Driving analysis system, driving analysis method, and driving analysis program

Abstract

This technology predicts the likelihood of slowing down during a marathon and provides users with useful information. [Solution] In the running analysis system 100, a prediction model is stored in the running analysis server 60 that predicts the pace changes in the latter half of the race using machine learning with motion history data that is measured over time from the beginning to the end of the race, using features corresponding to multiple types of motion analysis indicators, including at least running pace, as features that indicate the runner's motion state in a completed marathon. The measurement value acquisition unit 40 acquires motion history data that is measured over time with features corresponding to multiple types of motion analysis indicators during runs of less than the marathon race distance by the user 10, who is a marathon runner. Based on the motion history data measured over time during the user 10's run and the prediction model, the running analysis server 60 predicts the pace changes in the latter half of the race for the user 10 and notifies the user 10 of the prediction result.

Description

Technical Field

[0001] This invention relates to a technique for analyzing running motion.

Background Art

[0002] In recent years, due to the increasing health consciousness of people, the number of runners has been increasing. In particular, with the improvement of various measurement technologies using information terminals and wearable devices, runners can easily measure running information (time, distance, elevation, heart rate, movement information such as pitch and stride), record it as a running log, and use it for analyzing their own running results (see, for example, Patent Documents 1 and 2). The utilization of such running logs serves as a motivation for continuing and habitualizing running, and has been boosting the popularity of running.

Prior Art Documents

Patent Documents

[0003]

Patent Document 1

Patent Document 2

Summary of the Invention

Problems to be Solved by the Invention

[0004] Here, in a marathon, even a runner who has trained sufficiently may often experience a significant slowdown in the latter half or the final stage of the race. Such a significant slowdown is a phenomenon peculiar to long-distance races and is also called the so-called "30 km wall" or "35 km wall". Since a significant slowdown in the latter half of the race causes a significant delay in the finish time, preventing a significant slowdown is a common problem facing many runners.

[0005] This disclosure has been made in view of such problems, and its object is to provide a technique for predicting the possibility of stalling in a marathon and providing useful information to users. [Means for solving the problem]

[0006] To solve the above problems, a running analysis system in one aspect of the present disclosure includes: a model storage unit that holds a prediction model that predicts the pace changes in the latter half of a race by machine learning using motion history data measured over time from the beginning to the end of a race, with features corresponding to at least several types of motion analysis indicators, including running pace, as features indicating the motion state of a runner in a completed marathon; a measurement value acquisition unit that acquires motion history data measured over time with features corresponding to several types of motion analysis indicators during a run of less than the marathon race distance by a marathon runner; a prediction unit that predicts the pace changes in the latter half of a race by a user based on the motion history data measured over time during the user's run and the prediction model; a decision unit that determines the content of a notification to be sent to the user based on the prediction result by the prediction unit; and an output unit that outputs the notification content based on the decision by the decision unit.

[0007] Another aspect of this disclosure is a running analysis method. This method is a running analysis method for a running analysis system, comprising: a process in which the processor of the running analysis system acquires motion history data measured over time, which includes features corresponding to multiple types of motion analysis indicators, including at least running pace, as features indicating the user's motion state during a run of less than the marathon race distance by a marathon runner; a prediction model that predicts the pace changes in the latter half of a race by machine learning using motion history data measured over time from the beginning to the end of a race, which targets runners in completed marathons; motion history data measured over time during the user's run; a process of predicting the pace changes in the latter half of a race by a user based on these; a process of determining the content of a notification to be sent to the user based on the prediction result; and a process of outputting the notification content based on the determination.

[0008] Another aspect of this disclosure is a running analysis program. This program enables a computer to implement the following: a function to maintain a predictive model that predicts the pace changes in the latter half of a race by machine learning using running history data measured over time from the beginning to the end of a race, with features corresponding to multiple types of motion analysis indicators, including at least running pace, as features indicating the runner's motion state in a completed marathon; a function to acquire running history data measured over time for features corresponding to multiple types of motion analysis indicators during runs of a marathon runner less than the marathon race distance; a function to predict the pace changes in the latter half of a race for the user based on the running history data measured over time and the predictive model; a function to determine the content of a notification to be sent to the user based on the prediction result by the prediction unit; and a function to output the notification content based on the decision by the decision unit.

[0009] Furthermore, any combination of the above components, or any substitution of the components or expressions of this disclosure between methods, apparatus, programs, temporary or non-temporary storage media storing programs, systems, etc., are also valid forms of this disclosure. [Effects of the Invention]

[0010] According to this disclosure, it is possible to predict pace changes during a marathon and provide users with useful information. [Brief explanation of the drawing]

[0011] [Figure 1] This is a diagram showing the configuration of the driving analysis system. [Figure 2] This is a functional block diagram showing the various components of the driving analysis system. [Figure 3] This is a functional block diagram showing the various functions of the driving analysis server. [Figure 4] This diagram illustrates the correspondence between the classification of stall probability, the classification of pace, and the predicted pace. [Figure 5] This diagram illustrates the correspondence between the presence or absence of a stall, the classification of pace, and the predicted pace. [Figure 6]This diagram shows the relationship between the types of motion analysis metrics used in learning and the recall and precision of predictions. [Figure 7] This flowchart provides a schematic illustration of the processing steps in the driving analysis system. [Modes for carrying out the invention]

[0012] In this embodiment, features corresponding to motion analysis indices that indicate the runner's movement state are analyzed based on various information acquired by wearable devices worn by the user while running or devices carried while running. In long-distance races such as marathons, the pace changes in the latter half or final stages of the race are predicted based on motion history data for the distance run in the first half or middle of the race and a trained predictive model. The prediction results are notified to the runner before or during the race, and the user can adjust to an appropriate pace according to the notification.

[0013] When preparing for a marathon, it's common practice to predict your marathon time based on your recent times running shorter distances, such as 10km, a half marathon, or 30km. However, the relationship isn't simply a matter of doubling your half marathon time; the rate of increase in running time is greater than the rate of increase in distance, making simple calculations insufficient for accurate prediction. Many runners use a table of VDOT values ​​("Daniels' Running Formula," by Jack Daniels) to predict their marathon time beforehand and use it as a reference.

[0014] When the distance is longer than a marathon, it may lead to energy consumption exceeding the energy sources originally stored in the human body, or the fatigue in the muscles approaching the limit, resulting in an unintended significant stall in the latter half of the race. This is likely to occur mainly when running at a pace exceeding the lactate threshold in the first half of the race or when the energy replenishment cannot keep up sufficiently. A gradual deceleration stall is still acceptable, but a rapid recovery in a short time is difficult. Therefore, the stall accelerates, and it is not uncommon to be more than 30 minutes slower than the potentially achievable time. Such stalls are likely to occur around 30 - 35 km in the latter half of the race, so they are commonly called the "30 km wall" or "35 km wall".

[0015] On the other hand, if the running pace is excessively reduced to prevent stalling, there is a risk that the runner may fall behind in terms of personal record updates and achieving the target time. Therefore, it is actually difficult to run at a conservative pace in reality. Since many runners want to run at a pace that can maximize their abilities in races aiming for personal record updates, finding the optimal running pace can be said to be an eternal theme for runners.

