WO2022130709A1

WO2022130709A1 - Object identification device and object identification method

Info

Publication number: WO2022130709A1
Application number: PCT/JP2021/033451
Authority: WO
Inventors: 都堀田; 郭介牛場
Original assignee: 日立Astemo株式会社
Priority date: 2020-12-16
Filing date: 2021-09-13
Publication date: 2022-06-23
Also published as: DE112021005035T5; JP2022095079A

Abstract

Provided is an object identification device for, regardless of a host vehicle speed or relative distance, extracting a person's walking characteristics and identifying a pedestrian. This object identification device identifies a type for a moving object on the basis of image data captured by a camera, the object identification device being provided with: an object region cutout unit for cutting out, from the image data, an object region where the moving object is present; a characteristic amount calculation unit for calculating a characteristic amount of the object region; a cyclic fluctuation detection unit for detecting a cyclic fluctuation in the characteristic amount; and a moving object assessment unit for assessing a pedestrian on the basis of the detected cyclic fluctuation.

Description

Object identification device and object identification method

The present invention relates to an object identification device that identifies an object type from an image taken by a camera, and an object identification method.

In recent years, automobiles are required to be equipped with a function to identify the type of moving object in front of the vehicle in order to support driving and realize automatic driving. For example, in order to realize driving support that avoids a collision with a moving object in front, it is necessary to identify the type of the moving object and then perform brake control in consideration of the general moving speed of the moving object. ..

Here, as a method for identifying the type of a moving object captured in an image taken by a camera, a method using a classifier created by learning a large number of images of a specific object is generally used. However, with this identification method, it is difficult to accurately identify moving objects with similar shapes in the captured image, such as pedestrians and bicycles, because the apparent feature pattern on the image is learned. ..

Therefore, Patent Document 1 proposes a method for identifying a pedestrian using the standard deviation of the area of an area presumed to be human. For example, in claim 1 of Patent Document 1, "in a pedestrian detection device for a vehicle that detects a pedestrian from an image obtained by an image pickup means that images the front of the vehicle, a region estimated to be a human from the image is defined. A statistic showing the variation is calculated from the feature amount calculation means that cuts out and calculates the area of the cut out human estimation area as the feature amount, and the time-series data of the feature amount, and the calculated statistic is the judgment threshold. When it is larger than the above, the image corresponding to the human estimation region is provided with a pedestrian determining means for determining that the image is a pedestrian walking in a direction substantially perpendicular to the traveling direction of the vehicle, and the pedestrian determination is provided. The means is a pedestrian detection device characterized in that a moving average value of a fluctuation amount of the feature amount is calculated, and a dispersion or a standard deviation of the moving average value of a predetermined number of samples is calculated as a statistic indicating the variation. " Is disclosed.

Japanese Patent No. 4267171

According to the pedestrian detection device of Patent Document 1, a method of averaging the area of a human area calculated from an image in time series and determining that the person is a pedestrian when the standard deviation of the moving average is large is disclosed. .. However, the area of the pedestrian on the screen not only fluctuates in time series due to the walking behavior of the pedestrian itself, but also fluctuates when the vehicle equipped with the camera approaches the pedestrian.

In order to reduce the influence of such a change in distance to a moving object due to vehicle travel, a threshold value that dynamically fluctuates depending on the vehicle speed is set, and the current speed is obtained from the vehicle speed sensor. It was necessary to determine the type of moving object after setting the value. However, in order to set an accurate threshold value, it is necessary to use a highly accurate vehicle speed sensor, but since the detection accuracy at low speed is low with a general vehicle speed sensor mounted on a vehicle, an appropriate threshold is used at low speed. There was a problem that the value could not be set, and as a result, accurate pedestrian determination could not be performed at low speeds.

Therefore, an object of the present invention is to provide an object identification device and an object identification method capable of accurately determining the type of a moving object in front even when the own vehicle is traveling at a low speed.

The object identification device of the present invention for solving the above problems identifies the type of a moving object based on the image data taken by the camera, and cuts out an object region in which the moving object exists from the image data. A region cutting unit, a feature amount calculation unit that calculates the feature amount of the object region, a periodic fluctuation detection unit that detects the periodic fluctuation of the feature amount, and a moving object determination that determines a pedestrian based on the detected periodic fluctuation. It was equipped with a department.

