JP2000285245A - Method and device for preventing collision of moving body and recording medium - Google Patents

Abstract

PROBLEM TO BE SOLVED: To prevent a collision by picking up the image of the rear by a camera mounted on a user's vehicle and quickly and reliably detecting another vehicle from this image and making the user to surely take a countermeasure which prevents the trouble related to the collision or abnormal approach between both vehicles on the basis of the detection result. SOLUTION: A TV camera is used to observe the surrounding condition of he user's running vehicle, and another vehicle in the surroundings which may collide with the vehicle is detected, and an alarm is generated to the driver of the running vehicle on the basis of the detection result (steps S1 to S8). In particular, another vehicle which tries to overtake the vehicle is detected with a high precision, and the probability of collision is forecast in consideration of the turn signal, the handle steering angle, etc., of the user's own vehicle to give an alarm.

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a collision prevention apparatus and a collision prevention method for preventing a collision between vehicles when the vehicles experience an instantaneous state in which the vehicles approach each other, such as when one vehicle passes another vehicle. In particular, the present invention relates to a collision prevention apparatus and a collision prevention method in which a camera captures an image of a rear running state, detects another vehicle from the image, and issues an alarm to a vehicle equipped with the camera as necessary.

[0002]

2. Description of the Related Art In general, there are various causes for traffic accidents, and a considerable number of accidents are caused by lane changes or overtaking. Therefore, it is thought that if the driver of a running vehicle always knows the surrounding conditions, especially the driving state behind, it is often useful to avoid careless lane changes and steering operations. .

[0003] Conventionally, radars, television cameras, and the like have been frequently used to observe the surroundings of a running vehicle. As the radar, a laser radar or an ultrasonic radar is used. In the case of an environment recognition device using these radars, the relative speed to an obstacle and the distance to the obstacle can be detected, but the directivity is narrow, and it is difficult to use on a curved road.

On the other hand, in the case of a method using an image obtained from a television camera, as one aspect, it is possible to estimate a three-dimensional position of the vehicle from the position of the vehicle in the image, and to observe the observation range. Also has the feature that it becomes wider.
However, this method only recognizes an instantaneous situation around the vehicle and detects the position of the obstacle in the image,
It did not take into account temporal changes in position with obstacles.

For example, according to the technique described in Japanese Patent Application Laid-Open No. 2-287799, a driving lane is recognized from an image of a television camera mounted on a vehicle, and an obstacle exists in the driving lane. In such a case, a device has been proposed in which an alarm is generated when a turn signal is issued in that direction.
Specifically, this device obtains an area on the image different from the pixel density of the driving lane by binarization, and recognizes an obstacle in the driving lane.

However, in the case of this device, since only obstacles (such as other vehicles) are detected in the traveling lane detected earlier, the traveling lane cannot be detected or the traveling lane ends. In some cases, such as when there are two-wheeled vehicles, the detectability may not function well. Further, according to the method of this device, an obstacle existing on a road at a certain point in time is detected, and thus there is a possibility that a shadow or paint on the road is erroneously recognized as an obstacle. Furthermore, since only the presence of an obstacle is detected at one point at a certain time, there is a possibility that an alarm may be erroneously generated even for a vehicle that is slower than its own vehicle and relatively moves away.

Therefore, as another method of vehicle detection by image processing, a method of detecting a moving object such as a vehicle from an image continuously captured in time series has been proposed.

For example, “D. Murray et a
l. , Motion Tracking with
an Active Camera, PAMI, V
o. 16, NO. As described in US Pat. No. 5,449-559 ", this is a method of calculating a correspondence between time-series images by converting an image using a known camera movement amount, and extracting a moving object. , “Extraction of Moving Object from Moving Image Captured by Moving Camera” (IPSJ, No.5)
According to the 1st National Convention, 1995: Terakubo et al.), By arranging points of interest on images and searching for correspondence of regions having these points as vertices between time-series images, There has been proposed a method of obtaining an affine transformation parameter in two stages and extracting a moving object based on the obtained affine transformation parameter. Further, Japanese Patent Application Laid-Open No. Hei 7-110864 proposes a method of extracting a moving object by applying a two-dimensional Laplacian filter to a moving vector field and subjecting the output value to threshold processing.

