JP2018022220A - Behavior data analysis system, and behavior data analysis device, and behavior data analysis method - Google Patents

Abstract

PROBLEM TO BE SOLVED: To provide a behavior data analysis system that can reduce the throughput of image recognition to reduce processing time.SOLUTION: A behavior data analysis system 1 comprises: a behavior data collection device 3 that can record information on a movement of a movable body 2 driven by a driver as behavior data; and a behavior data analysis device 4 that analyzes the behavior data recorded by the behavior data collection device 3. The behavior data analysis device 4 evaluates the driving performed by the driver from the degree of an event occurring to the movable body 2 that is calculated from the behavior data, and changes a method for analysis processing on the behavior data on the basis of a result of the evaluation.SELECTED DRAWING: Figure 1

Description

  The present invention relates to a behavior data analysis system, a behavior data analysis device, and a behavior data analysis method.

2. Description of the Related Art A behavior data analysis system is known in which an imaging device such as a drive recorder is mounted on a vehicle, and acceleration and position information obtained from the imaging device, image data, and the like are analyzed to evaluate driving.
As one of such behavior data analysis systems, in order to reduce the evaluation effort and continuously perform evaluation, the evaluation reference object is automatically extracted by image recognition, and only the extracted part is passed to the evaluator. A behavior data analysis system to be presented has been proposed (see, for example, “Patent Document 1”).

However, the conventional behavior data analysis system that analyzes image data by image recognition has a problem that the processing time for the image recognition is too large and the processing time becomes extremely long.
An object of the present invention is to provide a behavior data analysis system that solves the above-described problems and can reduce the processing amount of image recognition to shorten the processing time.

  The present invention comprises a behavior data collection device capable of recording movement information of a moving vehicle to be driven as behavior data, and a behavior data analysis device for analyzing behavior data recorded by the behavior data collection device, the behavior data The analysis apparatus evaluates the operation of the moving body from the degree of the event that has occurred in the moving body calculated from the behavior data, and changes the analysis processing method of the behavior data based on the evaluation result. .

  According to the present invention, the processing amount of image recognition can be reduced as compared with the conventional case, and the processing time can be shortened by reducing the load of image processing.

1 is a block diagram of a behavior data analysis system to which an embodiment of the present invention is applied. It is a flowchart of the behavior data analysis method in one Embodiment of this invention. It is the schematic which shows the Japanese traffic light analyzed in one Embodiment of this invention. 1 is a schematic diagram illustrating a US traffic light analyzed in one embodiment of the present invention. FIG. It is the schematic explaining the color of the traffic light analyzed in one Embodiment of this invention. It is a block diagram of the label | marker detection part used for one Embodiment of this invention. It is a flowchart explaining the image processing in one Embodiment of this invention. It is a block diagram of temporary stop detection in one embodiment of the present invention. It is the schematic explaining the temporary stop detection in one Embodiment of this invention. It is the schematic explaining the temporary stop detection in one Embodiment of this invention. It is a block diagram of overspeed detection in one embodiment of the present invention. It is the schematic explaining the overspeed detection in one Embodiment of this invention. It is the schematic explaining the overspeed detection in one Embodiment of this invention. It is a block diagram of the behavior data analysis system used for the 2nd and 3rd embodiment of the present invention. It is a flowchart of the behavior data analysis method in the 2nd Embodiment of this invention. It is a flowchart of the behavior data analysis method in the 3rd Embodiment of this invention.

FIG. 1 shows a block diagram of a behavior data analysis system to which the first embodiment of the present invention can be applied. In FIG. 1, a behavior data analysis system 1 includes an imaging device 3 as a behavior data collection device that is attached to a vehicle 2 as a moving body and records an image from the vehicle 2, and a personal computer (hereinafter referred to as a PC) having a known configuration. 4 and server 5 are provided.
The image pickup device 3 includes a sensor group such as an image pickup element 3a, a GPS 3b, and an acceleration sensor 3c, and can acquire image information, position information, and acceleration information as behavior data from each sensor. Is stored in the storage unit 3d. The imaging device 3 may have a function of recording an image like a drive recorder.