According to the object identification device and the object identification method of the present invention, it is possible to accurately identify the type of a moving object in front even when the own vehicle is traveling at a low speed.

The block diagram which shows the structure of Example 1 of the object detection apparatus. The figure for demonstrating the processing contents of the object area cutout part and the feature amount calculation part. The figure which showed the walking behavior of a pedestrian in chronological order. The figure which showed the variation of the aspect ratio calculated from each object area of FIG. 3A. FIG. 3B is a diagram showing the relationship between the aspect ratio and the difference between the moving averages. Analysis result of FIG. 4A by the periodic fluctuation detection unit. The figure which showed the increase / decrease of the aspect ratio of FIG. 3B. Analysis result of FIG. 5A by the periodic fluctuation detection unit. The figure which shows the flow of the process of Example 1 of the object identification apparatus. The block diagram which shows the structure of Example 2 of the object identification apparatus. The figure which showed the running behavior of a bicycle. The figure which showed the running behavior of a standing bicycle.

Hereinafter, examples of the object detection device of the present invention will be described with reference to drawings and the like. It should be noted that the following description shows specific examples of the contents of the present invention, and the present invention is not limited to these descriptions, and is by those skilled in the art within the scope of the technical idea disclosed in the present specification. Various changes and modifications are possible. Further, in all the drawings for explaining the present invention, those having the same function may be designated by the same reference numerals, and the repeated description thereof may be omitted.

First, the object identification device 1 according to the first embodiment of the present invention will be described with reference to FIGS. 1 to 6.

FIG. 1 is a functional block diagram showing a schematic configuration of the object identification device 1, the stereo camera 2, and the vehicle control device 3 of the first embodiment mounted on the own vehicle.

The stereo camera 2 is an external sensor that captures a stereo image in front of the own vehicle using the image sensors of the left camera 2L and the right camera 2R and outputs it as image data D. The external sensor used in the present invention may be any sensor that can detect the distance to each object in the image data D, and the monocular camera is equipped with a ranging sensor such as a laser radar, a millimeter wave radar, or an ultrasonic sensor. It may be a combined external sensor.

The object identification device 1 identifies the type of a moving object around the own vehicle based on the stereo image (image data D) taken by the stereo camera 2, and the identification result is the identification result of the vehicle control device 3 via the vehicle-mounted network CAN. It is a device that outputs to. The types of moving objects identified by the object identification device 1 are other vehicles, motorcycles, bicycles, pedestrians, etc., but in this embodiment, a method for identifying pedestrians will be described, and in Example 2, pedestrians and bicycles will be described. The identification method of is explained.

The vehicle control device 3 is an ECU (Electronic Control Unit) that controls the acceleration system, braking system, steering system, etc. of the own vehicle, and collides with a moving object, etc., based on the identification result of the moving object by the object identification device 1. Appropriately control the braking and steering of the own vehicle so that Since a well-known technique can be used for controlling the own vehicle according to the identification result, detailed description of the vehicle control device 3 will be omitted below.

<Detailed structure of object identification device 1>
As shown in FIG. 1, the object identification device 1 includes an internal bus 10, an object area cutting unit 11, a feature amount calculation unit 12, a periodic fluctuation detection unit 13, a pedestrian determination unit 14, a camera interface IF ₁ , and a CAN interface IF. It has ₂ . Specifically, the object identification device 1 is a computer equipped with an arithmetic unit such as a microcomputer, a storage device such as a semiconductor memory, and hardware such as a communication device. Then, the arithmetic unit executes the program loaded in the storage device to realize each function of the object area cutting unit 11 and the like. In the following, the details of each unit will be omitted while appropriately omitting such a well-known technique. Will be described sequentially.

The camera interface IF ₁ acquires a stereo image (image data D) from the stereo camera 2. The acquired image data D is stored in a storage unit (not shown) through the internal bus 10.

The object area cutting unit 11 cuts out the object area R from the image data D stored in the storage unit. This is a process of extracting a region in which the same object is reflected from the image data D, and various methods can be used. For example, in a stereo image, the distance of each point on the image data D can be calculated from the disparity between the images, so that the image regions that are close to each other on the image data D and have a similar relative distance to the own vehicle are grouped. The object region R can be extracted by pinging.