[0009]

However, in the above-described conventional detection of a moving object (vehicle) using a time-series continuous image, a predetermined arithmetic processing is performed uniformly for the pixels of the entire image. Therefore, the calculation load of image processing is very large, and high-speed processing becomes difficult. In order to avoid this, a controller having a high calculation capability is required. In other words, there is a problem that a high real-time property such as prevention of collision of a vehicle is required, and it is not suitable for a device that can be mounted on a vehicle at a low cost.

[0010] The present invention has been made in view of such a situation of the prior art, and can reduce the amount of calculation when detecting another vehicle from an image of a camera mounted on the own vehicle. It is an object of the present invention to provide a collision prevention device and a collision prevention method that can speed up processing and can reliably detect another vehicle even if the device has a calculation capability that can be mounted on a vehicle. .

Further, the present invention captures an image of the rear from a camera mounted on the own vehicle, detects the other vehicle quickly and reliably from the image, and relates to a collision or abnormal approach of both vehicles based on the detection result. Another object of the present invention is to provide a collision prevention device and a collision prevention method that can surely take measures for avoiding a trouble beforehand.

[0012]

In order to achieve the above object, a collision preventing apparatus for a moving object according to the present invention comprises: an image pickup means for picking up an image of a situation around a moving object to be issued a collision prevention alarm; A detecting unit that searches for a limited detection area of each of the plurality of frames of images captured by the imaging unit in time series and detects another moving body; and the other moving state detected as the moving state of the target moving body. Prediction means for predicting the degree of possibility of collision between the two moving bodies from a relationship with the moving state of the moving body, and alarm generating means for issuing an alarm to the target moving body in accordance with the prediction result of the prediction means; It is characterized by having.

Preferably, the imaging means, the detection means, the prediction means, and the alarm generation means are all provided in the target moving body. For example, both of the moving bodies are vehicles. For example, the detecting means may determine position information of an edge existing on a road on which the target vehicle is traveling in the image and a vanishing point which is an intersection thereof, and the limited detection based on the position information Means for determining an area, means for detecting the amount of movement of the detection area between the time-series images, and only other vehicles approaching the target vehicle or traveling at the same speed as the vehicle based on the amount of movement And means for detecting

Preferably, the predicting means calculates a time required for an area of the vehicle to reach an end of the image on the image based on the detected movement amount of the another vehicle, Means for predicting a time until the other vehicle catches up with the target vehicle using a value. In this case, the alarm generation unit is configured to determine a possibility that the another vehicle will collide with the target vehicle based on the predicted time, and a unit that detects a change in a traveling direction of the target vehicle,
Means for issuing an alarm when it is determined that there is a possibility of the collision and a state in which the target vehicle is about to change the traveling direction to the lane in which the other vehicle is traveling is detected. Is also good. For example, the detecting means includes:
It is means for detecting a lighting state of a turn signal or a steering angle state of a steering wheel by a driver of the target vehicle.

[0015] In order to achieve the above object, a situation around a moving object to be issued a collision prevention alarm is imaged in a time series, and a limited detection area of each of the plurality of imaged frames is searched. To detect the degree of possibility of collision between the two moving bodies from the relationship between the moving state of the target moving body and the detected moving state of the other moving body. A warning is issued to the target moving body in accordance with the prediction result.

Further, in order to achieve the above object, there is provided a computer-readable recording medium on which a program for preventing collision of a moving object is recorded, wherein a computer-readable recording medium is provided around the moving object for which the collision prevention alarm is issued. Means for inputting a plurality of frames of images taken in time series by the imaging means for imaging the situation; means for searching for a limited detection area of each of the images to detect other moving objects; and the target moving object. Means for predicting the degree of the possibility of collision between the two moving bodies from a relationship between the moving state of the other moving body and the detected moving state of the other moving body, and an alarm is issued to the target moving body in accordance with the prediction result. A means for issuing a warning through a device.

