JP6476156B2 - Attitude estimation apparatus, method and program - Google Patents

Description

  The present invention relates to a posture estimation apparatus, method, and program for estimating a human posture, for example.

  Accompanying the widespread use of wearable terminals, posture estimation apparatuses that estimate the posture of a person using detection data of inertial sensors built into the terminals have been developed. In general, the posture of a person in a stationary state is estimated using a trigonometric function from the acceleration value detected by an acceleration sensor. This method is widely used as an application for estimating a posture because it is possible to estimate the posture of a person by using only one acceleration sensor and the calculation load is small. However, for example, when trying to estimate the attitude of a driver who is driving a car using an acceleration sensor, the value of the acceleration sensor changes due to acceleration / deceleration of the car, so it is difficult to accurately estimate the driver's own attitude. .

  Therefore, as a technique for estimating the driver's posture, for example, a method of estimating the posture from the image of the camera as described in Non-Patent Document 1, or the acceleration of the driver as described in Non-Patent Document 2 is used. There has been proposed a method for estimating the posture using two acceleration sensors, that is, a sensor for measuring and a sensor for measuring acceleration of an automobile.

Rajeev Gupta, et al. “Posture Recognition For Safe Driving”, Third International Conference on Image Information Processing (ICIIP), pp. 141-146 (2015) Masahiro Tada et al., “Measurement and Analysis Method of Driver Behavior Using Wireless Accelerometer”, IEICE Transactions D, Vol. 91, No.4, pp.1115-1129 (2008)

  However, in the method described in Non-Patent Document 1, it is necessary to install a camera for photographing the driver on the vehicle, and in the method described in Non-Patent Document 2, acceleration sensors are installed on both the driver and the vehicle. There is a need.

  The present invention has been made paying attention to the above circumstances, and intends to provide an attitude estimation apparatus, method, and program capable of accurately estimating the attitude of a person on the board using only a single acceleration sensor. Is.

  In order to solve the above-mentioned problem, a first aspect of the present invention is to receive acceleration data output from an acceleration sensor attached to a person riding in a vehicle, and based on the received acceleration data, Estimate the posture. Then, the angular velocity is calculated based on the time series change of the information representing the estimation result, and the filter is used to filter the information representing the estimation result using a filter whose parameters dynamically change in accordance with the angular velocity. By processing, a component derived from the movement of the vehicle included in the information representing the estimation result is suppressed.

  In a second aspect of the present invention, a filter that performs exponential moving average processing is used as the filter, and the smoothing coefficient of the filter is dynamically changed in accordance with the calculated angular velocity.

  According to a third aspect of the present invention, when the smoothing coefficient of the filter is dynamically changed, the calculated angular velocity is compared with at least one threshold value set in advance, and the threshold value is set based on the comparison result. Correspondingly, a preset smoothing coefficient is selected.

  According to the first aspect of the present invention, attention is paid to the fact that the frequency band of human motion is different from the frequency band of acceleration / deceleration of the vehicle, and the posture is changed using a filter whose parameters dynamically change according to the angular velocity. By performing the filtering process on the estimation result, it is possible to remove a noise component derived from the acceleration / deceleration of the vehicle. For this reason, it is possible to accurately estimate a change in the posture of a person on board by simply using a single acceleration sensor attached to the person.

  According to the second aspect of the present invention, a filter that performs exponential moving average processing is used, and the smoothing coefficient of the filter is dynamically changed according to the angular velocity, so that filtering processing is performed by calculation processing. It can be carried out.

  According to the third aspect of the present invention, the angular velocity is determined by comparing with a threshold value, and an appropriate smoothing coefficient is selected according to the determination result. Can be changed.

  That is, according to the embodiment of the present invention, it is possible to provide an attitude estimation apparatus, method, and program capable of accurately estimating the attitude of a person on board only by using a single acceleration sensor.

The block diagram which shows the function structure of the attitude | position estimation apparatus which concerns on one Embodiment of this invention. The flowchart which shows the process sequence and process content by the attitude | position estimation apparatus shown in FIG. The figure which shows the 1st example of the effect by the attitude | position estimation apparatus shown in FIG. 1 compared with the conventional method. The figure which shows the 2nd example of the effect by the attitude | position estimation apparatus shown in FIG. 1 compared with the conventional method.