[0016] In the VDOT value table and prior art, the time in the first distance shorter than a marathon is converted to the time in the second distance, such as a marathon, which is a longer distance. In these cases, it is premised on using the best time when the runner runs at their maximum output in the first distance. As a pre - race practice 2 - 6 weeks before the actual race, many runners do a race - pace run of 20 - 30 km at the target pace of the actual race. However, the time at this pace is considered to be at a slower pace than the best time when running the same distance at maximum output. Therefore, if the time in the pre - race practice is used as the time in the first distance and converted to the time in the second distance based on the VDOT value table and prior art, there is a possibility of an underestimated and conservative race pace or predicted time that undervalues the runner's ability.

[0017] According to the present disclosure, it is possible to predict stalls that often occur in the latter half to the end of a race, prevent the stalls, and optimize the running pace. Also, even for runners who do not have their own valid marathon time as a comparison target, such as first-time marathon runners or runners who have had a long period since their previous marathon, it is possible to predict the target pace and time based on a time less than the most recent marathon, and to formulate a race strategy using the prediction results. Further, based on the body's reaction affected by the environment specific to the actual race, such as the physical condition, weather, and undulation of the course on the race day, which is difficult to reproduce in preliminary training, it is possible to modify the race strategy in real time during the race.

[0018] The pace transition prediction model is learned using data measured in completed marathon races. The data used for learning are feature quantities corresponding to a plurality of types of motion analysis indexes indicating the motion state of a runner. The plurality of types of motion analysis indexes are parameters represented by values measured or detected by the measuring device itself worn or carried by the user who is the runner, or values calculated based on the measured or detected values. Note that "completed marathon" does not strictly mean whether the marathon has been actually held or not, but refers to data that has already been measured until the end of the race as data for learning the prediction model, and is merely expressed as "completed" to distinguish it from data being measured during a "marathon in progress". "Marathon" may be any long-distance race, and the distance needs to be a certain distance for comparison purposes, assuming a predetermined distance such as 42.195 km. Also, it may include marathons competing in other predetermined distances such as half marathons and ultramarathons.

[0019] Hereinafter, the present disclosure will be described with reference to the drawings based on preferred embodiments. In the embodiments and modifications, the same or equivalent components are denoted by the same reference numerals, and repeated explanations will be omitted as appropriate.

[0020] Figure 1 shows the configuration of the running analysis system 100. The running analysis system 100 includes a wristwatch-type device 12, a waist-worn device 14, an information terminal-type device 16, and a running analysis server 60, all of which can be worn by the user 10, who is a runner, during running exercise. The wristwatch-type device 12, the waist-worn device 14, and the information terminal-type device 16 are collectively referred to as the measuring device 20.

[0021] The wristwatch-type device 12 is a sports watch or smartwatch that acquires location information, movement information, etc. The wristwatch-type device 12 includes sensors such as a positioning module, motion sensor, heart rate sensor, and barometer, and acquires information such as date and time, location coordinates, altitude, heart rate, temperature, and cadence. The motion sensor basically consists of an inertial sensor that combines an accelerometer and a gyroscope. The waist-worn device 14 is an electronic device that is worn near the user's waist to acquire location information, movement information, etc. The information terminal-type device 16 is a portable information terminal such as a smartphone that acquires location information and movement information while being held by the user 10 in a pocket or elsewhere.

[0022] User 10 wears one or more measuring devices 20 while running in a race such as a marathon, and acquires location information and motion information. If multiple measuring devices 20 are worn, location information may be acquired with a wristwatch-type device 12 and motion information with a waist-mounted device 14, for example, the devices may be used differently depending on the information to be acquired.

[0023] The measuring device 20 is not limited to devices such as a wristwatch-type device 12, a waist-worn device 14, or an information terminal-type device 16, but may also be a device worn on or inside the runner's shoe. Alternatively, it may be a belt-type device that can be worn around the runner's chest, wrist, waist, or arm to acquire location information, movement information, and heart rate information.

[0024] Information such as distance traveled, time, heart rate, cadence, and stride measured by the measuring device 20 is displayed on the screens of the wristwatch-type device 12 and the information terminal-type device 16 while running. The user 10 can check their running status and activity status by looking at the screens of the wristwatch-type device 12 and the information terminal-type device 16 while running.

[0025] The driving analysis system 100 can be implemented with various hardware and software configurations. For example, the driving analysis system 100 may consist of just one of the following devices: a wristwatch-type device 12, a waist-worn device 14, an information terminal 50, or a driving analysis server 60, or it may consist of a combination of two or more of these devices.

[0026] For example, assuming that the measuring device 20 uses various general-purpose devices to detect the driving state and record it as a driving log, it may consist of only one of the information terminal 50 and the driving analysis server 60, or a combination of both. Alternatively, it may be implemented as a single device that includes all the software configurations included in the information terminal 50 and driving analysis server 60 shown in this figure. Therefore, regardless of the form of its hardware configuration, the driving analysis system 100 only needs to include at least the software configurations of the information terminal 50 and driving analysis server 60 shown in this figure.

[0027] User 10 runs while wearing at least one or all of the following devices as measuring devices 20: a wristwatch-type device 12, a waist-worn device 14, and an information terminal-type device 16. The measuring devices 20 transmit information to the running analysis server 60 via communication and receive analysis results from the running analysis server 60 via communication. However, since the communication means of the wristwatch-type device 12 and the waist-worn device 14 of the measuring devices 20 is short-range wireless communication, they do not communicate directly with the running analysis server 60, but rather synchronize information with the information terminal-type device 16 (which also functions as an "information terminal 50" described in detail later), and the information terminal 50 sends and receives information with the running analysis server 60. Thus, since the wristwatch-type device 12 and the waist-worn device 14 send and receive information with the running analysis server 60 via synchronization with the information terminal 50, it is assumed that the user possesses an information terminal 50. However, if user 10 does not need to receive notifications of prediction results while driving, it is not necessary to wear the information terminal 50 while driving, and synchronization with the information terminal 50 may be done after the driving is completed. As an alternative, the waist-worn device 14 may first synchronize information with the wristwatch-type device 12 via short-range wireless communication, and the wristwatch-type device 12 may then synchronize information with the information terminal-type device 16 (information terminal 50) via short-range wireless communication.

[0028] User 10 wears the measuring device 20 while running, primarily during long-distance running races such as marathons. However, measurements are not limited to during races; they may also be taken during long-distance training runs simulating races, or during shorter practice runs. User 10 starts measuring and recording the running log by operating a button on the measuring device 20 at the start of their run. During the run, the measuring device 20 uses a timer to measure the elapsed time from the start of recording as the running time and records location information for each date and time at predetermined time intervals. The measuring device 20 uses a built-in motion sensor to measure motion information such as pitch (steps per unit time, also called cadence), pelvic rotation and translation, and impact values. The measuring device 20 uses a built-in optical heart rate monitor to measure User 10's heart rate.

[0029] During the run and after the run log recording is completed, the measuring device 20 transmits information such as run time, location information, movement information, and heart rate to the run analysis server 60. When transmitting during the run, it may be transmitted at predetermined measurement intervals. A "predetermined measurement interval" refers to a measurement interval based on a predetermined elapsed time, such as every minute or every 5 minutes, or a measurement interval based on a predetermined running distance, such as every 100m, every 1km, or every 5km.