According to such a configuration, based on the motion information in the image, it is possible to detect only the overtaking vehicle as another moving body which may collide with the own vehicle as the moving body. The use of the temporal change in the relationship between the overtaken vehicle and the own vehicle is effective in stably and reliably preventing a collision.

[0018]

DESCRIPTION OF THE PREFERRED EMBODIMENTS One embodiment of the present invention will be described below with reference to FIG.
This will be described based on 12. In this embodiment, a vehicle that moves on a road surface in a real space or the like will be described as an example of a moving object.

The collision preventing device according to this embodiment is mounted on each vehicle. As shown in FIG. 1, this collision prevention device includes a camera 1 as an imaging unit.
1, a steering angle sensor 12, a blinker sensor 13, and a vehicle speed sensor 14, and a controller 15 having a CPU, a memory, and the like, and an alarm device 16.

An A / D converter 1 is provided between the camera 11, the steering angle sensor 12, and the vehicle speed sensor 14 and the controller 15.
7, 18 and a counter 19 are inserted respectively. Further, a driver 20 is interposed between the controller 15 and the alarm 16.

The controller 15 stores programs for performing various processes described later, and automatically activates the processes by turning on a key switch of the vehicle C.
These processes are continued until the key switch is turned off.

The camera 11 is an ordinary CCD camera, CM
An OS camera, an ITV camera, or the like can be used.
In the present embodiment, the camera 11 is attached to the rear window of the vehicle C (own vehicle Cs, other vehicle Co) with its field of view facing rearward, as shown in FIG. Camera 1
1 may be attached to any position as long as the position behind the vehicle C can be imaged. Data of images of roads, vehicles, and their surroundings captured by the camera 11 are processed into digital signals continuously through the A / D converter 17 and the like in a time-series manner, and sent to the controller 15.

The steering angle sensor 12 detects the operating angle of the steering wheel of each vehicle C. This operation angle signal is transmitted to the A / D converter 1
8 converts the signal into a digital signal, and the controller 1
Sent to 5.

The turn signal sensor 13 detects a state where the left or right turn signal of the vehicle C is turned on, and uses the detected result as an on / off signal having a logical value.
Send to 5.

Further, the vehicle speed sensor 14 detects the vehicle speed of the vehicle C, for example, as a pulse signal corresponding to the rotation of the wheels. The pulse signal is measured by the counter 19 as the number of pulses per unit time, and the measured value is supplied to the controller 15 in a digital amount state.

The vehicle speed, the blinker lighting state, or the steering wheel angle are used as parameters representing the running state of the own vehicle when determining the possibility of collision with another vehicle Co approaching the own vehicle Cs from behind. Used in.

As will be described later, the controller 15 outputs a digital amount warning signal to the driver 20 when predicting the possibility of a collision between the host vehicle Cs and another vehicle Co. This signal is converted into an analog signal of a required power by the driver 20 and supplied to the alarm 16. The alarm device 16 is, for example, a unit that emits light or sound (or sound) to give an alarm to the driver, and is provided at a position that can be heard from the driver's seat or a position that can be seen from the driver's seat.

Next, the operation of the collision prevention device will be described.

The controller 15 detects another vehicle Co approaching from behind based on the schematic algorithm shown in FIG. 3, and issues an alarm if necessary.

To explain this, the controller 15 inputs the image data from the camera 11 for each frame (step S1), and then stores the image data used in the previous processing (the image data one frame before) from the memory into the work area. (Step S2).

Next, the controller 15 performs a process for detecting another vehicle Co traveling behind the host vehicle Cs. This detection processing includes detection of a white line on the road surface (step S3), calculation of the amount of movement of the feature point in the image (step S4), and detection of the overtaking other vehicle Co (overtaking vehicle) (step S5). The processing of each of these steps will be described later in detail.

When another vehicle Co trying to pass the own vehicle Cs is detected by this detection processing, a time to catch up is predicted as described later (step S6). Then, if necessary, the overtaking vehicle C is passed to the driver of the own vehicle Cs.
An alarm is issued to notify that o is present (step S7). Thereafter, when such processing is continued (step S8), the image data of the current frame input in step S1 is stored for the next detection processing (step S8).
9).

