WO2012014972A1 - Vehicle behavior analysis device and vehicle behavior analysis program - Google Patents

Abstract

The objective of the present invention is not only to reduce the burden on an operator when determining, for instance, whether or not another vehicle was included as a cause of a rear-end accident case or a risk incident case, but to allow the determination to be made accurately, or to eliminate the need for the determination. In addition, the present invention is provided with a lateral edge extraction unit (202) for extracting a lateral edge X which exists within a reference image frame F_t for a predetermined time; a candidate area generation unit (203) for generating, with the lateral edge X as a reference, candidate areas (W2) which include all or a portion of other vehicles within the reference image frame F_t; and a candidate area tracking unit (204) for tracking the candidate areas (W2) in image frames other than the reference image frame F_t. Furthermore, the candidate area tracking unit (204) uses likelihood calculations to detect horizontal translation and resizing of textures of the candidate areas (W2) in tracking the candidate areas (W2).

Description

Vehicle behavior analysis apparatus and vehicle behavior analysis program

The present invention relates to a vehicle behavior analysis apparatus that analyzes the behavior of a vehicle using a moving image obtained by an in-vehicle camera such as a drive recorder, and a vehicle behavior analysis program used in this apparatus.

In recent years, vehicle-mounted drive recorders have been developed that automatically record images of the outside and inside of your vehicle (automobile) while driving, allowing you to analyze the objective situation in the event of an accident or near-miss event. Has been. For example, in a taxi, this type of drive recorder is installed for accident prevention measures through post-mortem analysis of daily driving, accumulation of knowledge related to accident prevention, objective evidence of the cause of an accident, investigation, etc. There is also a movement to do.

This drive recorder is composed of (1) a forward moving image of the own vehicle (front image), (2) acceleration of the own vehicle (front / rear or left / right direction), (3) vehicle speed pulse, (4) brake signal, and (5) winker signal. (6) A GPS position signal or the like can be acquired. By using all of this information, it becomes possible to analyze the behavior of the host vehicle in the event of an accident, a near-miss, etc. to a certain degree of accuracy.

On the other hand, the relative positional relationship between the host vehicle and the surrounding environment (including a preceding vehicle that approaches quickly) is generally obtained from a front moving image acquired by a camera built in the drive recorder. Then, the relative positional relationship between the host vehicle and the surrounding environment in the event of an accident or near-miss is estimated by the operator analyzing the recorded forward moving image by visual observation.

Specifically, as shown in Non-Patent Document 1, a rear-end collision case, a near-miss case, and the like are extracted from the collected forward moving image group by an operator's visual observation. Then, the operator designates the rear end width of the preceding vehicle shown in each frame of the recorded image, and the accident analysis is performed by estimating the inter-vehicle distance.

However, there is a problem that the task of visually checking each of a large number of collected forward moving images and extracting a rear-end collision case or a near-miss case is extremely complicated and burdensome. In addition, there are various judgment criteria for rear-end collision cases or near-miss cases, depending on the operator, and there is a problem that extraction of the cases and mistakes are likely to occur. For this reason, the extracted forward moving image may not be objectively a rear-end collision case or a near-miss case, and the above accident prevention measures using the forward moving image become inaccurate.

Accordingly, the present invention has been made to solve the above-mentioned problems all at once, and in a moving image obtained from an in-vehicle camera such as a drive recorder, a predetermined behavior that causes, for example, a rear-end collision case or a near-miss case, etc. In order to determine whether or not other vehicles are included, the main objective is not only to reduce the burden on the operator, but also to make the determination accurately or to make the determination unnecessary. It is what.

That is, the vehicle behavior analysis apparatus according to the present invention determines whether or not a moving image obtained by photographing the front or rear of the host vehicle with an in-vehicle camera includes another vehicle exhibiting a predetermined behavior with respect to the host vehicle. A horizontal edge extraction unit for extracting a horizontal edge existing in a reference image frame at a predetermined time constituting the moving image, and the reference image based on the horizontal edge obtained by the horizontal edge extraction unit. A candidate area generating unit that generates a candidate area that includes all or part of the other vehicle in a frame; and a candidate area tracking unit that tracks the candidate area in an image frame other than the reference image frame. The tracking unit uses likelihood calculation and does not detect the rotational movement and shape change of the texture of the candidate area, but detects the parallel movement and size change of the texture of the candidate area. And characterized in that it is intended to track the candidate region.