[0030] The measuring device 20 may calculate information such as running time, running distance, running speed, pitch, stride, altitude, and incline based on time information and location information, and include this calculated information in the running log and transmit it to the running analysis server 60. The measuring device 20 may also calculate various motion information of the runner based on information detected by the motion sensor, and include this calculated information in the running log and transmit it to the running analysis server 60. The measuring device 20 may also acquire weather information such as temperature, humidity, weather, wind direction, and wind speed corresponding to the time of running from a predetermined server, and include this weather information in the running log and transmit it to the running analysis server 60.

[0031] The information terminal 50 may be an information terminal such as a smartphone or tablet, or it may be a personal computer. The driving analysis server 60 is a server computer connected to the internet that sends and receives data with multiple user 10 information terminals 50. The driving analysis server 60 acquires information such as time information, location information, movement information, and heart rate as driving log data of user 10 received from the information terminal 50, along with the user 10's identification information and attribute information, and stores it along with various types of driving analysis index feature quantities calculated from each piece of information. The driving analysis server 60 may also acquire weather information such as temperature, humidity, weather, wind direction, and wind speed corresponding to the time of driving from a predetermined server and include that weather information in the driving log data for storage. In response to a request from the information terminal 50, the driving analysis server 60 sends the stored driving log data and analysis results based on the feature quantities of multiple types of driving analysis indexes to the information terminal 50.

[0032] The information terminal 50 and the driving analysis server 60 may be composed of a computer consisting of a CPU (Central Processing Unit), GPU (Graphics Processing Unit), RAM (Random Access Memory), ROM (Read Only Memory), auxiliary storage device, communication device, etc. The information terminal 50 and the driving analysis server 60 may each be composed of separate computers, or they may be implemented in a single computer or information terminal that combines the functions of both. In this embodiment, an example in which they are implemented in separate computers will be described.

[0033] Figure 2 is a functional block diagram showing the various components of the running analysis system 100. Figure 2 depicts functional blocks for the measuring device 20 and the information terminal 50, each realized through the coordination of various hardware and software configurations. Therefore, it will be understood by those skilled in the art that these functional blocks can be realized in various ways using hardware alone, software alone, or a combination thereof. The measuring device 20 is composed of a combination of hardware such as a microprocessor, display device, memory, communication module, positioning module, motion sensor, and optical heart rate monitor. The information terminal 50 is composed of a combination of hardware such as a microprocessor, touch panel, memory, communication module, positioning module, and motion sensor. The functions of the measuring device 20 and the information terminal 50 will be described below.

[0034] The measuring device 20 includes a communication unit 21, a time measurement unit 22, a position measurement unit 24, a motion detection unit 26, and a calculation unit 28. For example, a waist-worn device 14 is used as the measuring device 20. The time measurement unit 22 measures the running start time, i.e., the running time from the measurement start time, by counting a timer. The position measurement unit 24 measures the current position of the runner, user 10, using position information received from a satellite positioning system by a positioning module such as a GPS module. The motion detection unit 26 detects motion information such as the pitch of the runner, user 10, the rotation and translational movement of the pelvis, or impact values ​​using a motion sensor.

[0035] The calculation unit 28 calculates feature quantities corresponding to multiple types of motion analysis indicators based on travel time, location information, and motion information. The location information, motion information, and feature quantities corresponding to multiple types of motion analysis indicators are transmitted to the information terminal 50 via the communication unit 21. Although an example has been described in which the calculation unit 28 in the measuring device 20 calculates the feature quantities corresponding to multiple types of motion analysis indicators, these may also be calculated in the measurement value acquisition unit 40 in the information terminal 50, as described later, instead of the measuring device 20.

[0036] A wristwatch-type device 12 or an information terminal-type device 16 can also be used as the measuring device 20. When the wristwatch-type device 12 is used as the measuring device 20, the motion detection unit 26 of the wristwatch-type device 12 detects motion information such as the user 10's pitch using a motion sensor and detects physiological indicator information such as heart rate using an optical heart rate monitor. Physiological indicator information may also be included in the motion information. When the information terminal-type device 16 is used as the measuring device 20, the motion detection unit 26 of the information terminal-type device 16 detects motion information such as the user 10's pitch using a motion sensor. The information terminal 50 may also function as the information terminal-type device 16 used as the measuring device 20; in this case, for example, a single mobile terminal such as a smartphone may have all the functions of both the measuring device 20 and the information terminal 50. When the information terminal 50 is not used as the measuring device 20, the information terminal 50 is not limited to a smartphone; it may also be a tablet terminal or personal computer owned by the user 10.

[0037] The information terminal 50 includes an information acquisition unit 30, a measurement value acquisition unit 40, an input / output unit 51, and a communication unit 52. The information acquisition unit 30 receives location information and motion information of the user 10, who is a runner in a race, measured or detected at each progress point during the run by a measuring device 20 worn by the user 10, via the communication unit 52. Here, "progress point" refers to a point in time or distance during a run such as a race, and location information and motion information are recorded in the measuring device 20 for each time or distance progress point. Location information includes information such as the date and time of measurement, location coordinates, and altitude. Motion information includes information such as pitch, rotation and translational motion of the pelvis, or impact value. The information acquisition unit 30 may synchronize information with the measuring device 20 during the run and acquire location information and motion information as information on the running state during the run by the measuring device 20, or it may acquire the location information and motion information for the entire run all at once from the measuring device 20 after the run is completed.

[0038] However, as will be described later, users who wish to receive notifications of predicted pace changes in the latter half or final stages of a race while running need to synchronize information with the measuring device 20 during the run to obtain information on the progress of the race. Users who wish to obtain predicted pace changes in the latter half or final stages of a race as preliminary information for the race, based on the results of a practice run before the actual marathon, may obtain location information and movement information for the entire run all at once from the measuring device 20 after the run is completed.

[0039] The measurement value acquisition unit 40 acquires or calculates feature quantities of multiple types of motion analysis indicators that indicate the user's driving motion state for each predetermined measurement interval, based on the time-series position information and motion information acquired by the information acquisition unit 30.

[0040] The measurement acquisition unit 40 acquires, for example, lap pace, lap time, running time, running speed, pitch [steps / min], stride [m], stride-to-height ratio, trunk posterior tilt [°], vertical movement [cm], vertical movement-to-height ratio [%], hip sinking [%], pelvic lateral tilt [°], pelvic elevation [°], pelvic rotation [°], pelvic rotation timing, and lateral impact [m / s]. 2 ], push-off time [ms], ground contact time [ms], ground contact time percentage [%], landing impact [m / s] 2 ], kick-off acceleration [m / s 2 The system acquires feature quantities for multiple types of motion analysis indicators, such as [m / s], deceleration, and stiffness [kN / m·kg]. The motion analysis indicators may also include heart rate. The measurement acquisition unit 40 records the acquired feature quantities for multiple types of motion analysis indicators as time-series continuous motion history data.