As a result, such detection processing is continued until the engine is turned off, and the detection of a passing vehicle based on two images consecutive in time series, the possibility of collision, and
An alarm generation process is performed as necessary. Hereinafter, this processing will be described in detail.

(White Line Detection Process) In step S3, the controller 15 executes a white line detection process on the road.

In the white line detection processing, edges are extracted from each input time-series image, and the white line on the road surface in the image and the straight line edges forming the ridge line of the road and their intersections. This is a process of detecting a vanishing point. The extracted white line and edge processing result and vanishing point of the ridge line are used in a movement amount calculation process described later. In the following description, linear bodies that can distinguish lanes, such as a white line, a yellow line, and a pavement ridge line, are collectively described as a “white line”.

Now, assume that image data shown in FIGS. 4A and 4B is input from a vehicle C running on a road. FIG. 11A shows an input image obtained at time t,
FIG. 2B shows an input image obtained at time t + 1.
It is continuous in chronological order. The images shown here are images at arbitrary two consecutive times in an image group input from the camera 11.

The input image at each time obtained in chronological order includes a road, its side wall, a vehicle traveling behind, and the like. (Passing vehicle).

Here, the traveling range of the overtaking vehicle is limited by a white line or a road shoulder, and the location of the overtaking vehicle can be limited by extracting a ridge line formed by the white line or the road shoulder. . Also, when the rear of a vehicle traveling straight is photographed with a general television camera, the background is FOE (focus of expans).
(ion) so as to converge to a convergence point of a motion vector. In particular, when the vehicle travels along a white line between white lines that distinguish between parallel drawn lanes, the FOE will hit a vanishing point that is the intersection of the white lines on the screen.

Generally, in order to obtain the FOE from the motion vector obtained by the image processing, the motion of the background due to the motion of the camera and the motion of a moving object other than the background are divided, and the motion of the point having the motion of the background is divided. Since it is necessary to determine the concentration point of the vector, the amount of calculation becomes very large. On the other hand, the vanishing point can be obtained by obtaining a white line or a ridge line and calculating an intersection thereof, and thus can be obtained with a relatively small amount of calculation. Therefore, in the present embodiment, a vanishing point is obtained from a white line or a ridge line, and
A method used for calculating the movement amount assuming that the OE is used is adopted. Thereby, high-speed processing can be performed with a reduced amount of calculation.

The position on the image of the vanishing point, which is the concentration point of the white line (white line, yellow line, pavement ridge line, etc.), is uniquely determined from the internal parameters of the camera 11 and the angle between the camera 11 and the ground (road surface). Can be determined. Therefore, when the camera 11 is mounted on the vehicle C, an image of the road is first taken, and the position of the white line and the position of the vanishing point on the image taken at that time are stored. Is used as a default value when vehicle detection is performed. Also,
Since the default position may be shifted due to camera vibration or the like, if the position of the white line can be detected stably during operation of the apparatus, it is also preferable to update the default value to that position. It is.

The specific white line detection processing during running is executed as follows. In order to detect a white line, an edge component is first extracted from the screen. In the edge extraction, it is necessary to detect the direction and intensity of the edge, so a Sopel operator, a Laprasin Angaussian operator, or the like is used. However, any operator other than these operators can be used as long as it can determine the strength and direction of the edge.

Next, using the obtained edge intensities and directions, white line candidates near the default value of the white line are detected. For example, as shown in FIG. 5, when the default position PW of the white line dividing the lane in which the host vehicle travels is determined, the search (detection) range RW of the white line candidate includes the default position PW of the white line as shown in the figure. It can be limited to a narrow hatched area.

The size of the search range RW can be appropriately changed in consideration of the magnitude of vibration of the camera 11 mounted on the vehicle.