In such a case, a horizontal edge is likely to occur at the uppermost part (roof), the lowermost part (rear bumper or grounding part), or the central part (lower side of rear glass or upper trunk) of other vehicles. By setting a candidate area and tracking the candidate area, other vehicles can be automatically tracked. As a result, the operator can easily determine whether or not the vehicle includes another vehicle exhibiting a predetermined behavior using the tracking result of the candidate area tracking unit. Therefore, not only can the burden on the operator be reduced, but also the operator's judgment can be made accurately, or the judgment can be made unnecessary.

In addition, the candidate area generated based on the horizontal edge may include an area including only the sky and the road surface. Here, the candidate area tracking unit detects the translation and size change of the texture of the candidate area using likelihood calculation without detecting the rotational movement and the shape change of the texture of the candidate area. It is possible to prevent the candidate area including only the road surface from being tracked. Therefore, it is possible to reduce the calculation for tracking candidate areas including other vehicles. In other words, the candidate area including only the sky and the road surface cannot be kept unique by the texture feature alone, and therefore the candidate area tracking unit of the present invention has a tracking function for tracking the candidate area. No. On the other hand, since other vehicles included in the reference image frame are projected and approach the rear surface or the front surface, a change in the vehicle image can be captured only by parallel movement and size change of the same texture in the reference image frame. In addition, the texture features are complex enough to have uniqueness in the vicinity, and the candidate region tracking unit of the present invention can track the texture features of candidate regions including other vehicles.

As a specific embodiment of the candidate area tracking unit, it is conceivable that the candidate area tracking unit uses a particle filter.

In order to reduce the amount of calculation in the candidate area generation unit and the candidate area tracking unit by eliminating the horizontal edge that occurs at a position where no other vehicle can exist, eliminating the candidate area generated by the horizontal edge, It is preferable that the horizontal edge detection unit determines a horizontal edge to be output to the candidate area generation unit based on a position of the horizontal edge existing in the reference image frame.

In particular, the in-vehicle camera captures the front or rear of the host vehicle, and the position of the other vehicle is determined to some extent with respect to the host vehicle. Therefore, the lateral edge extraction unit performs the preceding or following in the reference image frame. It is desirable to output to the candidate area generation unit a horizontal edge located in a designated area that excludes areas where other vehicles cannot travel, or a lateral edge that has an overlap rate equal to or greater than a predetermined value.

The main cause of horizontal edges is the top (roof), bottom (rear bumper (front bumper) or grounding part), or center (lower side of rear glass (front glass) or upper trunk) of other vehicles. Therefore, in order to generate a candidate area so as to include other vehicles on the basis of the horizontal edge, the candidate area generation unit includes a plurality of rectangular candidate areas in the vertical direction so as to include one horizontal edge. It is desirable to generate. Here, generating the candidate area so as to include the horizontal edge includes not only that the candidate area includes the horizontal edge but also that the horizontal edge becomes the horizontal side of the candidate area.

Although it is conceivable for the operator to determine whether or not other vehicles exhibiting a predetermined behavior are included using the tracking result of the candidate area tracking unit, the determination factor of the operator is reduced as much as possible, In order to increase the objectivity of the determination, another vehicle in which the moving image approaches the host vehicle at a predetermined speed from the amount of movement of the feature points in the candidate region using the tracking result of the candidate region tracking unit It is desirable to provide an approaching vehicle determination unit that determines whether or not a vehicle is included.

For example, in order to be able to automatically extract a rear-end collision case or a near-miss case related to a rear-end collision from a storage unit that stores a plurality of moving image data photographed by an in-vehicle camera, the approaching vehicle determination unit includes: It is desirable to extract a moving image including another vehicle approaching the host vehicle at a predetermined speed from a group of moving images obtained by photographing the front or rear of the host vehicle with the in-vehicle camera.