[0041] The input / output unit 51 transmits the location information and operation information acquired by the measuring device 20, and the operation history data recorded by the measurement value acquisition unit 40, along with the user 10's attribute information, to the driving analysis server 60 via the communication unit 52 as a driving log. The input / output unit 51 displays the location information and operation information acquired by the measuring device 20 and the operation history data recorded by the measurement value acquisition unit 40 on the screen, and also displays information such as analysis results received from the driving analysis server 60 on the screen. The information received from the driving analysis server 60 includes information on race time and pace change prediction results as notification content to the user 10 determined by the driving analysis server 60. The input / output unit 51 displays the race prediction results received from the driving analysis server 60 on the screen. The input / output unit 51 may also notify the user 10 of the race prediction results received from the driving analysis server 60 in the form of an alarm by voice or vibration. Furthermore, the method of notifying the user 10 of the prediction results may involve transmitting the notification content to the measuring device 20, which then outputs a screen, audio, and vibration. The input / output unit 51 accepts operation input from the user 10. The input / output unit 51 may be configured as a touch panel in terms of hardware.

[0042] In this embodiment, an example was described in which the measurement value acquisition unit 40 acquires or calculates feature quantities of multiple types of motion analysis indicators. However, by assigning the functions of the measurement value acquisition unit 40 to the measurement device 20 or the driving analysis server 60, the information terminal 50 may not calculate the feature quantities of the motion analysis indicators. In the following description of the driving analysis server 60, an example will be described in which the driving analysis server 60 has a measurement value acquisition unit 80 that corresponds to the functions of the measurement value acquisition unit 40.

[0043] Figure 3 is a functional block diagram showing the functions of the driving analysis server 60. It depicts functional blocks that are realized by the cooperation of various hardware and software configurations for the driving analysis server 60. Therefore, it will be understood by those skilled in the art that these functional blocks can be realized in various ways by hardware alone, software alone, or a combination thereof. The driving analysis server 60 is composed of a combination of hardware such as a microprocessor, memory, display, and communication module. The driving analysis server 60 includes a communication unit 62, an information acquisition unit 64, a data storage unit 66, a learning processing unit 70, a measurement value acquisition unit 80, a model storage unit 84, a prediction unit 86, a decision unit 88, and an output unit 99.

[0044] The information acquisition unit 64 acquires the running log via the communication unit 62 and stores it in the data storage unit 66. The information acquisition unit 64 has a function equivalent to the information acquisition unit 30 of the measuring device 20. In other words, the information acquisition unit 64 acquires the location information and movement information of the user 10, who is the runner in the race, measured at each point along the run by the measuring device 20 worn by the user 10.

[0045] The measurement value acquisition unit 80 has a function equivalent to the measurement value acquisition unit 40 of the information terminal 50. That is, based on the information acquired by the information acquisition unit 64, the measurement value acquisition unit 80 acquires or calculates feature quantities of multiple types of motion analysis indicators that indicate the user's running motion state for each predetermined measurement interval, based on the time-series location information and motion information. The measurement value acquisition unit 80 acquires motion history data, which is measured over time by user 10, a marathon runner, during runs of less than the marathon race distance, and stores the feature quantities corresponding to multiple types of motion analysis indicators in the data storage unit 66.

[0046] The learning processing unit 70 uses machine learning on motion history data of numerous runners measured in already completed marathons to create a predictive model that predicts pace changes in the latter half or final stages of a race. The learning processing unit 70 performs machine learning using motion history data measured over time from the beginning to the end of the race, with features corresponding to multiple types of motion analysis indicators as features that indicate the runner's motion state. The predictive model created by the learning processing unit 70 is stored in the model storage unit 84.

[0047] The multiple types of motion analysis indicators used for learning may include at least running pace, as well as parameters that directly affect running pace, such as pitch and stride, and parameters that contribute to running efficiency (also called "running economy"), such as deceleration and vertical movement. The more motion analysis indicators used for learning, the higher the accuracy of stall prediction, but even when only running pace is used for learning, it is possible to predict with a certain degree of accuracy. The differences in prediction accuracy depending on which of the multiple types of motion analysis indicators acquired by the measurement value acquisition unit 80 is used for learning will be described later.

[0048] The features of the various motion analysis metrics used for learning may be preprocessed, such as by using representative values ​​for each interval when the race is divided into multiple intervals. Representative values ​​for the features may include values ​​that estimate the center of the distribution and values ​​that estimate the variability of the distribution, such as the mean and variance. Alternatively, the median and mean absolute deviation (MAD) may be used as representative values ​​for the features. By dividing the data into intervals and using representative values, overfitting due to an excessive number of feature variables can be suppressed, and the learning process can be reduced.

[0049] Each section is defined as a fixed distance, such as 1km or 5km, which is used as the unit for measuring lap times. However, as a variation, each section may be defined as a fixed time unit, such as 1 minute or 5 minutes. Regarding the average value, the average value itself may be used for the first section, while for the second and subsequent sections, the difference from the average value of the previous section may be used. The average value of the first section represents the characteristics of the running form, the difference from the average value of the previous section represents the characteristics of the change in running form, and the mean absolute deviation represents the characteristics of the reproducibility of the running form.

[0050] The pre-trained model used for prediction may be a regression model that uses features corresponding to multiple types of motion analysis indicators included in the motion history data as explanatory variables, and the ratio of the deceleration of the average pace in the second section of the race relative to the average pace in the first section of the race in the first half as the dependent variable. Alternatively, the pre-trained model used for prediction may be a classification model that uses features corresponding to multiple types of motion analysis indicators included in the motion history data as explanatory variables, and the dependent variable may be whether or not a deceleration of a predetermined ratio or more occurred in the second section of the race relative to the average pace in the first section of the race in the first half.

[0051] The prediction unit 86 predicts the pace progression of user 10 in the latter half of the race based on the operation history data measured over time during the user's run and a prediction model. The content predicted by the prediction unit 86 can take various forms, and the form is not limited to one, but may be a combination of multiple predictions.

[0052] One aspect of what the predictive model predicts is classifying the "probability of slowing down," which indicates the probability of slowing down in the latter half or final stages of a race, into several classes and predicting which class it falls into. In this case, a classification model such as a support vector machine or a neural network may be used as the pre-training model for the predictive model. In this embodiment, "slowing down" is defined in terms of how the relationship between distance, time, and pace changes from the first half of the race to the second half. For example, it is defined as a decrease of 15% or more in the average pace from 25km onwards in the second half of the race compared to the average pace in a predetermined section in the first half of the race, such as from 5km to 20km. This is because many runners begin to gradually slow down from around 25km in cases of slowing down.

[0053] Another aspect of what the prediction model can predict is whether or not a slowdown occurs in the latter half or final stages of the race, expressed as a binary value. In this case, the pre-training model used in the prediction model may be a classification model such as logistic regression, where the target variable is a binary value of 0 or 1 indicating whether or not a slowdown occurred.

[0054] Other aspects of what the prediction model predicts include the probability of a slowdown occurring in the latter half or final stages of the race, the time-series change in the rate of deceleration between the pace in the first half of the race and the pace in the second half, and the remaining time until completion. In this case, various regression models or statistical models can be used as the pre-trained model for the prediction model. For example, in addition to neural network models such as recurrent neural networks (RNNs) and Transformers, nonlinear mixed-effects models and functional simultaneous regression (FCR) models may also be used.

