculates walking parameters from the acquired movement trajectories of both legs (step S18). The analysis results of the stored data analysis and/or the walking parameters are output to the monitor 30 (step S19).

As described above, in the gait measurement system 100, even if the leg position of another person, such as an attendant, is erroneously detected as the observed value _yk , the observed value _yk can be prevented from being associated with the predicted positions of both legs by the first gate GA1 and can also be prevented by the second gate GA2. Therefore, it is possible to prevent the leg positions of another person from being erroneously tracked as the leg positions of the subject 1, and it is possible to track the leg positions of the subject 1 with high accuracy.

In the gait measurement system 100, when there are both an observed value _yk associated with the right leg F _R of the subject 1 and an observed value _yk associated with the left leg F _L of the subject 1 at the immediately preceding processing time, the correlation processing unit 27d sets a second gate GA2 centered on a position between the predicted positions of the right leg F _R and the left leg F _L. In this case, the second gate GA2 can be set centered on a position between the predicted positions of both legs with high accuracy.

In the gait measurement system 100, when there is only an observed value y _k associated with one of the legs at the immediately preceding processing time, the correlation processor 27 d sets a second gate GA2 that expands based on the predicted position of the one of the legs. In this case, the second gate GA2 can be set based on the predicted position of the one of the legs with the highest accuracy.

In the gait measurement system 100, when there is neither an observed value y _k associated with the right leg F _R nor an observed value y _k associated with the left leg F _L at the immediately preceding processing time, the correlation processing unit 27d sets a second gate GA2 centered on a position between the predicted positions of the right leg F _R and the left leg F _L. In this way, when it can be assumed that the accuracy of the predicted positions of both legs is not high, it is possible to set the second gate GA2 centered on a position between the predicted positions of both legs.

In the gait measurement system 100, the second gate GA2 is an ellipse with a ratio of its major axis to its minor axis of 8:5. This makes the above-mentioned effect of tracking the position of the legs of the subject 1 with high accuracy particularly effective when there is likely to be another person close to the subject 1 in the minor axis direction of the second gate GA2.

The gait measurement system 100 is a system used for the TUG test of the subject 1, and the second gate GA2 has a long shape extending from the chair 2 toward the marker 3. In this case, the second gate GA2 can be shaped to be suitable for the TUG test.

In the gait measurement system 100, the first gate GA1 includes a gate extending based on the predicted position of the right leg F _R of the subject 1 and a gate extending based on the predicted position of the left leg F _L of the subject 1. In this case, the two first gates GA1 can effectively prevent the erroneously detected observed value y _k from being associated with the predicted positions of both legs.

In the gait measurement system 100, the range of the first gate GA1 is variable, and the range of the second gate GA2 is fixed. In this case, the observation value _yk can be flexibly limited by the variable first gate GA1, and the observation value _yk can be reliably limited by the fixed second gate GA2.

In the gait measurement system 100, the weight w ⁽ⁿ⁾ _k is reduced for some particles that would have been erroneously estimated due to the left and right legs F being swapped in the starting phase. This makes it possible to suppress erroneous tracking in which the left and right legs F are tracked in a swapped position from the actual position in the starting phase. The behavior of the legs F of the subject 1 in the starting phase is characteristic, and erroneous tracking in which the left and right legs F are swapped is particularly likely to occur in the starting phase, so the effect of being able to suppress such erroneous tracking in the starting phase is effective.

In the gait measurement system 100, the predicted positions of both legs of the subject 1 are associated with the observed values for each of the plurality of particles. This allows for a plurality of possible associations. Then, a particle with any of the associations is selected from the plurality of particles based on the weight w ⁽ⁿ⁾ _k to determine the predicted positions of both legs and obtain a movement trajectory. This weight w ⁽ⁿ⁾ _k is calculated from the unique features found in the legs F of the walking subject 1, that is, the transition of the walking phase and the observation pattern. Therefore, it is possible to obtain a movement trajectory based on any of the plurality of possible associations that is most likely for the subject 1. Therefore, it is possible to suppress the acquisition of a movement trajectory based on an incorrect association, and it is possible to accurately track the leg positions of the subject 1. It is possible to reduce erroneous tracking.