Further, the vehicle behavior analysis program according to the present invention determines whether or not the moving image obtained by photographing the front or rear of the own vehicle with the in-vehicle camera includes another vehicle exhibiting a predetermined behavior with respect to the own vehicle. A vehicle behavior analysis program for judging, wherein a horizontal edge extraction unit for extracting a horizontal edge existing in a reference image frame at a predetermined time constituting the moving image, and a reference to the horizontal edge obtained by the horizontal edge extraction unit A candidate area generating unit that generates a candidate area that includes all or part of the other vehicle in the reference image frame, and a candidate area tracking unit that tracks the candidate area in an image frame other than the reference image frame. The candidate area tracking unit uses a likelihood calculation, and does not detect the rotational movement and shape change of the texture of the candidate area, To detect the translation and resizing texture auxiliary regions, and characterized in that to track the candidate region.

According to the present invention configured as described above, whether or not the moving image obtained from the in-vehicle camera such as the drive recorder includes another vehicle exhibiting a predetermined behavior that causes a rear-end collision case or a near-miss case, for example. In this determination, not only the burden on the operator can be reduced, but also the operator's determination can be made accurately, or the determination can be made unnecessary.

It is a figure showing typically the vehicle behavior analysis system of this embodiment. It is a figure which shows the component of the drive recorder of the same form. It is a functional block diagram of the vehicle behavior analysis apparatus of the same form. It is a flowchart which shows operation | movement of the vehicle behavior analysis apparatus of the form. It is a schematic diagram which shows the horizontal edge detection of the horizontal edge extraction part of the same form. It is a schematic diagram which shows the horizontal edge extraction which considered the designated area | region of the same form, and the said area | region. It is a schematic diagram which shows an example of the candidate area | region production | generation of the candidate area | region production | generation part of the same form. It is a schematic diagram which shows narrowing down of the candidate area | region of the candidate area | region production | generation part of the same form. It is a schematic diagram which shows the tracking aspect of the candidate area | region tracking part of the same form.

100: Drive recorder (vehicle camera)
200... Vehicle behavior analysis device F _t ... Reference image frame 202... Horizontal edge extraction unit X... Horizontal edge W1. ... Approaching vehicle judgment section

Hereinafter, an embodiment of a vehicle behavior analysis system according to the present invention will be described with reference to the drawings.

<1. System configuration>
As shown in FIG. 1, the vehicle behavior analysis system according to the present embodiment acquires a drive recorder 100 that captures the front of an automobile (own vehicle) V, and moving image data captured by the drive recorder 100, And a vehicle behavior analysis apparatus 200 that determines whether or not the moving image indicated by the moving image data includes another vehicle that exhibits a predetermined behavior with respect to the host vehicle V.

<2. Drive recorder>
The drive recorder 100 is attached to a windshield, and is a vehicle-mounted type that records the behavior of the automobile V1 and the surrounding situation during a certain period before and after an accident or near-miss, etc. In a single casing 2 , The basic components, that is, the detection unit 3, the information processing unit 8, the notification unit 4, the input unit 5, the communication unit 6, the detachable recording unit 7 and the like shown in FIG.

The detection means 3 senses a state relating to the behavior of the vehicle V, the surrounding situation, and the like, and outputs state data indicating the state. Here, three types of imaging means 31, an acceleration sensor 32, and a position sensor 33 are provided. At least. The imaging means 31 is, for example, a CCD camera that images a situation outside the vehicle ahead of the host vehicle and outputs state data (moving image data) indicating the image. The acceleration sensor 32 is configured by using, for example, the piezoresistive effect, and senses the acceleration of 1 to 3 dimensions (for example, front and rear, left and right, and up and down in the case of 3 dimensions) acting on the vehicle, and indicates the acceleration. State data (acceleration data) is output. The position sensor 33 is, for example, a GPS receiver that detects radio waves from a plurality of satellites, senses the position of the automobile V, and outputs state data (position data) indicating the position. In addition, the status data includes vehicle speed data transmitted from the vehicle speed sensor of the vehicle V, door opening / closing data indicating opening / closing of the door, brake data indicating ON / OFF of the brake, and the like. Received via connector CN. The connector CN is also used for a power source.

The notification unit 4 includes an LED 41 that is a light emitter exposed on the surface of the casing 2, an audio output body (not shown) such as a buzzer and a speaker built in the casing 2, and the like.