[0055] The decision unit 88 determines the content of the notification to be sent to the user 10 based on the prediction results from the prediction unit 86. The decision unit 88 determines the content of the notification to suggest the possibility of the user 10 slowing down in the latter half of the race. The decision unit 88 may also determine the content of the notification to suggest countermeasures, including a recommended pace, in accordance with the possibility of the user 10 slowing down in the latter half of the race. If there is a high possibility of slowing down, the decision unit 88 may also determine suggestions for improvement regarding the features of motion analysis indicators that could be considered as causes of slowing down.

[0056] The output unit 99 outputs the notification content to the information terminal 50 via the communication unit 62 based on the decision made by the determination unit 88. Depending on the notification content, the output unit 99 causes the information terminal 50 or measuring device 20 carried by the runner user 10 to execute an output using at least one of the following: voice, image, or vibration.

[0057] In this embodiment, an example was described in which the running analysis server 60 performs prediction processing as a configuration for predicting pace changes in the latter half of the race. In a modified example, the prediction of pace changes in the latter half of the race may be performed on the information terminal 50 without using processing on the running analysis server 60. In this case, some or all of the components described in Figure 3 may be provided to the measuring device 20 or the information terminal 50. However, in order to avoid heavy loads, the prediction processing using the prediction model in the measuring device 20 or the information terminal 50 during the race may be simplified to allow prediction with simplified calculation processing. For example, a determination process may be performed to classify the probability of stalling based on whether the feature quantities of a predetermined type of motion analysis index satisfy predetermined conditions.

[0058] As another variation, by further training the machine learning model with information indicating the type and attributes of shoes worn by runners during completed marathons, it may be possible to predict the pace changes in the latter half of the race according to the type and attributes of shoes worn by user 10. In particular, by training the model with attribute information such as sole thickness and whether or not the shoes contain a carbon plate, it becomes possible to predict before the race the advantage of user 10 wearing thick-soled shoes with carbon plates, which are mainly considered for advanced runners.

[0059] Figure 4 illustrates the correspondence between the classification of stall probability, the classification of pace, and the predicted pace. In the predicted pace table 110, the first column 111 shows the classification of stall probability, the second column 112 shows the pace classification, and the third column 113 shows the predicted pace. The first row 114 shows the case where the stall probability is "high probability" or "70-100%", the second row 115 shows the case where the stall probability is "medium probability" or "30-70%", and the third row 116 shows the case where the stall probability is "low probability" or "0-30%". The prediction unit 86 is assumed to predict one of the following using a prediction model based on the classification model: "high probability" in the first row 114, "medium probability" in the second row 115, or "low probability" in the third row 116. Alternatively, if the prediction unit 86 predicts the stall probability itself using a prediction model based on a regression model, it will be classified into one of the following categories: "70-100%" in the third column 113, "30-70%" in the second row 115, or "0-30%" in the third row 116.

[0060] If the predicted probability of stalling based on user 10's activity history data is a high 70-100%, then continuing at the same pace carries a high risk of stalling in the latter half of the race. Therefore, the decision unit 88 determines a notification recommending a generally more restrained pace for the first pace 117, second pace 118, and third pace 119. In this case, since the average pace from the activity history data is considered to be close to the state where user 10 is exerting maximum power, the second pace 118 is shown as the "appropriate pace" calculated from the VDOT value table or the pace and time of maximum power, as in conventional technology. As a pre-suggestion to user 10, the output unit 99 notifies the user that this pace will make it easier to maintain an even pace until the end. For user 10 during the race, the decision unit 88 determines a notification that suggests slowing down by the difference between the current pace and the "appropriate pace," displaying the recommended pace and completion time, and issuing a warning to encourage slowing down. User 10 suggests that slowing down to an "appropriate pace" increases the likelihood of avoiding a slowdown in the latter half or final stages of the race.

[0061] A pace approximately 3% (for example, 5-10 seconds) faster than the "appropriate pace" is indicated as the "challenge pace" for the first pace 117. In this case, the risk of slowing down is higher compared to the "appropriate pace," but it is a more restrained pace than the average pace of the operation history data. As a pre-race suggestion to user 10, if user 10's physical condition and the weather are good on the day of the race, or if user 10 has a strong desire to challenge themselves to achieve a record, the output unit 99 notifies the user along with advice on how to replenish fluids and energy and how to adjust the pace flexibly. For user 10 while running, the decision unit 88 determines the content of the notification, which includes displaying the challenge pace and the estimated completion time as a minimum deceleration suggestion, and a warning to encourage minimal deceleration.

[0062] A pace approximately 3% (for example, 5-10 seconds) slower than the "appropriate pace" is indicated as the "safe pace" for the third pace 119. As a pre-race suggestion to user 10, if user 10 has concerns about their physical condition or the weather on race day, or if they want to avoid a major slowdown and finish safely, the output unit 99 notifies them of advice regarding increasing their pace according to their physical condition and energy level from the middle of the race onward. For user 10 while they are running, the decision unit 88 determines the content of the notification, such as a suggestion to slow down significantly, displaying the safe pace and the time it will take to complete the race, and a warning to encourage significant deceleration.

[0063] In the modified version, the system may be configured to allow users to specify, as attribute information, which marathon strategy they intend to pursue: even pace, negative split, or positive split. An even pace is a strategy that aims to maintain a consistent pace from the beginning to the end of the marathon. A negative split is a strategy in which the pace is kept low in the first half and increased in the second half, resulting in a shorter time in the second half than in the first. A positive split is a strategy in which the pace is set aggressively from the beginning, and even if the time drops in the second half, the user perseveres until the end, resulting in a shorter time in the first half than in the second half. In this case, the decision unit 88 may select one of the following according to user 10's marathon strategy preference: "challenging pace," "appropriate pace," or "safe pace," or it may determine advice content in the notification that matches user 10's marathon strategy preference.

[0064] If the predicted probability of stalling based on user 10's movement history data is a moderate 30-70%, then the risk of stalling in the latter half of the race by continuing at the same pace is not necessarily high, but moderate. In this case, since the average pace of the movement history data is considered to be lower than user 10's maximum output, predicting the race pace and time in the latter half of the race using a VDOT value table or conventional technology would result in an underestimation. Therefore, without using a VDOT value table or conventional technology, a race pace and predicted time that is almost the same as the average pace of the movement history data is shown as the "appropriate pace" for the 5th pace 121. In this case, since there is still some risk of stalling, the decision unit 88 determines a notification that advises proactive energy replenishment. In addition, a pace approximately 3% (e.g., 5-10 seconds) faster than the "appropriate pace" is shown as the "challenging pace" for the 4th pace 120. A pace approximately 3% (e.g., 5-10 seconds) slower than the "appropriate pace" is shown as the "safe pace" for the 6th pace 122.

[0065] If the predicted probability of slowing down based on user 10's activity history data is a low 0-30%, then the risk of slowing down in the latter half of the race is low if the user continues running at the same pace. In this case, the average pace of the activity history data is considerably lower than user 10's maximum output, and it is considered possible to increase the pace in the latter half of the race. Therefore, a pace approximately 3% (e.g., 5-10 seconds) faster than the average pace of the activity history data is shown as the "appropriate pace" for the 8th pace 124. In this case, the decision unit 88 may decide on the notification content without giving user 10 any specific advice, or it may decide on positive notification content to reassure user 10, such as "You're doing well." Furthermore, a pace approximately 3% (e.g., 5-10 seconds) faster than the "appropriate pace" is shown as the "challenging pace" for the 7th pace 123. A pace roughly equivalent to the average pace of the activity history data is shown as the "safe pace" for the 9th pace 125.