  Information regarding the behavior of the vehicle 2 can also be acquired from the vehicle 2. Examples of information related to the behavior of the vehicle 2 include information such as accelerator opening, brake on / off and strength, steering angle, turn signal on / off, and the like, which are also stored in the storage unit 3d as behavior data. A storage medium 6 such as an SD card can be connected to the storage unit 3d, and various data stored in the storage unit 3d can be recorded in the storage medium 6.

The driver of the vehicle 2 can remove the storage medium 6 connected to the storage unit 3d from the imaging device 3 and take it out of the vehicle 2 when the vehicle 2 finishes traveling and gets off the vehicle 2. Data stored in the storage medium 6 is taken into the PC 4 via a reader 7 such as an SD card reader connected to the PC 4.
The PC 4 functioning as a behavior data analysis device is equipped with data analysis software as a processing unit 4a. The processing unit 4a performs various processes including image recognition to analyze the risk of driving the vehicle 2 by the driver. To do. The processed result is stored in the storage unit 4b, and the analysis result is output as a report via the output unit 4c at a predetermined timing.

A specific processing operation of the processing unit 4a is shown in FIG. The processing unit 4a first evaluates the driving risk based on the position data and acceleration data sent from the GPS 3b and the acceleration sensor 3c (ST01). The specific evaluation contents here include, for example, measuring the acceleration in the front-rear direction and determining a sudden braking when a predetermined threshold is exceeded, and measuring the number of sudden brakings. Similarly, it is also possible to measure the number of sudden handles by measuring the acceleration in the left-right direction.
In the evaluation using the position data, the speed is calculated based on the moving distance to evaluate whether the speed is excessive, or in combination with the map database that has the position information of the intersection with the stop sign. And the like.

  The above evaluation results are digitized to determine whether or not the driving risk is greater than or equal to a threshold value (ST02). If it is less than or equal to the threshold value, it is determined that the operation is safe and the process is terminated. If it is equal to or greater than the threshold, it is determined that further detailed evaluation is necessary, and after performing image recognition processing on the image data (ST03), driving evaluation is performed based on the acceleration data, position data, and image data (ST04). ). Specifically, the image recognition includes traffic signal recognition, sign recognition, lane recognition, forward vehicle recognition, and the like. A specific example of image recognition will be described below.

  First, the recognition of a traffic light will be described. The purpose of recognizing traffic lights is to support safe driving by detecting traffic signals and warning the driver and evaluating driving quality. For example, by using the signal detection result and detecting the driver's signal ignorance or sudden start, the driver can be warned and the driving quality can be evaluated.

FIG. 3 shows an example of a traffic signal in Japan, and FIG. 4 shows an example of a traffic signal in the United States.
As shown in FIGS. 3 and 4, the color and shape of the traffic lights are various. There are incandescent light sources and LED light sources in Japan, and the colors of the light sources differ depending on the light source. Specifically, color vividness, pixel color distribution, reflection due to a difference in background color, and the like can be given. Also, the colors of traffic lights in Japan shown in FIG. 3 and the colors of traffic lights in the United States shown in FIG. 4 are different from each other. When detecting a traffic light, it is necessary to cope with various changes in color information.

In addition, the shape of traffic lights is different in Japan and in the United States. The traffic signal in Japan shown in FIG. 3 is basically oval in outer shape, but the traffic signal in the United States shown in FIG. 4 is basically rectangular in shape. In addition, traffic lights in Japan are arranged in the horizontal direction, while signals in the United States are arranged in the vertical direction. The size of traffic lights in the United States is smaller than that of traffic lights in Japan, and the brightness of traffic lights in the United States is brighter than that of traffic lights in Japan.
Also, the color and shape of the traffic light differ depending on changes such as clear and rainy weather, morning and evening. Further, the image formation of the traffic light differs depending on the automatic exposure adjustment and automatic color adjustment (AWB) functions of the camera, which changes the color and shape.
As described above, due to the presence of various colors and shapes of traffic signals, high-accuracy traffic signal recognition has been a difficult task.