Embodiments according to the present invention will be described below with reference to the drawings.
[One Embodiment]
(Constitution)
FIG. 1 is a block diagram showing a functional configuration of a system including an attitude estimation apparatus according to an embodiment of the present invention. This system includes a posture estimation device 1 and an acceleration sensor 2.

  For example, the acceleration sensor 2 is attached to the center of the chest of the driver to be estimated. Although the acceleration sensor may be attached to the driver alone, it may be attached to the driver in a state where it is housed in a sensor unit together with other sensors such as a biosensor. The driver may wear the garment by pasting or sewing it at a corresponding position.

  The acceleration sensor 2 detects triaxial acceleration representing a change in the driver's posture at a constant sampling period, and transmits the detected data to the posture estimation device 1 via the wireless interface. As the wireless interface, for example, a wireless interface using a low-power wireless data communication standard such as Bluetooth (registered trademark) is used.

  The posture estimation device 1 is, for example, a wearable terminal worn on a driver's wrist, a smartphone or a tablet terminal housed in a driver's chest pocket or a glove box of a vehicle, and automatic driving control installed in a vehicle. Of computers. The posture estimation device 1 includes a control unit 11, a storage unit 12, and an input / output interface unit 13.

  The input / output interface unit 13 includes a wireless interface adopting the above-described low-power wireless data communication standard such as Bluetooth (registered trademark), and receives acceleration detection data transmitted from the acceleration sensor 2.

  The storage unit 12 uses a nonvolatile memory such as SSD (Solid State Drive) that can be written and read at any time as a storage medium. In the storage area, an acceleration storage unit 121, a pre-correction angle storage unit 122, A post-correction angle storage unit 123 is provided.

  The acceleration storage unit 121 is used for storing acceleration detection data transmitted from the acceleration sensor 2. The pre-correction angle storage unit 122 is used to store information that is calculated by the control unit 11 and that represents the angle of the front / rear inclination of the driver and the angle of the left / right inclination. The corrected angle storage unit 123 is used to store information that is calculated by the control unit 11 and that represents the corrected angle of the front / rear inclination of the driver and the corrected angle of the left / right inclination.

  The control unit 11 has a central processing unit (CPU) and a working memory. As a control function necessary for carrying out this embodiment, an acceleration acquisition unit 111, a moving average processing unit 112, A pre-correction angle calculation unit 113, an angular velocity calculation unit 114, and a post-correction angle calculation unit 115 are provided. Any of these functions 111 to 115 is realized by causing the CPU to execute an application program stored in a program area in the storage unit 12.

  The acceleration acquisition unit 111 takes in the triaxial acceleration detection data transmitted from the acceleration sensor 2 through the input / output interface unit 13 every predetermined time (for example, 1 second), and together with information indicating the detection time. The process which stores in the acceleration memory | storage part 121 is performed.

  The moving average processing unit 112 reads the triaxial acceleration detection data from the acceleration storage unit 121 every predetermined time (one second), and performs a process of calculating an average value of acceleration for each axis in the one second.

  The pre-correction angle calculation unit 113 calculates the driver's front / rear tilt angle and left / right tilt angle based on the three-axis acceleration average value calculated by the moving average processing unit 112. And the process which matches the calculated value with the information showing the said detection time, and stores it in the angle memory | storage part 122 before correction | amendment is performed. A specific calculation example will be described later.

  The angular velocity calculation unit 114 includes the latest pre-correction angle for one second stored in the pre-correction angle storage unit 122, the pre-correction angle for the previous one second, and the pre-correction angle calculation cycle (1 second). Is used to calculate the angular velocity. A specific calculation example will be described later.

  The post-correction angle calculation unit 115 is stored in the latest pre-correction angle stored in the pre-correction angle storage unit 122, the angular velocity calculated by the angular velocity calculation unit 114, and the post-correction angle storage unit 123. The corrected angle is calculated by applying a moving average process in which the smoothing coefficient changes dynamically using the corrected angle in the previous one second. Then, the calculated corrected angle is stored in the corrected angle storage unit 123 in association with the information indicating the detection time. A specific calculation example for calculating the corrected angle will be described later.

(Operation)
Next, the operation of the posture estimation apparatus configured as described above, that is, the posture estimation method will be described according to the processing procedure of the control unit 11. FIG. 2 is a flowchart showing the processing procedure and processing contents.