FIG. 2 is an example of image data D obtained by photographing a pedestrian P crossing a pedestrian crossing. When this image data D is used as a processing target, the object area cutting unit 11 detects a rectangular area including the pedestrian P as the object area R. In the following, the width of the object region R on the image data D is X, and the height is Y.

The feature amount calculation unit 12 calculates the feature amount of the object area R obtained by the object area cutting unit 11. The feature amount calculated here is a feature amount that can quantify the movement of the limbs of the pedestrian P. For example, the aspect ratio _{RX / Y} of the object region R that fluctuates with the walking of the pedestrian P, or the object region. The angle of the diagonal line of R and the like. Hereinafter, a case where the aspect ratio R _{X / Y} of the object region R is used as a feature amount will be described.

The aspect ratio R _{X / Y} , which is an example of the feature amount, is a value obtained by dividing the width X of the object region R set in FIG. 2 by the height Y. When the own vehicle approaches the pedestrian P, even if the posture of the pedestrian P is the same, the object region R on the image data D gradually expands, but the actual pedestrian P is the height (height). Therefore, by dividing the width X of the object region R by the height Y, the aspect ratio R _{X / Y} can be obtained as a normalized feature amount that is not affected by the distance or unit from the stereo camera 2. ..

FIG. 3A is an example of a time change of the object region R including the pedestrian P in FIG. Here, the pedestrian P and the object area R from time t to time t + 6 are indicated by P ₀ to P ₆ and R ₀ to R ₆ , respectively. As is clear from this figure, the height Y of the object region R increases as the own vehicle approaches the pedestrian P. Further, the width X of the object region R becomes larger as the own vehicle approaches the pedestrian P, and at the same time, it changes depending on the walking behavior of the pedestrian P moving his / her limbs back and forth. For example, the width X at time t and time t + 4 is narrower than the frames before and after, and the width X at time t + 2 and time t + 6 is wider than the frames before and after.

Therefore, when the feature amount calculation unit 12 calculates the aspect ratio _{RX / Y} for the object region R of the pedestrian P shown in FIG. 3A, the periodic fluctuation as shown in FIG. 3B is detected.

The periodic fluctuation detection unit 13 monitors the feature amount of the object region R calculated by the feature amount calculation unit 12 (for example, the aspect ratio _{RX / Y} of the object region R) in time series, and determines the presence or absence of the periodic fluctuation.

(First cycle fluctuation determination method)
Here, the first periodic fluctuation determination method by the periodic fluctuation detection unit 13 will be described with reference to FIGS. 4A and 4B. FIG. 4A is a graph in which the value of the aspect ratio R _{X / Y} of FIG. 3B and the value of the moving average Av, which is the average value of the aspect ratio R _{X / Y} of the past number frames, are plotted. From this graph, it can be seen that when the aspect ratio R _{X /} _{Y increases or decreases periodically, the aspect ratio R X / Y} periodically increases or decreases with respect to the moving average Av.

Therefore, in order to detect the periodic increase / decrease in the aspect ratio R _{X / Y, the periodic fluctuation detection unit 13 changes the aspect ratio R X / Y} with respect to the moving average Av from the aspect ratio R _X _{/ Y} obtained at each time. The magnitude is calculated as a "periodic fluctuation determination index", and the continuous state of the periodic fluctuation determination index is counted by the "periodic fluctuation determination index continuous counter (+ counter and-counter)".

First, the periodic fluctuation detection unit 13 obtains the value of the moving average Av from the value of the aspect ratio R _{X / Y} of the object region R of the past number frame. Further, it is determined whether the aspect ratio RX _{/ Y} at that time is larger (hereinafter referred to as “+”) or smaller (hereinafter referred to as “−”) than the moving average Av. Furthermore, the number of consecutive positive and negative states with respect to the moving average is obtained. Then, when the + number of consecutive times and the-number of consecutive times are substantially the same, the periodic fluctuation detection unit 13 assumes that the aspect ratio _{RX / Y} changes periodically.