As a method of detecting a traffic signal, a traffic signal recognition process is performed using the color and shape characteristics of the traffic signal. As shown in FIGS. 3 (a), (b), (c) and FIGS. 4 (a), (b), (c), the red, yellow and blue traffic lights are circular in shape, and FIG. As shown in (e), (f) and FIG. 4 (d), the shape of the arrow signal is an arrow shape.
Here, a method of recognizing a signal using the entire shape of the traffic light is conceivable, but the shape of the signal is various, the shape changes depending on the angle of photographing, and the size and resolution also change according to the distance. For this reason, it is difficult to recognize the signal shape using only the luminance image, and there is a problem that false recognition increases when it overlaps the background.
For this reason, a method of recognizing a signal using color and shape information is adopted. Signal signal candidate regions are extracted using the signal color information, and shape recognition is performed on the extracted signal candidate regions.

Here, signal recognition and candidate area extraction will be described.
The signal candidate area of the image input by the imaging device 3 is extracted using the color of the traffic light, but the separation performance differs due to different color exchanges when separating the pixels of the traffic light and the pixels other than the traffic light.
First, the signal image and other images are separated using the RGB values of the image signal input from the in-vehicle camera as they are. Next, the data of the signal image and the other image is converted into the rgb space, and the separation performance between the signal pixel and the other pixel is examined. Here, conversion formulas of r = R / (R + G + B), g = G / (R + G + B), and b = B / (R + G + B) are used.
Further, it is converted into a two-dimensional color space of (R / G, G / B), the signal pixel and other pixels are separated, and finally the RGB pixel is converted into the YUV color space. The separation performance was investigated. Here, the following formula is used for conversion between RGB and YUV.

As described above, the signal pixel and the other pixels can be separated. Here, an example of extracting a traffic signal candidate region using the YUV color space will be described.
An image sample of the input signal is converted from the RGB color space to the YUV color space, and the distribution of the UV space of the red, yellow, and blue signals used in Japanese traffic lights is shown in FIG. In FIG. 5, the vertical axis represents V and the horizontal axis represents U.
As is clear from FIG. 5, the pixel distributions of the blue signal and the red and yellow signals are separated from each other, but the red signal and the yellow signal partially overlap each other.

  Next, pixel extraction of a traffic light will be described. As shown in FIG. 5, in the UV color space, signal pixel candidates are extracted by threshold processing. Therefore, signal image samples in Japan and the United States are collected in advance, and the color distribution of each signal pixel is investigated to determine the color range. A threshold is determined. A signal pixel is extracted by providing a minimum value and a maximum value of U and V, respectively. Umin, Umax, Vmin, and Vmax are set for red, yellow, and blue, respectively, and threshold values are determined so as not to overlap pixels that are not signals. Here, signal pixel candidate regions are extracted by threshold processing in the YUV color space. In the present embodiment, the signal image extraction method in the UV color space is shown, but the signal pixel may be detected in another color space.

Next, signal candidate region extraction by pixel expansion processing will be described. In the threshold processing in the UV color space described above, the extracted signal pixels may be discontinuous. Therefore, the pixel area is expanded by the expansion process of the signal pixels to make the signal pixels continuous. In the present embodiment, vertical and horizontal 7-pixel expansion processing is performed on one pixel of signal pixels.
Next, the area threshold processing of the signal candidate region will be described. A background object having the same color as the traffic light, such as a red signboard or a blue signboard, is detected as a signal candidate area. For this reason, by providing the minimum area and the maximum area of the signal pixel as threshold values, it is possible to remove both the large area signboard and the small area noise.

  Next, signal shape recognition will be described. Signal candidate areas are extracted using the color information described above, and traffic signal shape recognition is performed on the extracted signal candidate areas. This is done by circular extraction of red, yellow and blue signals and arrow recognition of the arrow signal.