(1) Acquisition of Acceleration Data During driving, the acceleration sensor 2 detects triaxial acceleration representing a change in the posture of the driver, and the triaxial acceleration detection data is transmitted to the posture estimation device 1. Under the control of the acceleration acquisition unit 111, the posture estimation device 1 takes in the triaxial acceleration detection data transmitted from the acceleration sensor 2 from the input / output interface unit 13 for one second in steps S11 and S12. The acquired triaxial acceleration detection data is stored in the acceleration storage unit 121 of the storage unit 12 in association with information representing the detection time.

  Note that the triaxial acceleration detection data may be captured at a time interval shorter or longer than 1 second. In addition, the triaxial acceleration detection data may be transmitted and received as it is, but may be transmitted and received after being compressed by encoding for a certain time. In this way, it is possible to reduce the information amount of the triaxial acceleration detection data and save the storage capacity of the acceleration storage unit 121.

(2) Calculation of Pre-Correction Angle Under the control of the moving average processing unit 112, the posture estimation apparatus 1 reads the three axes of acceleration detection data from the acceleration storage unit 121 for one second in step S13, for each axis. The moving average value of the acceleration detection values is calculated. Subsequently, under the control of the pre-correction angle calculation unit 113, in step S14, based on the calculated moving average value of the three-axis acceleration detection values, the front and rear inclination angles and the left and right inclination angles of the driver are determined. Calculate each.

For example, X-axis in the lateral direction of the acceleration sensor 2 mounted on the chest central portion of the driver, the front-rear direction Y-axis, the respective acceleration when the vertical direction is Z axis a _x, a _y, When a _z The Euler angle is calculated by the following equation.
The angle of the driver's front / rear inclination and the angle of the left / right inclination calculated by the expressions (1) and (2) are corrected by the pre-correction angle calculation unit 113 in association with the information indicating the detection time. It is stored in the front angle storage unit 122.

(3) Post-correction angle calculation Next, the posture estimation apparatus 1 controls the angular velocity calculation unit 114 under the control of the angular velocity calculation unit 114, and in step S15, from the pre-correction angle storage unit 122, the angle of front-rear inclination and the left-right inclination angle in the latest one second And the angle of front / rear tilt and the angle of left / right tilt in the previous one second are read. Then, the difference between the angle of front / rear inclination and the angle of left / right inclination in the latest one second and the angle of front / rear inclination and the angle of left / right inclination in the previous one second are calculated, respectively. And calculate the angular velocity.

For example, if the latest pre-correction angle is θ _n , the previous one-second pre-correction angle is θ _n-1 , and the pre _- correction angle calculation period is T seconds, the angular velocity ω _n is ω _n = (θ _n − θ _n-1 ) / T (3)
Is calculated by

The posture estimation apparatus 1 is then stored in the post-correction angle storage unit 123 under the control of the post-correction angle calculation unit 115 and the calculated previous pre-correction angle θ _n and angular velocity ω _n for one second. The corrected angle θ ^ _n is calculated by applying a moving average process in which the smoothing coefficient dynamically changes using the value of the corrected angle θ ^ _n-1 of the previous one second.

For example, the n-th corrected angle estimation result theta ^ _n, n-th pre-correction angle estimation result theta _n, when the smoothing coefficient as a parameter alpha, as below formula exponential moving average processing table Is done.
θ ^ _n = (1-α) θ ^ _n-1 + αθ _n (4)
The value of α changes dynamically according to the value of the angular velocity calculated by the above-described equation (3) in the angular velocity calculation unit 114. The change method may be a heuristic rule base (decision tree) or may be determined using machine learning. The smaller the value of α, the stronger the effect of the low-pass filter because the effect of the angle after the previous correction becomes stronger.

For example, assuming that the angular velocity at the time of the driver's posture change is 10 degrees / second or more, and assuming that the angular velocity less than that is noise due to acceleration / deceleration of the vehicle, the value of α is based on the following formula: Determine.
if ω _n ≧ 10, α = 0.9
else, α = 0.0009
According to the above formula, if the angular velocity is greater than or equal to the threshold value, it is regarded as a change in the posture of the driver, and the effectiveness of the low-pass filter is weakened. .