FIG. 4B is an example of the analysis result of the graph of FIG. 4A by the periodic fluctuation detection unit 13. The row of FIG. 4B (a) shows the +/- state of the aspect ratio RX _{/ Y} with respect to the moving average Av of each time, which is the periodic fluctuation determination index of this determination method, and the row of FIG. 4B (b). The + counter value, which is the number of consecutive times of +, is shown, and the-counter value, which is the number of consecutive times of-, is shown in the row of FIG. 4B (c).

In this example, the periodic fluctuation determination index of time t + 1 and time t + 2 is +, the periodic fluctuation determination index of time t + 3 and time t + 4 is −, and the periodic fluctuation determination index of time t + 5 and time t + 6 is +. Therefore, the + counter value in FIG. 4B (b) is counted up once and twice at the times t + 1 and t + 2 in which the + of the periodic fluctuation determination index is continuous, and then the time when the periodic fluctuation determination index becomes −. At t + 3 and t + 4, the current counter value of 2 times is held, and the count-up is restarted from 1 time at the time t + 5 when the periodic fluctuation determination index changes to + again. Similarly, the-counter value is counted up once and twice at times t + 3 and t + 4 in which the periodic fluctuation determination index is continuous, and then at times t + 5 and t + 6 when the periodic fluctuation determination index becomes +. Holds the current counter value of 2 times.

In this way, when the + counter value and the-counter value are counted up, it can be detected that the number of consecutive times of + and the number of consecutive times of-both are 2 times at the time t + 4, and the periodic fluctuation detection unit 13 can detect that they match. , It can be determined that the aspect ratio _{RX / Y} has increased or decreased in a fixed cycle.

(Second cycle fluctuation judgment method)
Next, the second periodic fluctuation determination method by the periodic fluctuation detection unit 13 will be described with reference to FIGS. 5A and 5B. In FIG. 4B, the magnitude of the aspect ratio R _{X / Y} with respect to the moving average Av is used as the periodic fluctuation determination index, but in FIG. 5B, the aspect ratio R _{X / Y of the current frame with respect to the aspect ratio R X / Y} _one frame before is used. Increase / decrease is used as a periodic fluctuation judgment index. Since the increase _/ decrease in the aspect ratio R X / Y before the current frame with respect to the aspect ratio R _{X / Y} one frame before is shown by the broken line arrow in FIG. 5A, the periodic fluctuation detection unit 13 displays the graph in FIG. 5A. The analysis is performed as shown in FIG. 5B. In FIG. 5B as well as in FIG. _4B , at the time t + 4, the number of consecutive + and the number of consecutive − are both 2 times, which coincides with each other. It can be determined that the number has increased or decreased.

When the pedestrian determination unit 14 detects the periodic fluctuation in the periodic fluctuation detection unit 13, the pedestrian determination unit 14 determines that the moving object in the object region R is the pedestrian P.

When the pedestrian determination unit 14 detects the pedestrian P, the CAN interface IF ₂ transmits the determination result to the vehicle control device 3 via the in-vehicle network CAN. As a result, the vehicle control device 3 executes vehicle control on the premise that the own vehicle is close to the pedestrian P.

<Flow chart of processing by the object identification device 1>
The processing flow of the object identification device 1 of the present embodiment described above will be described with reference to the flowchart of FIG.

In step S1, the camera interface IF ₁ acquires image data D from the stereo camera 2.

In step S2, the object area cutting unit 11 cuts out the object area R from the image data D as shown in FIG.

In step S3, the object area cutting unit 11 collates the object area R cut out in step S2 with the object area R cut out at the previous time. In FIGS. 2 and 3A, for the sake of simplicity, a situation in which only one moving object exists in the image data D is illustrated, but in reality, there is a possibility that a plurality of moving objects exist in the image data D. Therefore, it is necessary to identify the object region R of the same object, detect the periodic fluctuation of the feature amount of the same object, and perform the pedestrian determination. Therefore, when the object area cutting unit 11 cuts out the object area R, a process of collating which area of the plurality of areas cut out at the previous time is the same object is performed. When collating objects between consecutive times, the actual distance to the object is due to the fact that the areas overlap on the screen, the area difference is less than a certain threshold value, and the stereo camera 2 is used. If it can be obtained, the difference between the actual distance values is equal to or less than a certain threshold value, and the like, the collation with the object cut out at the previous time is performed. Thereby, for each object region R, it is possible to determine whether or not there is a change due to reference to the past history. Here, it is assumed that n moving objects are detected.