First, signal circular recognition will be described. Shape recognition of red, yellow and blue signals is performed by circular recognition, which performs circle extraction by Hough transform (Hough transform).
Next, arrow recognition of an arrow signal will be described. This performs the candidate area detection of the arrow signal in the same color as the blue signal detection. The detected target area is detected and recognized by arrow template matching. If the average pixel difference between the target area and the template image is smaller than the threshold value, it is determined that an arrow signal is detected.

Next, the reduction of misrecognition by surrounding information will be described.
As shown in FIG. 3, in the traffic light in Japan, when the red signal is lit, the blue signal on the left side is dark and becomes a low luminance region. Similarly, when the yellow signal is lit, there are low luminance regions on both sides, and when the blue signal is lit, there is a low luminance region on the right side. With such a configuration, a threshold value in the pixel average value of the low luminance region is set around the target signal, and if the pixel average value in the target region is to exceed the threshold value, it is deleted from the recognition result as a false recognition.

  Since the traffic lights in the United States shown in FIG. 4 have each color lamp arranged in the vertical direction, when the red signal lights up, the lower blue signal is dark and becomes a low luminance region. When the yellow signal is lit, there is a low luminance region above and below, and when the blue signal is lit, there is a low luminance region above. With such a configuration, a threshold value in the pixel average value of the low luminance region is set around the target signal, and if the pixel average value in the target region is to exceed the threshold value, it is deleted from the recognition result as a false recognition.

Next, a white line detection algorithm by the PC 4 for lane detection performed inside the imaging device 3 will be described. In this method, an edge is searched from the vehicle 2 in the left-right direction, and a white line, a road edge, or the like is detected. When simplified, the edge is detected and a straight line is detected by Hough transform.
First, the lower end of the screen where the white line on the screen exists is set as the ROI area. Next, in order to reduce noise, a blurred image (an image is blurred by applying a primary IIR-LPF in the horizontal direction from the center of the screen). Then, the first edge having a certain threshold or more to the left of the vehicle center line with respect to the blurred image is searched, and a certain area around the searched point is marked as a road edge or a white line candidate point. .

Next, a road edge is detected from the candidate points by probabilistic Hough transform. If the number of votes for Hough transform is less than a certain value or if the length is less than a certain value, it is determined that a straight line cannot be detected. Also, straight lines found outside the appropriate angle range are filtered by dividing the screen into left and right areas.
Next, LPF (primary IIR-LPF) is applied to the slope and intercept of the found road edge on the time axis for each of the left and right regions. In addition, when the result is largely different from the result of the past frame, it is determined that the reliability is low and is excluded. If a certain number or more is found in successive frames, it is adopted as having high reliability.

Next, the coefficient of the quadratic function is changed with respect to the tip of the road edge, and the voting is performed with the curvature, and the bending method (coefficient) having the highest matching score with the feature point is adopted. The adopted curvature is smoothed by applying a first-order IIR-LPF to the time axis with l and r.
Finally, the detected coordinate fluctuation at the lower end of the road is checked. When the vehicle speed exceeds a certain speed and the coordinates of the lower end of the road change from a stable state, it is determined that the lane has been changed.

  Next, road sign detection processing by the PC 4 will be described. In FIG. 6, the imaging device 3 is usually mounted on the inside of the windshield of the vehicle 2 or the like, and takes an image of the vehicle 2 in the traveling direction forward. The image input unit 8 captures video captured by the imaging device 3 as a digital image that can be processed. The image processing unit 9 performs full-scale image processing such as correction, conversion, and filter necessary for performing recognition processing described later. The attention area extraction unit 10 sets an area for searching for a marker to be recognized from the entire image.

The template image holding unit 11 reads and holds a predetermined target sign image group collectively as a template image. The matching processing unit 12 performs matching using the template image from the above-described attention area image. As a matching method, various methods such as a luminance value comparison for each pixel and a matching using a shape feature can be considered.
The road sign determination unit 13 evaluates the matching result using a preset reference value, and performs detection determination of the road sign from the obtained evaluation result. Finally, the result output unit 14 outputs a road sign detection result as a determination result. Although the output destination varies depending on the system, for example, the detection result is displayed on the display.