Although only the white line of the lane in which the vehicle C is running is displayed here, other white lines and ridge lines can be detected and used for calculating vanishing points. However, in this case, it is difficult to define the default position of the white line other than where the host vehicle is running. Find only the white line. Next, another white line or a ridge line which passes through the vanishing point obtained from the white line constituting the own lane is searched. Thus, white lines and ridge lines other than the own lane can be detected.

Then, a vanishing point is obtained by using the least squares method using all finally obtained white lines and ridge lines. The obtained white line and vanishing point information and the edge extraction result are used for the next movement amount calculation.

(Moving Amount Calculation Processing) In step S4, the controller 15 calculates the moving amount of a feature point (feature region) on an image with respect to a time-series image (inter-frame interval) using two consecutive images. Is done.

Generally, for each image input in chronological order from the camera 11, the processing target area is limited by using information on white lines and vanishing points obtained by white line detection. By dividing the image into a plurality of small areas, the amount of movement of each small area between the time-series images is calculated. The calculated movement amount of each small area is used as the movement amount of the overtaking vehicle.

The movement amount calculation processing will be described with reference to FIGS.

The amount of movement of a feature point (feature region) between two images can be obtained by associating each feature point (feature region) between two images. For example, FIG.
The movement amounts obtained for the images (a) and (b) are shown in FIG.
An arrow shown in FIG. Thus, the movement amount can be expressed as a vector having a direction and a magnitude.

FIG. 6 shows an example of a specific method of calculating the amount of movement. This is performed by the controller 15 in step S4 of FIG.
This is executed as a process. As shown in FIG. 6, the search area RS of the overtaking vehicle is limited based on the position information of the white line and the vanishing point. For example, in FIG. 7, the area indicated by the hatched portion by the broken line becomes the search area RS, and the calculation load decreases as the search area becomes smaller. In addition, when detecting the other vehicle Co behind the own vehicle Cs, that is, the other vehicle Co traveling in the same lane as the own vehicle, the region RS ′ of the shaded portion with the solid line may be set as the search region. In FIG. 7, reference symbol PW indicates a white line of the own lane (default value), and VP indicates a vanishing point. Here, even if the vehicle is traveling in the next lane in the three-dimensional space,
When the vehicle is projected on a two-dimensional image, the vehicle may appear to protrude from between two white lines constituting an adjacent lane. Therefore, the search area is set to a butterfly shape in consideration of the influence of these projection transformations. is there. As a result, a vehicle traveling in the adjacent lane can be reliably searched.

In the actual search, first, the set search area is divided into a lattice (step S4a). Next, it is verified whether or not each lattice region has a strong edge component, and a lattice region having a strong edge component is set as a search candidate region (step S4b). When an image of the rear is taken by a camera attached to a car so that the road surface is parallel to the optical axis, the movement of other vehicles has a large X (horizontal) component in the image and does not move much in the Y (vertical) direction. Become. Therefore, the search area RS for an arbitrary tracking area can be limited to a horizontally long rectangular area as shown in FIG. Furthermore, if the detection target is limited to the amount of another vehicle passing the own vehicle, the search area is X on the left of the vanishing point.
Only in the minus direction of the axis.

Next, a corresponding area at the next time is searched for each of the search candidate lattice areas (step S4).
c). For example, each lattice area of the search candidate at time t is set as an initial position, and at time t + 1, a corresponding area is searched. That is, when each of the grid regions is moved within the search range RS, the region where the correlation value is the largest is detected. At this time, if the maximum correlation value is smaller than the threshold value, it means that the area cannot be dealt with.

Next, the amount of movement of each lattice area as a search candidate is determined (step S4d). Specifically, a difference (specifically, defined by the number of pixels) of the spatial distance between the initial position of each of the lattice regions of the search candidate at time t and their corresponding regions at time t + 1 is calculated and moved. Amount.

After that, various amounts of movement between successive images including the amount of position movement of the search candidate lattice area are stored (step S4e).