Here, the input means 5 is a button switch provided on the surface of the casing 2.

Here, the communication means 6 is communication hardware such as a wireless LAN or a mobile phone that is built in the casing 2 and transmits and receives radio waves to and from the base station or a vehicle behavior analysis device 200 described later.

The detachable recording means 7 is, for example, a CF memory card or an SD memory card that is detachably attached to the slot 2b that opens to the side of the casing 2.

The information processing means 8 is a so-called computer circuit having a CPU 81, an internal memory 82 (for example, a non-volatile memory), an I / O buffer circuit (which may include an AD converter) 83, etc. Built in the casing. The CPU 81 operates according to a program stored in a predetermined area of the memory 82 to perform control of each means and information processing.

In brief, the CPU 81 continuously and continuously stores various state data during traveling, that is, acceleration data, position data, moving image data, and the like in a temporary area (hereinafter also referred to as a temporary data storage unit) set in the memory 82. In the event that an event that indirectly indicates the occurrence of a near miss, an accident, an abnormality, or the like occurs, the status data over a certain period before and after that is stored in a normal area in the memory 82 (Hereinafter also referred to as a regular recording data storage unit) for recording.

Examples of the event include a case where the acceleration (deceleration) indicated by the acceleration data exceeds a predetermined reference value, a case where the time continues for a certain period, a case where the door is opened and closed, a case where the vehicle is turned off, etc. It corresponds to it. Here, depending on the event that occurred, data was triggered only when the vehicle speed was higher than the upper limit speed, when the vehicle speed was lower than the acceleration / deceleration, or when some other conditions such as the presence or absence of braking were met. Recording is performed so that useless data is recorded as little as possible.

Also, from the viewpoint of preventing useless data recording, it also has a learning function. In other words, before performing data recording, whether or not a near-miss or an accident has occurred is always notified to the driver by the notification means, and correct / incorrect input from the driver (for example, ON / OFF of the button switch 5) is made. I try to accept it. By repeating this, the driver's driving tendency is grasped to some extent, and for example, the driver's unique phenomenon that indirectly indicates an accident or the like is learned by changing a predetermined reference value of acceleration.

Furthermore, the recorded status data is weighted according to the situation at the time of recording, and is classified according to the importance to be recorded. Then, when the memory capacity becomes full or the like, it is configured such that new status data is recorded after being deleted from the less important ones.

The state data properly recorded in this way is transmitted to an analysis center (not shown) wirelessly at a specific place, or transferred to the removable recording means 7 and extracted, and brought to the analysis center. This is used for subsequent analysis using the vehicle behavior analysis apparatus 200.

<3. Vehicle Behavior Analysis Device>
The vehicle behavior analysis apparatus 200 refers to another vehicle (hereinafter also referred to as an accident target vehicle) that rapidly approaches the host vehicle at a predetermined speed from a group of moving images obtained by the drive recorder 100 mounted on a plurality of automobiles. It extracts moving images that contain them and supports post-mortem analysis. The specific device configuration of the vehicle behavior analysis apparatus 200 is a general purpose or dedicated computer equipped with a CPU, a memory, an input / output interface, an AD converter, etc., and a vehicle behavior analysis stored in a predetermined area of the memory. By cooperating the CPU and peripheral devices according to the program, as shown in FIG. 3, a data receiving unit 201, a moving image data storage unit D1, a horizontal edge extracting unit 202, a candidate region generating unit 203, a candidate region tracking unit The functions of the approaching vehicle determination unit 204, the analysis data storage unit D2, and the like having the functions of

Hereinafter, each part D1, D2, 201 to 204 will be described in detail together with the operation with reference to FIGS.

The data receiving unit 201 receives state data that is moving image data stored in the regular recording data storage unit of the drive recorder 100, and stores the state data in the moving image data storage unit D1 (FIG. 4, step S1). . The data receiving unit 201 may be configured by a receiver that receives status data transmitted by using a wireless LAN or the like by the communication means (transmitter) 6 provided in the drive recorder 100, or provided in the drive recorder 100. For example, the state data may be acquired via a CF card which is the detachable recording means 7 provided.