The detection processing operation will be described with reference to FIG. First, a template image is read as preprocessing (ST11). This is performed in an initialization process before the start of image input from the imaging device 3, and thereafter, the read image is held in the template image holding unit 11. The template image to be read is an image corresponding to the marker to be detected.
Next, the captured image of the imaging device 3 is input to the image input unit 9. Since images are output from the imaging device 3 at a constant interval (frame rate), input processing is performed each time (ST12). The input image is subjected to image processing suitable for later recognition processing such as distortion correction, image processing such as filtering, and grayscale conversion in the image processing section 9 (ST13).

  When the region where the detection target sign appears in the field of view of the image can be narrowed down in advance, the image is extracted as a region of interest (ROI) and necessary image processing such as gray conversion is performed (ST14). Next, matching processing between the attention area image and the template image is performed by the matching processing unit 12 (ST15). In step ST11, the template images to be used are matched in order with all of the marker images to be detected held in the template image holding unit 11. The term “matching” here refers to comparing the features of the template image and the inspection image (ROI). In this embodiment, when normalized cross-correlation is used, a correlation value is calculated by the following equation (ST16). Further, the template image used here is an image including the entire sign part.

In the matching result, the road sign determination unit 13 performs detection determination based on whether or not the highest correlation value exceeds a preset reference value (ST17). If there is no template image exceeding the correlation value, no result is detected, and if there is one template image exceeding the correlation value, the result output unit 14 outputs that the label of the template image has been detected (ST18). When there are a plurality of template images exceeding the reference value, processing such as selecting the template with the maximum value is performed. After the determination, the process returns to step ST12 in order to proceed to the next image inspection.
Among the configurations described above, the sign detection unit 15 is configured by the image input unit 8, the image processing unit 9, the attention area extraction unit 10, the template image holding unit 11, the matching processing unit 12, the road sign determination unit 13, and the result output unit 14. Has been.
The PC 4 uses the image recognition result based on the processing operation described above to detect temporary stop, detect lane changes, and ignore red signals.

  Next, temporary disregard detection will be described. FIG. 8 shows a block diagram used for the process of detecting suspension disregard. The sign detection unit 15 and the pause detection unit 16 described above are used for this temporary stop detection. The temporary stop detector 16 receives speed data and input image time information sent from the vehicle 2. The input image time information is time data when an image is input by the imaging device 3. The pause detection unit 16 detects an event that ignores pause based on the speed data and the input image time information.

FIG. 9 shows a case where the vehicle 2 stops after detection of the stop sign, and FIG. 10 shows a case where the vehicle 2 does not stop ignoring the stop sign.
The vehicle stop check period is the period set by the parameter “stop search time” from the detection of the stop sign, and it is determined that the vehicle has stopped if the vehicle speed falls below the parameter “deemed stop speed” during this period. To do. Similarly, it is determined that the vehicle has not stopped when it is not less than the stop speed within the stop search time, and the temporary stop detection unit 16 outputs an event detection flag for ignoring the temporary stop.
The event detection flag sent from the pause detection unit 16 is input to the PC 4 together with the sign detection flag sent from the sign detection unit 15. The PC 4 receiving each flag determines whether or not the vehicle 2 has been temporarily ignored.

  Next, red signal neglect will be described. This detects a red signal ignoring event using the result of recognizing the red signal. When a red signal is detected by the imaging device 3, it is detected that the signal is ignored when the vehicle 2 passes through the red signal road region at a certain speed or higher. At this time, since the vehicle 2 approaches the red signal and crosses the stop line, the size of the detected signal region is larger than a certain size threshold.

Next, lane change detection will be described. When both white lines are effective at speeds of 40 km / h or more (excluded because there is a case where the vehicle crosses the diagonal line in order to pass on a narrow road at low speed), the white line detected at the lower end When the difference from the previous frame of the X coordinate is equal to or higher than the threshold after LPF (low pass filter) is applied, it is detected that the lane has been changed. If it is less than 40km / h or white line is invalid, wait for lane stability.
Depending on how long the lane change was detected by the above-mentioned method and the lane change was made again after the previous lane change, for example, event flag value 0: no event (basic), lane change within 1: 9 seconds, 2: Set a condition flag such as lane change within 6 seconds, lane change within 3: 3 seconds. This flag is transferred to the server 5 to detect dangerous driving that frequently repeats lane changes.