In step S4, the feature amount calculation unit 12 sets the reference counter i of the object area R to 1.

In step S5, the feature amount calculation unit 12 determines whether or not the type flag of the object area R of the reference counter i is set. The type flag is a variable assigned to the object area R, and is set as "unidentified" when the object area R is detected for the first time. If the object type flag of the reference counter i is "unidentified" in step S5, the identification determination is required, so the process proceeds to step S6. If it is "pedestrian" instead of "unidentified", it is determined that there is already a pedestrian in the object area R of the reference counter i. The process proceeds to step S12 for incrementing the count of the counter i.

In step S6, the feature amount calculation unit 12 calculates the feature amount of the object area R of the reference counter i. For example, as shown in FIGS. 3A and 3B, the aspect ratio R _{X / Y} is calculated with respect to the time-series object region R.

In step S7, the periodic fluctuation detection unit 13 calculates the periodic fluctuation determination index of the object region R of the reference counter i. Hereinafter, the case where the periodic fluctuation determination index is +/- of the aspect ratio RX _{/ Y} with respect to the moving average Av shown in FIG. 4B will be described. For example, at time t + 3, since the aspect ratio RX _{/ Y} is smaller than the moving average value Av, the periodic fluctuation detection unit 13 calculates − as a periodic fluctuation determination index of the object region R.

In step S8, the periodic fluctuation detection unit 13 determines whether or not the periodic fluctuation determination index of the object region R of the reference counter i has changed from the previous time. For example, in FIG. 4B, +, which was the periodic fluctuation determination index at time t + 2, has changed to − at time t + 3, so the process proceeds to step S9. As described above, the step S8 is a process of determining the timing at which the sign of the periodic fluctuation determination index of the object region R of the reference counter i is changed. On the other hand, if no change is found in the sign of the periodic fluctuation determination index, the process proceeds to step S12 in which the counter is incremented to the object area R of the next reference counter i because it is not the determination timing.

In step S9, the continuous counters of the object area R of the reference counter i are compared. Specifically, the + counter value and the-counter value are compared, and if they are almost the same, it is considered that there is a periodic fluctuation and the process proceeds to step S10. For example, at time t + 3 in FIG. 4B, since the + counter value is “2” and the −counter value is “1”, the process proceeds to step S11 in which the + counter value or the −counter value is incremented as if they do not match. On the other hand, at time t + 4, the + counter value is "2" and the-counter value is also "2". Therefore, it is determined that the aspect ratio _{RX / Y} has changed periodically, and the process proceeds to step S1. ..

In step S10, since periodic fluctuations were observed in the aspect ratio RX _{/ Y} in the previous step S9, the pedestrian determination unit 14 sets the type flag of the object area R of the reference counter i to “pedestrian”.

On the other hand, in step S11, since the periodic fluctuation determination index of the object region R of the reference counter i does not change, or the + counter and the-counter do not match, the periodic fluctuation cannot be determined yet, so that the determination after the next time is performed. Therefore, the current states of the + counter and the-counter are updated, and the process proceeds to step S12. For example, in the case of time t + 2 in FIG. 4B, since the periodic fluctuation determination index of the current time is "+" following the previous time, the + counter is incremented by 1 as shown in 5a, and the change is changed for-. Since there is no-counter is set to 0 as it is.

In step S12, since the processing for the object area R of the reference counter i is completed, the reference counter i is incremented by 1 in order to set the next object area R as the determination target.

In step S13, it is determined whether or not the reference counter i exceeds the number of objects n set in step S3. If it does not exceed, the unprocessed object area R remains, so the process returns to step S5, and the process for the next object area R is executed. If it exceeds, the processing of the current time is terminated.

Regarding the matching determination between the + counter and the-counter in step S9, the following methods may be adopted for more accurate determination.

(1) In order to avoid erroneous judgment due to noise value fluctuation, the + counter and the-counter are judged to match when each is twice or more.