The moving image data storage unit D1 stores and accumulates state data that is moving image data captured by the drive recorder 100 (FIG. 4, step S2). In the present embodiment, the moving image data storage unit D1 is configured to systematically store a plurality of moving image data obtained by a plurality of automobiles V.

Horizontal edge detection section 202 acquires the moving image data stored in the moving image data storage unit D1, extracts the horizontal edge X existing in the reference image frame F _t of the moving image indicated by the video data (FIG. 4, step S3). Here, the reference image frame _Ft is an image frame at a predetermined time t constituting a moving image. Specifically, for example, an image frame at a time when the driver steps on the brake to generate an acceleration of a predetermined value or more. It is.

Then, as shown in FIG. 5, the horizontal edge extraction unit 202 regards the reference image frame Ft as a luminance image, and, for example, the horizontal edge X in the following three stages ( 1) to (3) (three conditions). To extract.
(1) Edge detection by Canny filter, (2) Horizontal edge extraction by original filter convolution operation, (3) Threshold processing by length / gradient / location after labeling processing. Can be considered.

When the length of the horizontal edge extracted by such a method is l _ID , and the area where the designated front area W1 overlaps the horizontal edge X is l _in , when l _in / l _ID > th _S , the front area W1 It is considered to exist. On the other hand, when l _in / l _ID ≦ th _S , the horizontal edge X does not exist in the front area W1, and is excluded from the horizontal edge X that generates the candidate area W2. Note that th _S is a predetermined constant. Here, the front area (designated area) W1 is an area excluding an area where the accident target vehicle cannot travel, and is symmetrical with respect to the central axis of the image frame as shown in FIG. From the predetermined position to the predetermined position, and the area below the predetermined position gradually expands to form a generally inverted funnel shape. The front region W1 is not limited to the reverse funnel shape, and may have other shapes.

As shown in FIG. 7, the candidate area generation unit 203 acquires the horizontal edge data extracted by the horizontal edge extraction unit 202 based on the above three conditions, and uses the horizontal edge X as a reference in the other image frame Ft. A candidate area W2 including all or part of the vehicle is generated (FIG. 4, step S4). Specifically, the candidate area generation unit 203 generates a plurality of rectangular candidate areas W2 in the vertical direction so as to include one horizontal edge X. Note that the horizontal edge data includes the pixels constituting the horizontal edge X and the label ID assigned by the labeling process.

In detail, the candidate area generation unit 203 creates the following line segment from the horizontal edge X extracted by the horizontal edge extraction unit 202.

Here, 0 ≦ t ₁ ≦ 1. P is a point on the line segment,

It is. Here, 0 ≦ α ≦ 1 is a constant for considering that both ends of the horizontal edge line segment are contracted due to the effect of the original filter applied to the horizontal edge X, and 0 ≦ t ₂ ≦ 1 is a horizontal line segment. It is a constant for taking the y coordinate between {y _min , y _max }.

The candidate area generation unit 203 assumes that the horizontal line segment is the uppermost part (roof), the lowermost part (rear bumper / grounding part), or the central part (lower side of rear glass / upper trunk) of the target vehicle. Generate. Specifically, as shown in FIG. 7, the candidate area generation unit 203 has a square having the above horizontal line segment (horizontal edge X) as the upper side, a square having the lower side, and a line connecting the midpoint of the right side and the midpoint of the left side. generating a square reference image frame F _t having as. That is, three square areas (candidate areas W2) are generated from one horizontal line segment. When each side protrudes from the image plane (reference image frame F _t ), the coordinates corresponding to the edge of the image frame are given.

Next, as shown in FIG. 8, the candidate area generation unit 203 performs a process of screening out the poor candidate condition as the position of the generated candidate area W2. Specifically, the candidate area generation unit 203 excludes candidate areas W2 in which there is a contradiction in projective geometry from a large number of candidate areas W2 generated by the previous processing.

Detail candidate area generation unit 203, when considered the lower side of the candidate area W2 and the ground point, when the ground point is present below the reference image frame F _t (that is, when sufficiently close to the host vehicle V) candidate region W2 If is too small, it is considered that a contradiction has occurred. On the other hand, when the ground point is present above the reference image frame F _t (that is, when sufficiently far into the vehicle V) when the candidate area W2 is too large, regarded as inconsistent is occurring.