  Next, the overspeed detection will be described. FIG. 11 shows a block diagram used for the overspeed detection process. The above-described sign detection unit 15 and overspeed detection unit 17 are used for this temporary stop ignore detection. The overspeed detection unit 17 receives speed data and input image time information sent from the vehicle 2 and a sign detection flag sent from the sign detection unit 15. The overspeed detection unit 17 detects the event of ignoring the pause based on the speed data and the input image time information.

FIG. 12 shows a case where the vehicle 2 has not exceeded the speed after detecting the speed sign, and FIG. 13 shows a case where the vehicle 2 has exceeded the speed after detecting the speed sign.
The period to check for overspeed is the period set by the parameter “speed measurement time” from the detection of the speed limit sign, and if the vehicle speed exceeds the parameter “limit speed” during this period, to decide. Similarly, if the speed limit is not exceeded within the speed measurement time, it is determined that the speed has not exceeded. The speed excess detection unit 17 outputs an event detection flag with a speed excess when it is determined that the speed has been exceeded.
The event detection flag sent from the overspeed detection unit 17 is input to the PC 4 together with the detection limit speed flag sent from the sign detection unit 15. The PC 4 receiving each flag determines whether the vehicle 2 has exceeded the speed.

  As described above, in the first embodiment of the present invention, as shown in FIG. 2, first, driving evaluation is performed based on acceleration data and position data, and then image recognition processing is performed when the driving risk is equal to or higher than a threshold value. For further driving evaluation. In this way, by performing driving evaluation in two stages, behavior data that can reduce image recognition with a high processing load as much as possible, and can reduce the processing amount of image recognition and shorten the processing time. An analysis system can be provided.

  In the first embodiment, image data, position data, and acceleration data are transferred to the PC 4 by the storage medium 6 such as an SD card. However, as shown in FIG. 14, portable communication means such as 3G or LTE is used. A case of transferring directly to the server 5 is also conceivable. In this case, the server 5 includes a processing unit 5a including a CPU, ROM, RAM, HDD, external I / O, and the like, and functions as a behavior data analysis apparatus. In this configuration, since a communication fee is generated according to the amount of data to be communicated, it is necessary to reduce the amount of data to be communicated as much as possible. The processing operation in this configuration will be described below as a second embodiment with reference to the flowchart shown in FIG.

  First, a request for acceleration data and position data is transmitted from the server 5 side to the imaging device 3 side (ST21), and the imaging device 3 transmits data to the server 5 in response to the request (ST22). Here, the acceleration data and the position data may be constantly transmitted because the data amount is small. Thereafter, acceleration data and position data are analyzed on the server 5 side to evaluate the driving risk (ST23). If the degree of risk is less than or equal to the threshold value, it is determined that the vehicle is safe driving, and the process ends.

If the degree of risk is greater than or equal to the threshold value in step ST23, the server 5 side sends an image data transmission request to the imaging device 3 side (ST24), and the imaging device 3 transmits the image data to the server 5 in response to the request. (ST25).
When analyzing acceleration data and position data, extract not only driving evaluation but also dangerous scenes that you want to use later for image recognition such as sudden braking, sudden steering, excessive speed, etc. Thus, it is possible to designate the total data time required for the image data transmission request. Thereby, it is possible to reduce the capacity of the image data by extracting only particularly necessary scenes. In this manner, image recognition processing is performed on the image received on the server 5 side (ST26), and driving evaluation based on acceleration data and position data is performed (ST27).
With this configuration, the amount of image recognition processing can be reduced as compared to the conventional case, and the processing time can be shortened by reducing the load of image processing.