On the other hand, in addition to the processing of steps S4a to S4c, step 4f is performed.
The processing of 4d and S4e may be performed. That is,
If the search candidate grid regions have already been correlated at each time in the past, information such as luminance and edge components has been accumulated for each region. In step S4f, such information is used. The corresponding area is sequentially searched for a new image. That is, by adding step S4f, the area can be tracked continuously over several frames. At this time, the search may be performed for all search candidate grid regions obtained in the past, but since the processing amount is enormous, only the grid regions after a certain point in the past may be targeted, or at each time, By continuing the search for only the region having a high evaluation value corresponding to, the processing can be performed at higher speed.

Here, as a criterion for evaluation, not only a correlation value but also a difference value of brightness or a difference value of edge strength can be used. Further, although the region-based method has been described here, a moving amount detection method using strong feature points, a method of obtaining the motion of all points by using a gradient method, and the like can be adopted. Further, it is also possible to use the movement amount obtained by these methods. Further, when the movement amount obtained by the search is accumulated for two or more past frames, the movement amount at an arbitrary time (frame) interval can be obtained by using the accumulated data.

The information on the amount of movement of each area thus obtained is used for passing vehicle detection.

(Overtaking Vehicle Detection Process) In step S5, the controller 15 detects an overtaking vehicle (other vehicle) Co that is going to overtake the own vehicle Cs.

In this process, generally, based on the calculated movement amount of the search candidate lattice region, regions in which regions having the same amount of movement are present are extracted, and the direction of the movement amount of each region is determined. This is a process of recognizing an area where a collision is more likely than a size as an overtaking vehicle. This passing vehicle Co
Is used for alarm generation processing.

Specifically, as shown in FIG. 9, a threshold value of the speed V of the overtaking vehicle Co is set (step S5).
a). For example, when the threshold value V = 2, a grid region having a movement amount of 2 pixels or less is excluded, so that the possibility that noise or information of a stationary object is erroneously recognized as a passing vehicle is reduced,
The detection accuracy is improved. This threshold is 1 for the entire search area.
You can set one value, but if you want,
The threshold value V can be changed for each location in the search area.

Next, voting of the movement amount in the parameter space is performed (step S5b). Specifically, as shown in FIG. 10, a histogram of the movement amount included in each band region BR set on the search region is created. Several peak values appear in each histogram, which are detected (step S5c).

These peak values correspond to the speeds of each passing vehicle Co. Therefore, the histogram is divided by peak values, an area where the difference from each peak value is within the threshold value is detected, and a candidate for the overtaking vehicle Co is detected. Further, the positional relationship in the image of the regions having the respective velocities is examined, the adjacent regions are merged, and the individual passing vehicles Co are detected (step S5d).

Here, when the own vehicle Cs is moving forward,
All the backgrounds move in a direction converging on the FOE, but the overtaking vehicle Co moves in a direction spouting from the FOE. Therefore, it is possible to extract only the region where the speed in the springing direction is equal to or greater than 0, and set it as a candidate for the region representing the passing vehicle Co.

Based on the above procedure, the overtaking vehicle is detected by merging the search areas having similar positions and moving amounts on the image.

(Catching Time Prediction Process) Next, FIG.
The processing corresponding to step S6 of
5, the time at which the overtaking vehicle Co catches up with the own vehicle Cs is predicted. This process is shown in FIG.

In the present apparatus, since the angle of view of the camera 11 is known in advance, when the overtaking vehicle Co arrives at the left and right ends of the image, the overtaking vehicle Co is compared with the own vehicle Cs in the three-dimensional scene. You can ask if you are in a position.
Therefore, the position of the tip of the overtaking vehicle Co on the screen is calculated first (step S6a). Next, when the vehicle continuously moves forward at the current speed on the screen of the overtaking vehicle Co, the time to reach the edge of the camera screen in a few seconds is calculated. Then, the final time until the overtaking vehicle Co is overtaken is predicted in consideration of the relative distance in the three-dimensional scene between the overtaking vehicle Co and the own vehicle Cs when the overtaking vehicle Co reaches the screen edge (step S).
6b).

Information such as the detected position of the area of the overtaking vehicle Co and the time until overtaking is used in the alarm generation processing.

(Alarm Generation Processing) Next, as processing corresponding to step S7 in FIG.
If necessary, an alarm is issued to the own vehicle Cs. This process is shown in FIG.