Therefore, the evaluation reference value l _Base is given based on the y coordinate y _b of the lower side of the candidate area W2.

Here, height is the height [unit: pixel] of the front image, and the upper left origin in the image plane, the right direction of the x axis is positive, and the lower direction of the y axis is positive. σ _l is a constant that controls a coefficient for reducing the evaluation reference value l _Base when going upward. As described above, the candidate area generation unit 203 of the present embodiment uses the evaluation reference value l _Base and deletes the case where the length of the lower side of each candidate area W2 does not satisfy the following conditions.

Here, β and γ are degrees of freedom of the size of the candidate area W2 to be accepted, and β ≦ γ.

In addition, the candidate area generation unit 203 determines that the upper side y coordinate y _t is too biased downward at the time t (the following equation (6)) or the lower side y coordinate y _b is the time t. If there is too much bias upward (Equation (7) below), a simple contradiction is also inspected.

As described above, the candidate area generation unit 203 generates a plurality of candidate areas W2 in the reference image frame F _{t at} time t, and outputs the candidate area data to the approaching vehicle determination unit 204.

The approaching vehicle determination unit 204 includes an accident target vehicle in the candidate region W2 by tracing back a plurality of image frames constituting the moving image for each candidate region W2 obtained by the candidate region generation unit 203. The reliability indicating whether or not to be used is calculated (FIG. 4, step S5). Here, the image frame traced back in the past, a plurality of image frames back in reference image frame F _t in the past was the starting point. In the present embodiment, there are 40 frames that are continuous in time series, starting from the reference image frame _Ft as the starting point. In addition, the reliability is calculated by inspecting conditions regarding characteristics of the accident target vehicle. In other words, the conditions defined in this embodiment are as follows.

(1) Tracking is possible with a tracking function that does not take into account rotational movement of textures and changes in shape. (2) Sufficient corners are included in candidate area W2. (3) The tracked candidate area W2 becomes smaller each time the time goes back.

Requirement (1) is determined because the vehicle subject to the accident is approached with only the rear and rear surfaces reflected in the moving image, and therefore, the variation of the same texture can be described only by translation and size change in the candidate area W2. In addition, since the texture feature is complex enough to have uniqueness in the vicinity, it can be sufficiently tracked with a weak tracking function. In the present embodiment, there is a possibility that a candidate area W2 including only the sky and the road surface is selected as the candidate area W2, but these candidate areas W2 are not unique in the neighborhood only by texture features, With the weak tracking function, the tracking result is easily broken.

Condition (2) is set in order to confirm whether the candidate area W2 has a sufficiently complex texture. Sufficient corners are observed at the back of the car, but on the other hand, sufficient corners are often not observed on the road surface and the sky area described above.

Requirement (3) is used to determine whether the candidate area W2 obtained at time t is approaching rapidly. In the condition (3), the degree of approach based on the size change history of the candidate area W2 obtained from the tracking result is observed.

In order to inspect for each condition, the approaching vehicle determination unit 204 of the present embodiment performs tracking by the particle filter (used for the conditions (1) and (3)), and detection of the corner in the area by the Harris operator (for the condition (2)). Use) and the calculation of the amount of movement of the corner (use for condition (2)), the reliability is calculated using the result, and the accident target vehicle is detected.

The approaching vehicle determination unit 204 has a function as a candidate region tracking unit, and as shown in FIG. 9, as a result of tracking the candidate region W2 from time t to n frames (for example, n = 40) by the particle filter, The coordinates (x (ti), y (ti)) and scale information (length of one side) s (ti) of the candidate area W2 at an arbitrary time ti are obtained (FIG. 4). Step S5-1). However, FIG. 9 shows only the image frame F _t-1 one frame before the reference image frame F _t and the image frame F _t-2 two frames before. The particle filter uses likelihood calculation, does not detect the rotational movement and shape change of the texture of the candidate area W2, detects the parallel movement and size change of the texture of the candidate area W2, and detects the candidate area. W2 is tracked.

At this time, the approaching vehicle determination unit 204 leaves only the candidate area W2 that satisfies the following two conditions and performs the next evaluation.