In the gait measurement system 100, the weight w ⁽ⁿ⁾ _k is calculated based on the transition of the gait phase. In this case, the weight w ⁽ⁿ⁾ _k of each particle can be calculated taking into consideration that the gait phase changes periodically.

In the gait measurement system 100, the weight w ⁽ⁿ⁾ _k is calculated based on the transition of the observation patterns corresponding to both legs. In this case, the weight w ⁽ⁿ⁾ _k of each particle can be calculated taking into consideration that the observation patterns of both legs change periodically.

In the gait measurement system 100, the data analysis unit 24 includes a leg detection unit 26, a predicted position calculation unit 27c, and a correlation processing unit 27d. The leg detection unit 26 detects the observation value y _k from the distance information. The predicted position calculation unit 27c calculates the predicted positions of both legs for each of the plurality of particles based on the positions of both legs at past times. The correlation processing unit 27d sets a gate GA based on the predicted positions for each of the plurality of particles, and randomly associates the predicted positions of both legs of the subject 1 with either the observation value y _k or the observation result with no observation value y _k present within the gate GA. In this case, the association for each particle can be performed using the gate GA set based on the predicted positions.

In the gait measurement system 100, the data analysis unit 24 has an update processing unit 27e and a weight calculation unit 27f. The update processing unit 27e updates the state quantity of the leg F for each of the plurality of particles after the correlation processing unit 27d performs correspondence. The weight calculation unit 27f calculates the weight w ⁽ⁿ⁾ _k from the transition of the walking phase and the observation pattern for each of the plurality of particles after the update processing unit 27e performs updating. In this case, the weight w ⁽ⁿ⁾ _k of each particle can be specifically calculated in tracking the leg position using the particle filter.

In the gait measurement system 100, the predicted position calculation unit 27c calculates the predicted positions of both legs by predicting the positions of both legs using a Kalman filter. The update processing unit 27e performs an update process for the Kalman filter. In this case, the leg positions of the subject 1 can be tracked with high accuracy using the Kalman filter.

In the gait measurement system 100, the data analysis unit 24 includes a resampling unit 27g. The resampling unit 27g resamples a plurality of particles so that they are selected and replicated based on the weight w ⁽ⁿ⁾ _k . In this case, the number of particles required for tracking the leg position using a particle filter can be reduced. It becomes possible to hold the association hypothesis and estimate the state quantity with a small number of particles.

Figures 13 and 14 are diagrams explaining matching using a particle filter. Figure 13(a) shows the result of matching for the first particle PT1 at time k, Figure 13(b) shows the result of matching for the second particle PT2 at time k, and Figure 13(c) shows the result of matching for the third particle PT3 at time k. Figure 14(a) shows the result of matching for the first particle PT1 at time k+1, Figure 14(b) shows the result of matching for the second particle PT2 at time k+1, and Figure 14(c) shows the result of matching for the third particle PT3 at time k+1.

In the examples of FIG. 13(a) to FIG. 13(c), the erroneous detection of observed values y _k ¹ and y _k ² is detected. In the first particle P1, the predicted position of the left leg is associated with the observed value y _k ¹ , and the predicted position of the right leg is associated with the observed value y _k ^2. In the second particle P2, the predicted position of the left leg is associated with the observed value y _k ² , and the predicted position of the right leg is associated with the observed value y _k ^1. In the third particle P3, the predicted position of the left leg is associated with an observation result of no observed value, and the predicted position of the right leg is associated with the observed value y _k ^2. In the third particle P3, since there is no observed value associated with the predicted position of the left leg, the predicted position of the left leg is acquired as the position of the left leg. In such an example, the weight w ⁽¹⁾ _k > weight w ⁽²⁾ _k > weight w ⁽³⁾ _k .