This warning is issued when the overtaking vehicle Co, which may collide, exists around the own vehicle Cs to issue a turn signal in that direction to indicate the intention to change lanes or to change the lane. This is a process to warn the driver that there is a possibility of collision when the steering wheel is turned to the direction.

Specifically, as shown in FIG. 12, the controller 15 determines whether or not the overtaking vehicle Co has been detected (step 7a). If it is determined that the overtaking vehicle Co has been detected in this determination, then it is determined whether the catching time is equal to or less than a threshold (step S7).
b). That is, when the passing vehicle Co exists, it is determined whether or not the vehicle can catch up within the time set as the threshold.

In this determination, when it is determined that the overtaking vehicle Co catches up with the own vehicle Cs in a time equal to or less than the threshold value (that is, catches up soon), the controller 15 then causes the driver of the own vehicle Cs to overtake the overtaking vehicle. It is determined whether or not Co is trying to change the traveling direction of the vehicle to the lane traveling while catching up (step S7).
c). This determination is made while reading the detection signals of the steering angle sensor 12 and / or the turn signal sensor 13 and monitoring the steering operation or the turn signal operation.

If the determination of the direction change is YES, it means that the overtaking vehicle Co is approaching from behind, and if the traveling direction is changed as it is, there is a high risk of collision with the following vehicle. Therefore, the controller 15
Sends a control signal for alarm to the alarm 16
Generate an alarm from. That is, when a turn signal is issued in the direction of the overtaking vehicle to indicate the intention to change lanes in that direction, or when the driver turns the steering wheel to change lanes in that direction, the alarm 16 emits a sound (voice) and / or light. Warns the driver.

Therefore, the driver can surely know the danger of collision with the following vehicle, take appropriate measures such as stopping the lane change, and contribute to safe driving.

Further, as a further advantage, according to the method of the present embodiment, since the movement of each area of the image can be known, it is possible to extract only vehicles that pass the own vehicle. Further, since two continuous images with little temporal change are used, the influence of the threshold value can be suppressed as compared with a method such as binarization, and a highly accurate collision prevention device can be realized. Furthermore, in a three-dimensional scene, even if a vehicle actually exists on the driving lane, it may appear that no obstacle exists on the driving lane in the image depending on the installation position of the camera. ,
On the basis of the motion information, the overtaking vehicle in the image can be selectively detected with high accuracy.

It is to be noted that the alarm may be generated in an embodiment in which, when there is another vehicle that may collide around the own vehicle, the driver is notified by sound or light that the vehicle is present in that direction. is there. Further, since the controller 15 recognizes the position and speed of the overtaking vehicle, the controller 15 changes the color, intensity, or sound intensity of the light in accordance with the time until the overtaking vehicle catches up based on the information, so that the overtaking vehicle can be recognized. A variant is possible in which the presence is informed to the driver in advance.

(Modification 1) One of the embodiments described above.
One modification will be described with reference to FIGS.

This embodiment uses the camera 1 as described above.
The present invention relates to an apparatus for detecting a moving amount of a characteristic portion moved between image frames from a time-series continuous image captured in 1. In particular, in order to stably detect the amount of movement of a characteristic part from an image, a strong edge in the image is used for the characteristic part (movement amount calculation target).
The present invention relates to an apparatus for detecting a movement amount of a characteristic portion at high speed by using a constraint on movement of an edge in an image.

This detection device is mounted on the own vehicle in the collision prevention device described in the above-described embodiment, and effectively functions as a device that captures an image of a succeeding vehicle and detects the amount of movement of a characteristic portion from a time-series image thereof. I do. In order to achieve this, the detection device includes imaging means such as a camera, detection means for detecting a characteristic portion such as a strong edge from a plurality of time-series images sequentially output from the imaging means, and Calculating means for calculating the amount of movement of the characteristic part by limiting the search direction. The detecting means and the calculating means are constituted by, for example, a computer.