  In the second embodiment, the degree of driving risk is evaluated and the image recognition process is not performed when the evaluation value is equal to or less than the threshold value. However, in the case of a driver with a low evaluation value (performing dangerous driving) A configuration may be adopted in which image recognition processing is performed for all driving and the image recognition processing is performed only when sudden braking is applied to a driver with a high evaluation value (performing safe driving). This is because it has been experimentally found that a driver who has caused an accident in the past (has performed dangerous driving) has a large number of sudden brakings. The image processing method is changed based on the number of times and how to do it. Hereinafter, this processing operation will be described as a third embodiment.

The driving evaluation in the third embodiment is performed based on the number of sudden brakings or the manner of sudden braking. Sudden braking is likely to occur in the case of a near-miss accident just before an accident and is generally correlated with the accident rate. . That is, it can be said that a driver with a lot of sudden braking has a tendency of dangerous driving. Therefore, in the present embodiment, the number of sudden brakings per predetermined time or predetermined distance (for example, how many sudden brakings are applied per hour) and the timing of sudden braking with respect to the traveling distance or traveling time (for example, as compared with a 0-50 km section) Driving evaluation is carried out using a method in which the number of sudden brakings in the 50 to 100 km section is a type in which the concentration decreases due to fatigue.
Detection of sudden braking includes a method of examining acceleration from a change in speed with a speed sensor, a method of examining acceleration from a change in GPS position information, a method of examining acceleration using an acceleration sensor, and the like. In this embodiment, the acceleration sensor 3c is used. The case where acceleration is detected will be described.

The processing operation in the third embodiment will be described based on the flowchart shown in FIG.
First, the server 5 acquires acceleration data, position data, and image data collected by the imaging device 3 (ST31). Each data also records the time of recording.
Next, the server 5 calculates the number of times of deceleration (the number of times of sudden braking) at an acceleration of 0.3 G (1G = 9.8 m / s ² ) or more from the acceleration data (ST32), and a predetermined time (for example, from the recording time) The number of sudden brakings is calculated every 1 hour (ST33). In this embodiment, sudden braking is decelerated by acceleration of 0.3 G or more, but this value can be set arbitrarily.

Next, it is determined whether or not the calculated number of times of sudden braking exceeds a predetermined number of times set in advance (ST34), and if it exceeds the predetermined number of times, all recorded image data is analyzed. (ST35). If the predetermined number of times has not been exceeded, only the image data around the time when sudden braking is recorded is analyzed among the recorded image data (ST36).
With the above-described configuration, it is possible to determine the risk level of the driver under conditions that are finer than those of the second embodiment, and by appropriately processing the image data for analysis corresponding to various risk levels each time. The driving evaluation according to the driver can be performed in a short time.

In the third embodiment, the rate of sudden braking depends on the terrain. For example, sudden braking is unlikely to occur in areas where land is wide and roads are wide, and sudden braking is likely to occur in areas such as urban areas where roads are narrow and people are likely to jump out.
Therefore, using the map data recorded in advance in the server 5 or the imaging device 3 and the acquired position data, a point where sudden braking has occurred is calculated, and it is determined whether or not the point is a region where frequent sudden braking occurs. In the case of frequent braking, the number of sudden brakings may be multiplied by a predetermined coefficient. For example, when driving on a one-lane road in an urban area, the number of sudden braking is counted by multiplying by a factor of 0.5, and when traveling on a two-lane road that is not an urban area, the number of sudden braking is counted as 1. The processing method may be changed depending on whether or not the final number of sudden braking exceeds a predetermined value. With this configuration, it is possible to perform driving evaluation in consideration of the sudden braking occurrence rate that varies depending on regional characteristics.

  In 3rd Embodiment, it is good also as a structure which changes the predetermined frequency shown to step ST34 of FIG. 16 using a positional information. For example, when the percentage of traveling on one lane road accounts for 80%, the predetermined number of times is reduced, and when the percentage of traveling on two lane roads accounts for 80%, the predetermined number of times is increased. And Thereby, the driving | running evaluation according to a more detailed driving | running condition can be performed.