(2) The match judgment between the + counter and the-counter is not an exact match, and an error of about 50% is allowed. For example, it is assumed that + matches twice in a row and-matches three times in a row.

(3) It is possible to make a more robust judgment by judging when the number of consecutive + and the number of consecutive-matches for the first time, but when the number of consecutive matches is two or more consecutive times. In the examples of FIGS. 4B and 5B, since the second coincidence is detected at the time t + 6, it is determined that the period fluctuation is performed at this time.

With the above configuration, considering that the height (height) of the same pedestrian is constant even if the distance to the moving object changes due to the movement of the own vehicle, the aspect ratio _{RX / Y} is used as a normalized feature amount. By calculating, the walking behavior of the pedestrian P can be accurately detected. Also, when the value of the diagonal angle of the object region R is used as the feature amount, the walking behavior of the pedestrian P is accurately performed based on the normalized index, as in the case of using the aspect ratio _{RX / Y.} The effect of being able to detect is obtained.

According to the object identification device of the present embodiment described above, even when the own vehicle is traveling at a low speed, the normalized feature amount is used to accurately identify that the moving object in front is a pedestrian. It becomes possible.

Next, the object identification device 1 according to the second embodiment of the present invention will be described with reference to FIGS. 7 to 9. It should be noted that the common points with the first embodiment are omitted.

FIG. 7 is a functional block diagram showing the configuration of the object identification device 1 of the second embodiment. The object identification device 1 of FIG. 7 has a configuration in which an object actual height calculation unit 15 is added to the object identification device 1 of FIG. 1 and the pedestrian determination unit 14 is replaced with a pedestrian / bicycle determination unit 14a. Hereinafter, a configuration different from that of the first embodiment will be described.

In this embodiment, an object actual height calculation unit 15 is provided in order to obtain the height of the moving object on the actual coordinates.
The object actual height calculation unit 15 calculates the actual height of the object region R from a stereo image from a stereo camera, for example, and uses the output of a laser radar, a millimeter wave radar, or the like to calculate the actual height of the object region R. The actual height may be calculated. Hereinafter, an example of obtaining the actual height of the object region R using a stereo image from the stereo camera 2 will be described.

The object actual height calculation unit 15 calculates the height in actual coordinates with respect to the object area R cut out by the object area cutting unit 11. Using a stereo image, if the number of pixels of the height Y of the object region R whose distance from the stereo camera 2 is known is known, the actual height of the object region R can be obtained by Equation 1.

Actual height (mm) = (pixel pitch x distance (mm) x height (pixel)) / focal length (mm) ... Equation 1
The periodic fluctuation detection unit 13 calculates the periodic fluctuation determination index in the same manner as in the first embodiment, and updates the + counter and the-counter. In addition, in this embodiment, the history of the actual height acquired by the object actual height calculation unit 15 is acquired, and the average value and the standard deviation in the time series direction of the same object region are calculated.

The pedestrian / bicycle determination unit 14a determines whether or not the periodic fluctuation is detected by the matching determination between the + counter and the-counter, as in the pedestrian determination unit 14 of the first embodiment. Then, when the periodic fluctuation is detected, the standard deviation of the actual height is further referred to, and if the standard deviation is equal to or less than the threshold value, the object region R cut out by the object region cutting portion 11 is determined to be a pedestrian. , If it is larger than the threshold value, it is judged as a bicycle.

The processing by the pedestrian / bicycle determination unit 14a will be described with reference to FIG. FIG. 8 is a diagram showing the running behavior of a bicycle crossing the front of the own vehicle. Here, it is assumed that the actual height of the object region R has already been calculated. In this case, while the height of the person's head does not change, the position of the lower end of the object region R fluctuates up and down due to the up and down of the position of the foot riding the bicycle, and as a result, the height of the object region R periodically changes. fluctuate. That is, since the standard deviation of the height of the object region R on the actual coordinates obtained by the periodic fluctuation detection unit 13 becomes large, it can be determined that the fluctuation of the aspect ratio R _{X / Y} is caused by the fluctuation in the height direction. This is a feature seen when analyzing an image of a bicycle, and when such a feature is found, the pedestrian / bicycle determination unit 14a determines that the moving object in the object region R is a bicycle. Can be done.