In the examples of FIG. 14(a) to FIG. 14(c), the observed value y _k+1 ¹ and the observed value y _k+1 ² are detected. In the first particle P1, the predicted position of the left leg is associated with the observed value y _k+1 ² , and the predicted position of the right leg is associated with the observed value y _k+1 ^1. In the second particle P2, the predicted position of the left leg is associated with the observed value y _k+1 ² , and the predicted position of the right leg is associated with the observed value y _k+1 ^1. In the third particle P3, the predicted position of the left leg is associated with the observed value y _k+1 ¹ , and the predicted position of the right leg is associated with the observed value y _k+1 ^2. In such an example, the weight w ⁽³⁾ _k+1 >> the weight w ⁽²⁾ _k+1 > the weight w ⁽¹⁾ _k+1 . As a result, for example, the first particle PT1 is eliminated and the third particle PT3 is duplicated by resampling.

15 is a diagram showing the tracking results of the TUG test by the gait measurement system according to the comparative example and the gait measurement system 100 according to the embodiment. In the test results, the embodiment corresponds to the gait measurement system 100. Comparative example 1 corresponds to a system similar to the gait measurement system 100, except that the matching using the particle filter is not performed and the second gate GA2 is not set. Comparative example 2 corresponds to a system similar to the gait measurement system 100, except that the above process of reducing the weight w ⁽ⁿ⁾ _k for some particles that are erroneously estimated due to the left and right legs F being switched in the starting phase is not performed and the second gate GA2 is not set.

The number of subjects 1 to be tested was 138. "Switching of left and right legs" in the erroneous tracking content in the figure refers to a situation in which the left leg F _L of the subject 1 is tracked as the right leg F _R and/or the right leg F _R is tracked as the left leg F _L. "Mistracking of attendant's legs" in the erroneous tracking content in the figure refers to a situation in which the attendant's legs are mistakenly tracked as the legs F of the subject 1.

As shown in FIG. 15, the overall tracking success rate is improved in the embodiment compared to Comparative Examples 1 and 2. This confirms the effect of the gait measurement system 100, which is capable of tracking the leg position of the subject 1 with high accuracy.

Furthermore, in the Example, the number of times the legs of the attendant were erroneously tracked was reduced compared to Comparative Examples 1 and 2. This confirms the effect of the gait measurement system 100, which is capable of suppressing erroneous tracking of the leg positions of the attendant (other person) as the leg positions of the subject 1. In particular, in the Example, the number of times the legs of the attendant were erroneously tracked was reduced regardless of the phase, which indicates that the effect is due to the setting of the second gate GA2 rather than the above process of reducing the weight w ⁽ⁿ⁾ _k in the starting phase.

Furthermore, in the example, the number of interchanges of the left and right legs F during the starting phase is significantly reduced compared to Comparative Example 2. This confirms the effect of the gait measurement system 100, which is capable of suppressing erroneous tracking in which the left and right legs F are tracked in an interchanged position from the actual position during the starting phase. Since this effect is brought about during the starting phase, it can be seen that the above process of reducing the weight w ⁽ⁿ⁾ _k during the starting phase is particularly contributory to this effect.

The gait measurement system 100 uses a small laser range sensor 10 and a pressure sensor 15 capable of acquiring distance information on a two-dimensional plane, thereby enabling tracking of the positions of both legs with fewer misidentifications during TUG test measurements and improving measurement accuracy. The gait measurement system 100 does not perform real-time analysis during the walking test, but instead performs analysis (post-analysis) by the data analysis unit 24 after the walking test is completed. This makes it possible to calculate precise movement trajectories and walking parameters. Of course, the gait measurement system 100 also makes it possible to analyze walking characteristics by the data analysis unit 24 while the walking test is being performed.

[Modification]
One aspect of the present invention has been described above, but this aspect of the present invention is not limited to the above embodiment, and may be modified or applied to other things without changing the gist of the claims.