Based on the above concept, the corrected angle calculation unit 115 first determines in step S16 whether or not the angular velocity ω _n calculated by the angular velocity calculation unit 114 is greater than or _equal to a preset threshold value. If the angular velocity ω _n is equal to or greater than the threshold value, the smoothing coefficient α is set to a large value in step S17, and the latest pre-correction angle θ _n and the previous one second are used by using the smoothing coefficient α. The value of the corrected angle θ ^ _n-1 is subjected to exponential moving average processing according to the above equation (4), and the latest corrected angle θ ^ _n for one second is calculated.

On the other hand, if the value of the angular velocity ω _n is less than the threshold value, the smoothing coefficient α is set to a smaller value in the above case in step S18, and the latest correction for one second is performed using the set smoothing coefficient α. Exponential moving average processing is performed using the above equation (4) for the previous angle θ _n and the previous corrected angle θ ^ _n-1 for 1 second, and the latest corrected angle θ ^ _n for 1 second is obtained. calculate.

  The corrected angle calculation unit 115 calculates the corrected angle of the front / rear inclination of the driver and the corrected angle of the left / right inclination calculated by the exponential moving average process in step S17 or S18 in the latest one second in step S19. The post-correction angle storage unit 123 stores the post-correction angle of the driver's posture.

  The post-correction angle of the driver's posture stored in the post-correction angle storage unit 123 is read into, for example, an automatic driving control device, and is used to determine whether the automatic driving mode can be switched to the manual driving mode. Is done.

(effect)
As described above in detail, the posture estimation apparatus 1 according to the embodiment receives the triaxial acceleration detection data output from the acceleration sensor 2, calculates the moving average value of each axis every second, Based on the moving average value, the driver's front / rear inclination angle and left / right inclination angle are calculated. Then, the angular velocity ω _n is calculated from the difference between the angle of the front and rear inclination and the angle of the previous one second for each of the left and right inclination angles, and if the calculated angular velocity ω _n is equal to or greater than a preset threshold value. Using the smoothing coefficient α set to a large first value, the exponential moving average for the most recent 1 second pre-correction angle θ _n and the previous 1 second post-correction angle θ ^ _n-1 Processing is performed to calculate the latest corrected angle θ ^ _n for 1 second. On the other hand, if the value of the angular velocity ω _n is less than the threshold value, the latest pre-correction angle θ _n for one second and one value are used using the smoothing coefficient α set to the second value smaller than the first value. It performed an exponential moving average processing on the values of the corrected angle theta ^ _n-1 of the previous one second, and to calculate the corrected angle theta ^ _n of the latest 1 second.

  Therefore, the noise component derived from the acceleration / deceleration of the vehicle can be suppressed by the filtering process, thereby accurately estimating the change of the driver's posture only by using the single acceleration sensor 2 attached to the driver. It becomes possible to do.

  For example, suppose that a 25 Hz acceleration sensor is attached to the center of the driver's chest and the circuit is driven at a speed of about 40 km / h. Then, under this condition, the angle of the driver's front / rear inclination and the angle of the left / right inclination are applied using the method according to the embodiment of the present invention, that is, the filter using the exponential moving average process in which the smoothing coefficient α dynamically changes. When the estimation was performed using the method and the conventional method using a filter using a moving average process for 1 second of triaxial acceleration, the results shown in FIG. 3 were obtained.

  In FIG. 3, the horizontal axis represents time [seconds], the vertical axis represents the driver's front-rear inclination angle and left-right inclination angle [°], and A1 represents the left-right inclination angle according to one embodiment. A2 indicates the angle of front and rear inclination according to one embodiment. In addition, B1 shows the angle of the left-right inclination by the conventional method, and A2 shows the angle of the front-back inclination by the conventional method.

  The driver has always been driving in the same posture, but in the conventional method, as shown in B2 and B1 of FIG. This is because a change in acceleration due to acceleration / deceleration of the vehicle is added as noise. On the other hand, according to one embodiment, as shown in A1 and A2 of FIG. 3, the influence (noise component) of the acceleration change due to acceleration / deceleration of the vehicle is almost eliminated, and the driver's posture is constant. An estimation result showing was obtained.

  FIG. 4 shows a posture estimation result when the driver takes a forward leaning posture at a rate of about once every 30 seconds while traveling under the same conditions as in FIG. In FIG. 4, A5 indicates the estimation result for the forward tilt posture according to the embodiment, and B5 indicates the estimation result for the forward tilt posture according to the conventional method. As shown in FIG. 4, in the conventional method, it is estimated that the driver's left / right inclination angle B3 is changing during curve driving, but according to one embodiment, the left / right inclination angle A3 is substantially constant. It can be seen that the angle change A5 of the forward / backward inclination can be accurately captured.