[0066] Figure 5 illustrates the correspondence between stalling / delaying, pace classification, and predicted pace. In the predicted pace table 130, the first column 131 indicates stalling / delaying, the second column 132 indicates pace classification, and the third column 133 indicates predicted pace. The first row 134 indicates the case where the predicted result for stalling / delaying is "stalling will occur," and the second row 135 indicates the case where the predicted result for stalling / delaying is "not stalling." The prediction unit 86 predicts "stalling will occur" in the first row 134 or "not stalling" in the second row 135 using a prediction model that uses a classification model.

[0067] If the prediction of whether user 10 will stall based on their activity history data is "they will stall," then continuing at the same pace carries a high risk of stalling in the latter half of the race. Therefore, the decision unit 88 determines a notification recommending a generally more restrained pace for the first pace 136, second pace 137, and third pace 138. These paces are the same as those shown in Figure 4 as the first pace 117, second pace 118, and third pace 119.

[0068] If the prediction of whether user 10 will stall based on user 10's activity history data is "no stalling," then a pace approximately 3% (for example, 5-10 seconds) faster than the "appropriate pace" is shown as the "challenge pace" for the first pace 117. In this case, the risk of stalling is higher than at the "appropriate pace," but it is a more restrained pace than the average pace from the activity history data. As a pre-race suggestion to user 10, if user 10's physical condition and the weather are good on the day of the race, or if user 10 has a strong desire to challenge themselves to achieve a record, the output unit 99 notifies the user along with advice on how to replenish fluids and energy and how to adjust their pace flexibly. For user 10 while they are running, the decision unit 88 determines the content of the notification, which includes the display of the challenge pace and its completion time as a minimum deceleration suggestion, and a warning to encourage minimal deceleration.

[0069] Figure 6 shows the relationship between the types of motion analysis metrics used for learning and the prediction recall and precision. The applicant performed machine learning using a binary classification model that predicts whether or not a stall occurs based on motion history data of many runners, and performed predictions using the trained model. The prediction recall and precision for each combination of motion analysis metrics are shown in Motion Analysis Metric Table 150. Column 151 shows the types of motion analysis metrics used for learning as explanatory variables, Column 252 shows the prediction recall, and Column 353 shows the prediction precision. Prediction recall is the proportion of cases where a stall was predicted to occur based on motion history data when a stall actually occurred. Prediction precision is the proportion of cases where a stall actually occurred when a stall was predicted based on motion history data.

[0070] The recall and precision when features corresponding to all types of motion analysis indicators acquired by the measurement acquisition unit 40 are used as explanatory variables for learning are shown in row 154. In this case, the recall and precision were the highest, with a recall of 80% and a precision of 80%. However, when features of a large number of parameters, such as more than 20 types of motion analysis indicators, are used as explanatory variables for learning, the time and load required for machine learning increase. Furthermore, calculating and frequently sending and receiving features of all these motion analysis indicators during a race for use in prediction also places a heavy processing load on the system. Therefore, it is desirable to strike a balance between the number of parameters used as explanatory variables for learning and the prediction accuracy.

[0071] Here, when all features other than pace, pitch, stride, deceleration, and vertical movement from the multiple types of motion analysis indicators acquired by the measurement unit 40 are used as explanatory variables for learning, the recall and precision are 60% and 65%, respectively (line 6, 159). It can be seen that despite still using a large number of parameters, this is a significant decrease compared to when features corresponding to all types of motion analysis indicators are used as explanatory variables for learning (line 1, 154). Furthermore, when all features other than pace from the multiple types of motion analysis indicators acquired by the measurement unit 40 are used as explanatory variables for learning, the recall and precision are 65% and 70%, respectively. It can be seen that despite still using a large number of parameters, this is a significant decrease compared to when features corresponding to all types of motion analysis indicators are used as explanatory variables for learning (line 1, 154).

[0072] On the other hand, when only pace is used as an explanatory variable for training, the recall and precision are 75% and 77%, respectively (row 4, page 157). Despite having only one parameter, the difference from when features corresponding to all types of motion analysis indicators are used as explanatory variables for training (row 1, page 154) is not necessarily large. Furthermore, when pace, pitch, and stride are used as explanatory variables for training, the recall and precision are 77% and 78%, respectively (row 3, page 156), showing a slight increase. Moreover, when pace, pitch, stride, deceleration, and vertical movement are used as explanatory variables for training, the recall and precision are 78% and 79%, respectively (row 2, page 155). While the recall decreases slightly, the precision is actually higher than when all of the multiple types of motion analysis indicators are used as explanatory variables for training.

[0073] Based on the above, it is preferable that the features of the motion analysis indicators used for learning as explanatory variables include at least pace, and it is even preferable to further include pitch and stride, and it is even preferable to further include deceleration and vertical movement. Furthermore, if additional motion analysis indicators are added with the aim of further improving the recall and precision shown in line 2, paragraph 155, indicators related to running economy are considered to be more likely to be related to stalling, so indicators such as ground contact time, ground contact time percentage, stiffness, and trunk posterior tilt may be added. Heart rate may also be added as an explanatory variable. Heart rate is an effective physiological indicator for determining whether or not the lactate threshold is exceeded, and therefore may improve the accuracy of stall prediction. However, even without adding heart rate as an explanatory variable, changes in the runner's posture and movement may appear earlier than changes in physiological indicators, so by using pace as an explanatory variable at least, it is possible to capture the signs of stalling and ensure sufficient prediction accuracy.

[0074] Figure 7 is a flowchart illustrating the processing steps in the running analysis system. This flowchart primarily explains the relationship between the timing of acquiring running logs during a race, the timing of predicting pace changes in the latter half of the race based on features corresponding to motion analysis indicators, and the timing of notifying the prediction results.

[0075] When user 10 starts recording a running log using a measuring device 20 such as a wristwatch-type device 12 or an information terminal-type device 16 during a marathon race (S10), the measuring device 20 acquires location information and motion information (S12), and calculates feature quantities corresponding to multiple types of motion analysis indicators based on the acquired location information and motion information (S14). Until a predetermined time interval or predetermined distance interval has elapsed (N in S16), steps S12 and S14 are repeated, and once the predetermined time interval or predetermined distance interval has elapsed (Y in S16), the motion history data is sent to the information terminal 50 or the running analysis server 60 (S18). If user 10 instructs to end recording the running log (Y in S20), the process ends. If there is no instruction to end recording (N in S20), the process from S12 to S20 is repeated until the first half of the race has elapsed (N in S22). It should be noted that the first half of the race, as used here, does not strictly have to be half the distance of a marathon. It can be defined as a predetermined distance where signs of slowing down tend to begin, for example, 25km, and the section up to passing 25km can be considered the first half of the race.