In the above embodiment, the weight w ⁽ⁿ⁾ _k is calculated from the transition of the walking phase and the transition of the observation pattern of the subject 1, but the weight w ⁽ⁿ⁾ _k may be calculated from only one of them. In the above embodiment, the above formula 19 for calculating the weight w ⁽ⁿ⁾ _k is expressed as the product of the second term related to the transition of the observation pattern and the third term related to the transition of the walking phase, but it may be expressed as the sum or difference of the second and third terms. In the above embodiment, the transition of the walking state is not limited to the transition of the walking phase and the transition of the observation pattern, and may be a transition of the walking state based on the positions of both legs.

In the above embodiment, the resampling unit 27g resamples a plurality of particles, but in some cases, the resampling unit 27g may not be necessary. In the above embodiment, the movement trajectory of the leg F is obtained from the positions and velocities of both legs of the particle with the largest weight w ⁽ⁿ⁾ _k , but this is not limiting. For example, the movement trajectory may be obtained from any of the particles with a weight w ⁽ⁿ⁾ _k equal to or greater than a predetermined value. In short, any of the associated particles may be selected based on the weight w ⁽ⁿ⁾ _k , and the movement trajectory may be obtained from the selected particle.

In the above embodiment, the gait measurement system 100 is applied to measure walking characteristics in a TUG test, but this is not limited thereto, and the gait measurement system 100 can be applied to measure walking characteristics in, for example, a multi-target stepping test (MTST) and various other walking trainings.

In the above embodiment, the shape, size, central position, etc. of the first gate GA1 and the second gate GA2 are not particularly limited. Various shapes, sizes, and central positions may be adopted as the shape, size, central position, etc. of the first gate GA1 and the second gate GA2 depending on various factors. In the above embodiment, the size of the first gate GA1 is variable, but the size of the first gate GA1 may be fixed. In the above embodiment, the size of the second gate GA2 is fixed, but the size of the second gate GA2 may be variable.

For example, when the gait measurement system 100 is applied to measure walking characteristics in the MTST, the shape of the second gate GA2 may be a perfect circle in consideration of the characteristics of the MTST. For example, the shape of the second gate GA2 may be elongated in the direction of travel of the subject 1, and in this case, when the subject 1 turns the marker 3, the elongated direction of the second gate GA2 changes so as to rotate 180° in response to the turn. For example, at least one of the shape, size, and center position of the second gate GA2 may be different for each phase of the walking test.

In the above embodiment, the second gate GA2 is set to expand based on the predicted positions of both legs according to the observation value _yk at the immediately preceding processing time, but the second gate GA2 may be set to expand according to the past observation value _yk without being limited to the observation value _yk at the immediately preceding processing time. In the above embodiment, the data analysis unit 24 acquires the movement trajectory of the leg F of the subject 1 using a particle filter, but the movement trajectory of the leg F of the subject 1 may be acquired without using a particle filter (i.e., without performing matching using a particle filter).

In the above embodiment, a laser range sensor 10 is used as the distance information acquisition unit, but various sensors and devices can be used as long as they can acquire distance information at a position at a specified height from the floor surface. For example, an infrared sensor capable of acquiring the above-mentioned two-dimensional planar distance information may be used as the distance information acquisition unit. The above embodiment includes one laser range sensor 10, but multiple sensors may be included. In the above embodiment, a pole 40 is used as the reference member, but this is not limited to this. Various members may be used as the reference member as long as they are capable of aligning the laser range sensor 10 with the marker 3.

In the above embodiment, the laser range sensor 10 is positioned so that the laser light L is irradiated from the rear side of the subject 1 approaching the marker 3 (moving away from the chair 2), but the laser range sensor 10 may also be positioned so that the laser light L is irradiated from the front side of the subject 1 approaching the marker 3. In some cases, the laser range sensor 10 may be positioned so that the laser light L is irradiated from the side of the subject 1 approaching the marker 3. In the above embodiment, the laser range sensor 10 is positioned under the chair 2, but the position of the laser range sensor 10 is not limited, and the laser range sensor 10 may be positioned according to the irradiation mode of the laser light L to the subject 1.