[Other Embodiments]
In the embodiment, the value of the angular velocity ω _n is determined using one threshold value, and two types of smoothing coefficients α are selected according to the determination result, and the angle is subjected to exponential moving average processing. However, the present invention is not limited to this, and the angular velocity ω _n is further finely identified using two or more thresholds, and a smoothing coefficient α set in advance for each region of the identified angular velocity ω _n is used. Thus, the angle may be subjected to exponential moving average processing.

For example, the threshold for determining the angular velocity is
if ω _n ≧ 10, α = 0.9
else if 10> ω _n ≧ 5, α = 0.09
else, α = 0.0009
It is determined.
In this way, the estimated value of the posture angle can be corrected by selectively using the three types of smoothing coefficient α according to the magnitude of the angular velocity ω _n .

  In the embodiment, the case where the acceleration sensor 2 is provided separately from the posture estimation device 1 has been described as an example. However, the acceleration sensor may be provided in the posture control device. The person whose posture is to be estimated is not limited to the driver but may be a navigator sitting in the passenger seat or a passenger sitting in the rear seat. In addition, the configuration of the posture estimation apparatus, its processing procedure and processing contents, the type of posture to be estimated, and the like can be variously modified and implemented without departing from the gist of the present invention.

  In short, the present invention is not limited to the above-described embodiment as it is, and can be embodied by modifying the constituent elements without departing from the scope of the invention in the implementation stage. Further, various inventions can be formed by appropriately combining a plurality of constituent elements disclosed in the embodiment. For example, some components may be deleted from all the components shown in the embodiment. Furthermore, you may combine suitably the component covering different embodiment.

  DESCRIPTION OF SYMBOLS 1 ... Posture estimation apparatus, 2 ... Acceleration sensor, 11 ... Control unit, 12 ... Memory | storage unit, 13 ... Input / output interface unit, 111 ... Acceleration acquisition part, 112 ... Moving average process part, 113 ... Pre-correction angle calculation part, 114 ... angular velocity calculation section, 115 ... post-correction angle calculation section, 121 ... acceleration storage section, 122 ... pre-correction angle storage section, 123 ... post-correction angle storage section.

Claims (7)

Means for receiving acceleration data output from an acceleration sensor attached to a person riding in the vehicle;
Means for estimating the posture of the person based on the received acceleration data and outputting information representing the estimation result;
Means for calculating an angular velocity based on a time series change of information representing the output estimation result;
Derived from the movement of the vehicle included in the information representing the estimation result by providing a filter whose parameters dynamically change according to the calculated angular velocity, and filtering the information representing the estimation result by the filter A posture estimation apparatus comprising correction means for suppressing a component to be corrected.   The posture estimation apparatus according to claim 1, wherein the correction unit includes a filter that performs exponential moving average processing as the filter, and dynamically changes a smoothing coefficient of the filter in accordance with the calculated angular velocity.   3. The correction means compares the calculated angular velocity with at least one preset threshold value, and selects a preset smoothing coefficient corresponding to the threshold value based on the comparison result. Posture estimation device. A posture estimation method executed by a posture estimation device that estimates the posture of a person riding in a vehicle,
Receiving acceleration data output from an acceleration sensor worn by the person;
A step of estimating the posture of the person based on the received acceleration data and outputting information representing the estimation result;
A process of calculating an angular velocity based on a time series change of information representing the output estimation result;
By using a filter whose parameter dynamically changes in accordance with the calculated angular velocity, and filtering the information representing the estimation result by the filter, the movement of the vehicle included in the information representing the estimation result is detected. A posture estimation method comprising: a correction process for suppressing derived components.   5. The posture estimation method according to claim 4, wherein the correction process uses a filter that performs exponential moving average processing as the filter, and dynamically changes a smoothing coefficient of the filter in accordance with the angular velocity.   6. The correction process compares the calculated angular velocity with at least one preset threshold value, and selects a preset smoothing coefficient corresponding to the threshold value based on the comparison result. The attitude estimation method described.   The program which makes a computer function as each means with which the attitude | position estimation apparatus of any one of Claims 1 thru | or 3 is provided.