Similarly, FIG. 9 is a diagram showing the running behavior of the standing bicycle B. In the case of freezing, the position of the human head moves up and down periodically with the cycle of riding the bicycle, so that the standard deviation of the height on the actual coordinates becomes large as in FIG. Therefore, even when the image data D as shown in FIG. 9 is captured, the pedestrian / bicycle determination unit 14a can determine that the moving object in the object region R is a bicycle.

In Example 1, an example of detecting the walking behavior of a pedestrian by detecting the periodic fluctuation of the aspect ratio R _{X / Y} of the object region R on the image data D has been described. This has the effect that when the object is a pedestrian, the width variation normalized by the height can be detected because the height is invariant.

On the other hand, in Example 2, pedestrians and bicycles can be distinguished by further considering the periodic fluctuation of the actual height of the object region R on the actual coordinates. That is, if the height fluctuation on the actual coordinates is smaller than the threshold value, it is determined to be a periodic fluctuation due to the width fluctuation of the pedestrian, and if it is larger than the threshold value, it is determined to be the periodic fluctuation due to the height fluctuation of the bicycle.

According to the object identification device 1 of the present embodiment described above, even when the own vehicle is traveling at a low speed, the normalized feature amount is used to accurately identify whether the moving object in front is a pedestrian or a bicycle. It becomes possible to do.

1 ... object identification device, 11 ... object area cutting unit, 12 ... feature amount calculation unit, 13 ... periodic fluctuation detection unit, 14 ... pedestrian determination unit, 14a ... pedestrian / bicycle determination unit, 15 ... object actual height calculation Department, IF ₁ ... Camera interface, IF ₂ ... CAN interface, 2 ... Stereo camera, 2L ... Left camera, 2R ... Right camera, 3 ... Vehicle control device, D ... Image data, P ... Pedestrian, B ... Bicycle, R ... Object area, W ... Width of object area, H ... Height of object area, RX _{/ Y} ... Aspect ratio of object area, Av ... Moving average of aspect ratio

Claims

An object identification device that identifies the type of moving object based on the image data taken by the camera.
An object area cutout portion that cuts out an object area in which a moving object exists from the image data,
A feature amount calculation unit that calculates the feature amount of the object region,
A periodic fluctuation detection unit that detects periodic fluctuations of the feature amount, and
A moving object determination unit that determines pedestrians based on the detected periodic fluctuations,
An object identification device characterized by being equipped with.
In the object identification device according to claim 1,
The feature amount calculation unit is an object recognition device characterized in that the aspect ratio of the object area in the image data is calculated as the feature amount of the object area.
In the object identification device according to claim 1,
The feature amount calculation unit is an object recognition device characterized in that the diagonal angle of the object area in the image data is calculated as the feature amount of the object area.
In the object identification device according to claim 1,
The moving object determination unit is characterized in that it determines a moving object in the object region as a pedestrian when the value of the feature amount periodically moves up and down the moving average value of the feature amount. Device.
In the object identification device according to claim 1,
The moving object determination unit determines that a moving object in the object region is a pedestrian when the ratio of the number of consecutive times in which the value of the feature amount exceeds the moving average of the feature amount and the number of consecutive times in which the value does not exceed the moving average of the feature amount is substantially the same. An object recognition device characterized by
In the object identification device according to claim 1,
The moving object determination unit determines that a moving object in the object region is a pedestrian when the value of the feature amount substantially matches the ratio of the number of consecutive times exceeding the previous value of the feature amount and the number of consecutive times not exceeding the previous value of the feature amount. An object recognition device characterized by
In the object identification device according to claim 1,
The moving object determination unit calculates the height on the actual coordinates of the object region from the distance to the moving object and the height on the image data of the moving object, and the periodic fluctuation of the height on the actual coordinates. An object recognition device, characterized in that the moving object is determined to be a bicycle based on the above.
It is an object identification method that identifies the type of moving object based on the image data taken by the camera.
A step of cutting out an object area in which a moving object exists from the image data,
The step of calculating the feature amount of the object region and
The step of detecting the periodic fluctuation of the feature amount and
Steps to determine pedestrians based on detected periodic fluctuations,
An object identification method characterized by being equipped with.