Condition (a) is for determining whether the area of the candidate area W2 continues to shrink, and condition (b) is for determining whether the candidate area W2 is moving near the center of the image frame. .

Next, the approaching vehicle determination unit 204 performs corner extraction by the Harris operator at all times, and tracks from the sum of corners, the amount of movement of the corners, or the sum of the amounts of movement of the corners, which are feature points in the candidate area W2. Assess the reliability of Here, “reliability” means whether or not there is a vehicle in the candidate area W2, that is, whether or not the candidate area W2 is tracking another vehicle, or is the other vehicle approaching the host vehicle? It is one index for determining whether or not.

When the reliability calculated for each candidate area W2 is the final output of the approaching vehicle determination unit 204, and the maximum reliability is greater than or equal to the threshold, the approaching vehicle determination unit 204 uses the candidate area W2 as the accident target vehicle area. The accident target vehicle area data indicating the accident target vehicle area is output to output means such as a display (FIG. 4, step S5-3). Further, the approaching vehicle determination unit 204 stores analysis data indicating the determination result in the analysis data storage unit D2.

<Effect of this embodiment>
According to the vehicle behavior analysis system according to the present embodiment configured as described above, the vehicle is laterally moved by the uppermost part (roof), the lowermost part (rear bumper or grounding part), or the central part (lower side of rear glass or upper trunk) of other vehicles. Since the edge X is likely to occur, by setting the candidate area W2 with the horizontal edge X and tracking the candidate area W2, it becomes possible to automatically track other vehicles. As a result, the operator can easily determine whether or not the vehicle includes an other vehicle exhibiting a predetermined behavior, using the tracking result of the vehicle approach determination unit 204 having a function as a candidate area tracking unit. Therefore, not only can the burden on the operator be reduced, but the operator's judgment can be made unnecessary.

In addition, the vehicle approach determination unit 204 having a function as a candidate region tracking unit may detect the parallel movement of the texture of the candidate region W2 using likelihood calculation without detecting the rotational movement and the shape change of the texture of the candidate region W2. Since the size change is detected, it is possible to prevent the candidate area W2 including only the sky or the road surface from being tracked.

<Other modified embodiments>
The present invention is not limited to the above embodiment.

For example, the inter-vehicle distance of the preceding vehicle t may be estimated. At this time, for example, the following procedure is performed using the geometric relationship in the camera model. However, as preconditions, the focal length f (pixel), the mounting height h (cm), the image center y _c (pixel) (the origin of the image plane is the upper left, and the y axis is the downward direction in the positive direction), and the mounting angle θ. (°) (The mounting angle is positive in the depression direction and negative in the elevation direction, and 0 ° is horizontal with the road surface.), Y coordinate y _in (pixel) _in the image coordinates of the contact point (candidate area W2 Is used).

At this time, first, a deviation y _off between the infinity line and the image center y _c is obtained from y _off = f × tan (| θ |). Next, in order to determine whether the ground point is below the infinity line, the y coordinate y _inf of the vanishing point is obtained from the following equation.

However, sign is a sign function, and when y _inf <y _in , the ground point is below the infinity line, and when y _inf > y _in , the ground point is above the infinity line.

And, if the grounding point is below the infinity line, the inter-vehicle distance dist is obtained from the following formula.

In the above-described embodiment, the candidate area generation unit generates three candidate areas W2 based on the horizontal edge X. However, the present invention is not limited to this, and the two candidate areas W2 may be generated in the vertical direction. It is also possible to generate four or more candidate regions W2. By calculating the rear-end collision risk from the inter-vehicle distance and the acceleration of the host vehicle thus obtained, it can be used as data for driver's driving evaluation and driving technique improvement.

Furthermore, the vehicle approach determination unit of the above embodiment may determine the sudden approach of the accident target vehicle based only on the size change history of the candidate area W2.

In addition, in the embodiment, the vehicle approach determination unit has a function as a candidate region tracking unit. However, the vehicle approach determination unit and the candidate region tracking unit are provided separately, and the tracking result of the candidate region tracking unit is provided. It may be configured such that the vehicle approach determination unit determines the rapid approach of the subject object vehicle using.