[0076] When the 25km mark, which is the first half of the race, has passed (Y in S22), the running analysis server 60 predicts the pace change in the second half of the race based on the operation history data up to that point and the prediction model (S24). Based on the prediction result, the running analysis server 60 determines the content of the notification to the user 10 (S26) and outputs the notification content to the information terminal 50 (S28). The information terminal 50 notifies the user 10 of the content based on the information received from the running analysis server 60 (S30).

[0077] The present disclosure has been described above based on embodiments. The embodiments are illustrative, and it will be understood by those skilled in the art that various modifications are possible in combinations of their components and processing processes, and that such modifications are also within the scope of the present disclosure. Furthermore, the above-described embodiments can be generalized to obtain the following embodiments.

[0078] [Aspect 1] A model memory unit holds a predictive model that predicts pace changes in the latter half of a race using machine learning with motion history data measured over time from the beginning to the end of a race, which includes features that represent the motion state of runners in completed marathons, and features corresponding to multiple types of motion analysis indicators, including at least running pace. A measurement value acquisition unit that acquires motion history data by measuring feature quantities corresponding to the multiple types of motion analysis indicators over time during a run by a user who is a marathon runner, which is less than the distance of the marathon race, A prediction unit predicts the pace progression of the user in the latter half of the race based on the operation history data measured over time during the user's run and the prediction model. A determination unit that determines the content of the notification to be sent to the user based on the prediction result from the prediction unit, An output unit that outputs notification content based on the decision made by the aforementioned decision unit, A driving analysis system characterized by having the following features.

[0079] In the running analysis system of Embodiment 1, it is possible to obtain prediction results regarding the likelihood of slowing down in the second half of a marathon race based on running data from practice runs that are not at maximum power and running data from the first half of the race. In addition, the runner can receive notifications before or during the race according to the likelihood of slowing down, allowing them to easily adjust to an appropriate pace.

[0080] [Aspect 2] The measurement acquisition unit acquires motion history data by measuring feature quantities corresponding to the multiple types of motion analysis indicators for runners in a marathon in progress over time until partway through the race. The running analysis system according to embodiment 1, characterized in that the prediction unit predicts the pace progression from a measurement point during the race to the end of the race based on the operation history data measured over time up to the middle of the race and the prediction model.

[0081] In the driving analysis system of the second embodiment, based on the body's reactions influenced by the unique environment of the actual race, such as physical condition, weather, and course undulations, which are difficult to reproduce in practice runs, it is possible to easily grasp the possibility of slowing down in the second half of the race in real time during the race and adjust the race strategy accordingly.

[0082] [Aspect 3] The running analysis system according to embodiment 1 or 2, characterized in that the aforementioned multiple types of motion analysis indicators further include at least one of pitch and stride.

[0083] In the running analysis system of embodiment 3, it is possible to predict the pace changes in the latter half of the race with higher accuracy than when pace is used as the explanatory variable alone.

[0084] [Aspect 4] The driving analysis system according to any one of embodiments 1 to 3, characterized in that the aforementioned multiple types of motion analysis indicators further include at least one of multiple types of motion analysis indicators that contribute to driving efficiency, including deceleration and vertical movement.

[0085] In the running analysis system of Embodiment 4, it is possible to predict the pace changes in the latter half of the race with higher accuracy than when pace is used as the explanatory variable.

[0086] [Aspect 5] The aforementioned prediction model is a machine learning-based regression model that uses features corresponding to multiple types of motion analysis indicators included in the motion history data as explanatory variables, and the rate of deceleration of the average pace in the second section of the race relative to the average pace in the first section of the race in the first half as the dependent variable. The running analysis system according to any one of embodiments 1 to 4, characterized in that the prediction unit predicts the rate of deceleration of the user's pace in the latter half of the race as the pace progression.

[0087] In the running analysis system of Embodiment 5, the user, who is a runner, can objectively understand how much deceleration may occur in the latter half of the race and adjust their pace accordingly.

[0088] [Aspect 6] The aforementioned prediction model is a classification model developed through machine learning, with explanatory variables being features corresponding to multiple types of motion analysis indicators included in the motion history data, and the dependent variable being whether or not a deceleration of a predetermined percentage or more occurred in the second section of the race relative to the average pace of the first section in the first half of the race. The driving analysis system according to any one of embodiments 1 to 5, characterized in that the prediction unit predicts whether or not a deceleration of a predetermined percentage or more will occur in the latter half of the user's race as a pace progression.

[0089] In the running analysis system of embodiment 6, the user, who is a runner, can objectively understand the possibility of slowing down in the latter half of the race and adjust their pace accordingly.

[0090] [Aspect 7] The driving analysis system according to any one of embodiments 1 to 6, characterized in that the decision unit determines, as the notification content, content that suggests the possibility of the user slowing down in the latter half of the race.

[0091] In the driving analysis system of embodiment 7, the possibility of slowing down in the latter half of the race can be objectively identified and the pace can be adjusted accordingly.

[0092] [Aspect 8] The driving analysis system according to any one of embodiments 1 to 7, characterized in that the decision unit determines, as the notification content, content that suggests countermeasures including a recommended pace according to the user's possibility of deceleration in the latter half of the race.

[0093] In the driving analysis system of embodiment 12, the possibility of losing speed in the latter half of the race can be objectively identified and prevented before it occurs.

[0094] [Aspect 9] The running analysis system according to any one of embodiments 1 to 8, characterized in that the output unit causes a device carried by the runner to output at least one of the following: voice, image, or vibration, in accordance with the notification content.

[0095] In the driving analysis system of embodiment 9, the possibility of slowing down in the latter half of the race can be easily identified, and the pace can be adjusted accordingly.

[0096] [Aspect 10] A position measurement unit that acquires the runner's location information, A motion detection unit that detects the runner's movements, The system further comprises a calculation unit that calculates feature quantities corresponding to multiple types of motion analysis indicators based on the position information and motion information, The driving analysis system according to any one of embodiments 1 to 9, characterized in that the measurement value acquisition unit acquires the feature quantities calculated by the calculation unit.

[0097] In the driving analysis system of embodiment 10, the possibility of slowing down in the latter half of the race can be easily grasped by a device worn by the user, and the pace can be adjusted accordingly.

[0098] [Aspect 11] A driving analysis method for a driving analysis system, The processor of the aforementioned driving analysis system A process for acquiring motion history data by measuring over time features corresponding to multiple types of motion analysis indicators, including at least running pace, as features representing the user's motion state during running of a distance less than the marathon race distance, by a user who is a marathon runner, A predictive model that predicts the pace changes in the latter half of a race by machine learning using motion history data measured over time from the beginning to the end of a race, with features corresponding to multiple types of motion analysis indicators targeting runners in completed marathons; and a process that predicts the pace changes in the latter half of a race for the user based on motion history data measured over time during the user's running. A process of determining the content of the notification to be sent to the user based on the results of the prediction, The process of outputting the notification content based on the aforementioned decision, A driving analysis method characterized by comprising the following features.

[0099] In the running analysis method of embodiment 11, it is possible to obtain prediction results regarding the likelihood of slowing down in the second half of a marathon race based on running data from practice runs that are not at maximum power and running data from the first half of the race. In addition, the runner can receive notifications before or during the race according to the likelihood of slowing down, allowing them to easily adjust to an appropriate pace.