In the above embodiment, subject 1's standing up (moving out) from chair 2 and sitting down on chair 2 were determined based on pressure information from pressure sensor 15, but in addition to determination by pressure sensor 15 or omitting pressure sensor 15, the standing up and sitting down may be determined based on the detection results of laser range sensor 10. In the above embodiment, subject 1's standing up and sitting down were detected based on pressure information, but only standing up may be detected based on pressure information, or only sitting down may be detected based on pressure information.

The present invention can be understood as a gait measurement method implemented by the gait measurement system 100. The present invention can be understood as a gait measurement program that causes a computer to execute the gait measurement method (each process of the gait measurement system 100).

The components in the above embodiments and modifications are not limited to the materials and shapes described above, and various materials and shapes can be applied. The components in the above embodiments or modifications can be applied to the components in other embodiments or modifications as desired. Parts of the components in the above embodiments or modifications can be omitted as appropriate without departing from the gist of one aspect of the present invention.

1...Subject (subject), 10...Laser range sensor (distance information acquisition unit), 24...Data analysis unit, 26...Leg detection unit (observation value detection unit), 27c...Predicted position calculation unit, 27d...Correlation processing unit, 27e...Update processing unit, 27f...Weight calculation unit, 27g...Resampling unit, 100...Gait measurement system, F...Leg, GA1...First gate, GA2...Second gate, L...Laser light (detection wave).

Claims (8)

A gait measurement system for tracking a leg position of a walking subject, comprising:
a distance information acquiring unit that emits a detection wave so as to scan along a horizontal direction and acquires distance information over time relating to a distance to an object that has reflected the detection wave based on a reflection state of the detection wave;
a data analysis unit that acquires a movement trajectory of the subject's legs based at least on the distance information acquired by the distance information acquisition unit,
The data analysis unit
a leg detection unit that detects one or more observation values that are position candidates of the leg based on the distance information;
A predicted position calculation unit that calculates predicted positions of both legs of the subject based on at least the observed values at past times;
a correlation processor that correlates the observed values with the predicted positions of the legs,
The correlation processing unit
setting a first gate that spreads based on the predicted positions of both legs, and setting a second gate that spreads based on the predicted positions of both legs according to the observed values at past times;
The gait measurement system associates the predicted positions of both legs with the observed values that exist within an area where the first gate and the second gate overlap. The gait measurement system according to claim 1, wherein the correlation processor sets the second gate extending from a position between the predicted position of the right leg and the predicted position of the left leg when both the observed value associated with the right leg of the subject and the observed value associated with the left leg of the subject exist at the immediately preceding processing time. The gait measurement system according to claim 1 or 2, wherein the correlation processor sets the second gate extending from a predicted position of one of the legs when only the observed value associated with one of the legs exists at the immediately preceding processing time. The gait measurement system according to any one of claims 1 to 3, wherein the correlation processor sets the second gate extending from a position between the predicted position of the right leg and the predicted position of the left leg when both the observed value associated with the right leg of the subject and the observed value associated with the left leg of the subject at the immediately preceding processing time are not present. The gait measurement system according to any one of claims 1 to 4, wherein the second gate is an ellipse with a ratio of its major axis to its minor axis being 8:5. 1. A system for use in gait testing of the subject, comprising:
The walking test is a test in which the subject walks from a seated state in a chair toward a marker disposed away from the chair, turns around the marker, and then returns to the chair and sits down again;
6. The gait measurement system according to claim 1, wherein the second gate is elongated in a direction from the chair to the marker. The gait measurement system according to any one of claims 1 to 6, wherein the first gate includes a gate that extends based on a predicted position of the subject's right leg and a gate that extends based on a predicted position of the subject's left leg. the range of the first gate is variable;
8. The gait measurement system according to claim 1, wherein a range of the second gate is fixed.

Images