In addition, the vehicle behavior analysis device does not have the vehicle approach determination unit, and displays the tracking result of the candidate area tracking unit on the screen so that the operator can determine whether or not the target vehicle is a sudden approach. You may comprise.

In addition, in the above embodiment, the drive recorder is provided on the windshield and the front of the host vehicle is imaged to determine the presence or absence of a preceding vehicle that exhibits a predetermined behavior. It may be provided in the rear part of the host vehicle such as rear glass, and the rear side may be imaged to determine the presence or absence of a following vehicle exhibiting a predetermined behavior.

In addition, it goes without saying that the present invention is not limited to the above-described embodiment, and various modifications can be made without departing from the spirit of the present invention.

According to the present invention, in determining whether or not a moving image obtained from an in-vehicle camera such as a drive recorder includes another vehicle that exhibits a predetermined behavior that causes, for example, a rear-end collision case or a near-miss case, the operator In addition to reducing the burden, it is possible to make the determination accurately, or to make the determination unnecessary.

Claims (8)

1. It is determined whether or not the moving image obtained by photographing the front or rear of the host vehicle with the in-vehicle camera includes another vehicle that exhibits a predetermined behavior with respect to the host vehicle.
   A horizontal edge extracting unit that extracts a horizontal edge existing in a reference image frame at a predetermined time constituting the moving image;
   A candidate area generation unit that generates a candidate area including all or part of the other vehicle in the reference image frame with reference to the horizontal edge obtained by the horizontal edge extraction unit;
   A candidate area tracking unit that tracks the candidate area in an image frame other than the reference image frame,
   The candidate area tracking unit uses likelihood calculation, does not detect the rotational movement and shape change of the texture of the candidate area, detects the parallel movement and size change of the texture of the candidate area, A vehicle behavior analysis apparatus that tracks candidate areas.
2. The vehicle behavior analysis device according to claim 1, wherein the lateral edge detection unit determines a lateral edge to be output to the candidate region generation unit based on a position of a lateral edge existing in the reference image frame.
3. In the reference image frame, the horizontal edge extraction unit excludes a region where a preceding or subsequent other vehicle cannot travel, and is located in a specified region or an overlap rate of a predetermined value or more with the specified region. The vehicle behavior analysis apparatus according to claim 1, wherein a lateral edge having a value is output to the candidate area generation unit.
4. The vehicle behavior analysis device according to claim 1, wherein the candidate region generation unit generates a plurality of rectangular candidate regions in the vertical direction so as to include one horizontal edge.
5. An approaching vehicle that uses the tracking result of the candidate region tracking unit to determine whether or not the moving image includes another vehicle that approaches the host vehicle at a predetermined speed from the movement amount of the feature point in the candidate region The vehicle behavior analysis apparatus according to claim 1, further comprising a determination unit.
6. The approaching vehicle determination unit extracts a moving image including another vehicle approaching the host vehicle at a predetermined speed from a group of moving images obtained by photographing the front or rear of the host vehicle with the in-vehicle camera. The vehicle behavior analysis apparatus according to claim 5.
7. The said approaching vehicle judgment part can calculate the inter-vehicle distance of the own vehicle and another vehicle from the extracted moving image, and can calculate the collision risk from the inter-vehicle distance and the acceleration of the own vehicle. Vehicle behavior analysis device.
8. A vehicle behavior analysis program for determining whether or not a moving image obtained by photographing the front or rear of the host vehicle with an in-vehicle camera includes another vehicle exhibiting a predetermined behavior with respect to the host vehicle,
   A horizontal edge extracting unit that extracts a horizontal edge existing in a reference image frame at a predetermined time constituting the moving image;
   A candidate area generation unit that generates a candidate area including all or part of the other vehicle in the reference image frame with reference to the horizontal edge obtained by the horizontal edge extraction unit;
   A computer is provided with a function with a candidate area tracking unit that tracks the candidate area in an image frame other than the reference image frame,
   The candidate area tracking unit uses likelihood calculation, does not detect the rotational movement and shape change of the texture of the candidate area, detects the parallel movement and size change of the texture of the candidate area, A vehicle behavior analysis program for tracking candidate areas.
   

Images