The driving recorder used as the imaging device 3 can be overwritten to enable long-term recording, and is not overwritten by a collision or when the driver presses a stop switch. Therefore, when the number of sudden brakings recorded in the driving recorder is larger than a predetermined value, the image data is transmitted to the server 5, and when the number of sudden brakings is smaller than the predetermined value, the image data is transmitted to the server 5. It is good also as a structure overwritten without.
Further, the image data may be stored as it is when the number of sudden braking is large, and the image data may be compressed and stored when the number of sudden braking is small.

In the third embodiment, the number of sudden brakings in a predetermined time is measured as the degree of the event that has occurred in the vehicle 2, and the driving evaluation method that is the analysis process is changed based on this. Others may be used. Examples of the degree of the event include the number of sudden handles in a predetermined time, the number of lane changes in a predetermined time, and the like.
Also, if the number of sudden braking is greater than the predetermined value, analyze whether white line is ignored by white line detection, analyze whether signal is ignored, and analyze whether to ignore temporary stop. If the number is too small, only the analysis of whether or not the pause is ignored may be performed. As a result, the processing load can be reduced by excluding the analysis of data that need not be analyzed with different types of analysis.

  In each of the first to third embodiments described above, the behavior data analysis system, the behavior data analysis device, and the behavior data analysis method when the driver drives the vehicle 2 that is a moving body have been described. The possible range is not limited to a moving body driven by a driver. The mobile body to which the present invention is applicable may be, for example, a mobile body that is operated by remote operation, or a mobile body that is driven by machine control by an information processing apparatus or the like, that is, a mobile body that is driven automatically.

  The preferred embodiments of the present invention have been described above. However, the present invention is not limited to such specific embodiments, and unless specifically limited by the above description, the present invention described in the claims is not limited. Various modifications and changes are possible within the scope of the gist. The effects described in the embodiments of the present invention are merely examples of the most preferable effects resulting from the present invention, and the effects of the present invention are limited to those described in the embodiments of the present invention. is not.

1 Behavior data analysis system 2 Mobile body (vehicle)
3 Behavior data collection device (imaging device)
4 Behavior data analyzer (PC)
5 Behavior data analyzer (server)

JP 2013-206031 A

Claims (7)

A behavior data collection device capable of recording movement information of a moving vehicle to be driven as behavior data;
A behavior data analysis device for analyzing behavior data recorded by the behavior data collection device;
The behavior data analysis device evaluates the operation of the moving body from the degree of the event generated in the moving body calculated from the behavior data, and changes the behavior data analysis processing method based on the evaluation result Analysis system. The behavior data analysis system according to claim 1,
A behavior data analysis system, wherein the behavior data collection device includes an imaging device. In the behavior data analysis system according to claim 1 or 2,
The behavior data analysis system, wherein the event is sudden braking, and the behavior data analysis device changes an analysis processing method of the behavior data based on the number of sudden brakings within a predetermined time. In the behavior data analysis system according to claim 1 or 2,
The behavior data analysis system, wherein the event is a sudden handle, and the behavior data analysis device changes the analysis processing method of the behavior data based on the number of sudden brakings within a predetermined time. In the behavior data analysis system according to claim 3 or 4,
The behavior data analysis system characterized in that the behavior data includes position data of the moving body, and the behavior data analysis device changes an analysis processing method of the behavior data based on the position data.   An event occurring in the moving body is calculated from behavior data that is movement information of the moving body to be driven, and the driving of the moving body is evaluated from the calculated degree of the event, and the behavior data is calculated based on the evaluation result. Behavior data analysis device that changes the analysis processing method.   A first step of recording movement information of the driven moving body as behavior data; a second step of evaluating the driving of the moving body based on the degree of an event that has occurred in the moving body based on the behavior data; And a third step of determining whether or not the evaluation is equal to or greater than a predetermined threshold. If the evaluation is equal to or greater than the predetermined threshold, the image recognition process is performed and then the operation is performed again including the image data. A behavior data analysis method for performing the fourth step to be evaluated.

Images