[0100] [Aspect 12] The system has a function to maintain a predictive model that predicts pace changes in the latter half of a race using machine learning with motion history data measured over time from the beginning to the end of the race, which includes features that represent the motion state of runners in completed marathons, and features corresponding to at least multiple types of motion analysis indicators, including running pace. A function to acquire motion history data by measuring feature quantities corresponding to the multiple types of motion analysis indicators over time during runs of a user who is a marathon runner, which are shorter than the distance of the marathon race, A function that predicts the pace progression of the user in the latter half of the race based on the operation history data measured over time during the user's run and the prediction model, A function to determine the content of the notification to be sent to the user based on the results of the prediction, A function to output notification content based on the aforementioned decision, A driving analysis program characterized by enabling a computer to perform this analysis.

[0101] In the running analysis program of embodiment 12, prediction results regarding the likelihood of slowing down in the second half of a marathon race can be obtained based on running data from practice runs shorter than a marathon (not at maximum power) and running data from the first half of the race. In addition, the runner can receive notifications before or during the race according to the likelihood of slowing down, allowing them to easily adjust to an appropriate pace.

[0102] In a modified version, the learning processing unit 70 may create a different prediction model for each marathon race and use a prediction model specialized for the marathon race in which the user is participating. That is, the data for runners in the same marathon will have the same weather conditions and course topography, and by creating a prediction model that has been trained only on such identical data, it is possible to provide more accurate prediction results to users running the same marathon course. In yet another modified version, the accuracy of the prediction results may be improved by extracting and using prediction models for marathons that have similar weather conditions and course topography from multiple models.

[0103] In yet another variation, different prediction models could be created for each level of running ability, and predictions could be made using the model that best suits the user's running ability level. For example, a prediction model could be created by dividing runners into those with personal best marathon times of less than 2.5 hours, less than 3 hours, less than 3.5 hours, less than 4 hours, and 4 hours or more, and training the data of each runner. By using such prediction models tailored to different levels of running ability, it is possible to provide users with more accurate prediction results. [Explanation of Symbols]

[0104] 10 Users, 20 Measuring devices, 24 Position measurement unit, 26 Motion detection unit, 28 Calculation unit, 40 Measurement value acquisition unit, 50 Information terminal, 60 Driving analysis server, 80 Measurement value acquisition unit, 84 Model storage unit, 86 Prediction unit, 88 Decision unit, 99 Output unit, 100 Driving analysis system.

Claims

1. A model memory unit holds a predictive model that predicts pace changes in the latter half of a race using machine learning with motion history data measured over time from the beginning to the end of a race, which includes features that represent the motion state of runners in completed marathons, and features corresponding to multiple types of motion analysis indicators, including at least running pace. A measurement value acquisition unit that acquires motion history data by measuring feature quantities corresponding to the multiple types of motion analysis indicators over time during a run by a user who is a marathon runner, which is less than the distance of the marathon race, A prediction unit predicts the pace progression of the user in the latter half of the race based on the operation history data measured over time during the user's run and the prediction model. A determination unit that determines the content of the notification to be sent to the user based on the prediction result from the prediction unit, An output unit that outputs notification content based on the decision made by the aforementioned decision unit, A driving analysis system characterized by having the following features.

2. The measurement acquisition unit acquires motion history data by measuring feature quantities corresponding to the multiple types of motion analysis indicators for runners in a marathon in progress over time until partway through the race. The running analysis system according to claim 1, characterized in that the prediction unit predicts the pace progression from a measurement point during the race to the end of the race based on the operation history data measured over time up to the middle of the race and the prediction model.

3. The running analysis system according to claim 1 or 2, characterized in that the aforementioned multiple types of motion analysis indicators further include at least one of pitch and stride.

4. The driving analysis system according to claim 1 or 2, characterized in that the aforementioned multiple types of motion analysis indicators further include at least one of multiple types of motion analysis indicators that contribute to driving efficiency, including deceleration and vertical movement.

5. The aforementioned prediction model is a machine learning-based regression model that uses features corresponding to multiple types of motion analysis indicators included in the motion history data as explanatory variables, and the rate of deceleration of the average pace in the second section of the race relative to the average pace in the first section of the race in the first half as the dependent variable. The running analysis system according to claim 1 or 2, characterized in that the prediction unit predicts the rate of deceleration of the user's pace in the latter half of the race as the pace progression.

6. The aforementioned prediction model is a classification model developed through machine learning, with explanatory variables being features corresponding to multiple types of motion analysis indicators included in the motion history data, and the dependent variable being whether or not a deceleration of a predetermined percentage or more occurred in the second section of the race relative to the average pace of the first section in the first half of the race. The driving analysis system according to claim 1 or 2, characterized in that the prediction unit predicts whether or not a deceleration of a predetermined percentage or more will occur in the latter half of the user's race as a pace progression.

7. The driving analysis system according to claim 1 or 2, characterized in that the decision unit determines, as the notification content, content that suggests the possibility of the user slowing down in the latter half of the race.

8. The driving analysis system according to claim 1 or 2, characterized in that the decision unit determines, as the notification content, content that suggests countermeasures including a recommended pace according to the user's possibility of deceleration in the latter half of the race.

9. The running analysis system according to claim 1 or 2, characterized in that the output unit causes a device carried by the runner to output at least one of the following: voice, image, or vibration, in accordance with the notification content.

10. A position measurement unit that acquires the runner's location information, A motion detection unit that detects the runner's movements, The system further comprises a calculation unit that calculates feature quantities corresponding to multiple types of motion analysis indicators based on the position information and motion information, The driving analysis system according to claim 1 or 2, characterized in that the measurement value acquisition unit acquires the feature quantities calculated by the calculation unit.

11. A driving analysis method for a driving analysis system, The processor of the aforementioned driving analysis system A process for acquiring motion history data by measuring over time features corresponding to multiple types of motion analysis indicators, including at least running pace, as features representing the user's motion state during running of a distance less than the marathon race distance, by a user who is a marathon runner, A predictive model that predicts the pace changes in the latter half of a race by machine learning using motion history data measured over time from the beginning to the end of a race, with features corresponding to multiple types of motion analysis indicators targeting runners in completed marathons; and a process that predicts the pace changes in the latter half of a race for the user based on motion history data measured over time during the user's running. A process of determining the content of the notification to be sent to the user based on the results of the prediction, The process of outputting the notification content based on the aforementioned decision, A driving analysis method characterized by comprising the following features.

12. The system has a function to maintain a predictive model that predicts pace changes in the latter half of a race using machine learning with motion history data measured over time from the beginning to the end of the race, which includes features that represent the motion state of runners in completed marathons, and features corresponding to at least multiple types of motion analysis indicators, including running pace. A function to acquire motion history data by measuring feature quantities corresponding to the multiple types of motion analysis indicators over time during runs of a user who is a marathon runner, which are shorter than the distance of the marathon race, A function that predicts the pace progression of the user in the latter half of the race based on the operation history data measured over time during the user's run and the prediction model, A function to determine the content of the notification to be sent to the user based on the results of the prediction, A function to output notification content based on the aforementioned decision, A driving analysis program characterized by enabling a computer to perform this analysis.

Images