As a characteristic portion having a strong edge, for example, FIG.
As shown in FIG. 3, a trunk 31 of a tree beside the road, a front bumper 32 of another vehicle running behind, a pole part 3 of a road sign.
2 and so on. For example, as shown in FIG. 14, when a characteristic part is detected in the pole part 33 of the road sign, the search range (direction) is limited (constrained) on an extension line connecting this characteristic part and the vanishing point VP, and the characteristic is detected. The movement amount of the part (in this case, the own vehicle) can be obtained at high speed and with high accuracy.

As described above, by limiting the edge strength of the characteristic portion whose movement amount is to be calculated and the search direction when calculating the movement amount, the movement amount of the characteristic portion can be detected at high speed and with high accuracy. .

(Modification 2) Another modification of the above embodiment will be described with reference to FIG.

This modification is also similar to that of the camera 1 as described above.
The present invention relates to an apparatus for detecting a moving amount of a feature amount moved between image frames from a time-series continuous image captured in step 1 with high accuracy.

When detecting the amount of movement of a characteristic portion between time-series images, if a repetitive pattern such as a guide rail 34 or a zebra pattern 35 is reflected in the search direction as shown in FIG. An error may occur in the calculation of the movement amount of the feature amount due to the repeated pattern of the value.

Therefore, a camera as imaging means attached to the rear of the vehicle, a means for calculating a feature amount in the search direction and its change from a plurality of time-series images taken by the camera, and a repetitive pattern having a predetermined period or more. Means for judging whether or not there is, and means for prohibiting calculation of the amount of movement of the feature amount for the existence range, assuming that an accurate movement amount cannot be calculated when it is determined that there is a repetitive pattern. On the other hand, when such a repetitive pattern no longer exists, a means for restarting the calculation of the movement amount is provided,
As a result, only a stable movement amount can be accurately calculated. Here, each of the above means can be functionally realized by causing a computer to execute a predetermined program.

This detection device is mounted on the own vehicle in the collision prevention device described in the above-described embodiment, and can be used as a device that captures an image of a following vehicle and detects the amount of movement of a characteristic portion from a time-series image. it can.

In the above-described embodiment and its modifications, the detecting means, the calculating means, and the calculating means are functionally configured by causing a computer (controller) to execute a predetermined program. An electronic circuit such as a circuit may be configured to perform the same function.

Although the above-described embodiment and its modifications have been described by exemplifying a vehicle running on a road as a moving body, the moving body of the present invention includes a train and an aircraft.

[0088]

As described above, according to the present invention, based on motion information in an image, only an overtaking vehicle as another moving object which may collide with the own vehicle as the moving object is detected. By using the detected temporal change in the relationship between the overtaking vehicle and the own vehicle, it is possible to contribute to a stable and reliable collision prevention effect.

[Brief description of the drawings]

FIG. 1 is a block diagram showing a schematic configuration of a vehicle-mounted collision prevention device according to an embodiment of the present invention.

FIG. 2 is a schematic view showing an example of a state in which the camera is mounted on a vehicle.

FIG. 3 is a schematic flowchart of a collision prevention process executed by a controller of the apparatus.

FIG. 4 is an image diagram illustrating a calculation of a moving amount of an overtaking vehicle.

FIG. 5 is a view for explaining a relationship between a default position of a white line and a white line search area.

FIG. 6 is a schematic flowchart illustrating the calculation of a movement amount.

FIG. 7 is a diagram illustrating a relationship between a lane area and a search range of an overtaking vehicle.

FIG. 8 is a view for explaining an example of another search range of an overtaking vehicle.

FIG. 9 is a schematic flowchart illustrating an overtaking vehicle detection process.

FIG. 10 is an explanatory diagram of a band area for detecting an overtaking vehicle.

FIG. 11 is a schematic flowchart illustrating a process of predicting a catch-up time of an overtaking vehicle.

FIG. 12 is a schematic flowchart illustrating an alarm process.

FIG. 13 is a diagram of a modified example illustrating detection of a movement amount of a feature amount.

FIG. 14 is a view for explaining the relationship between detection of a movement amount of a feature amount and a search direction in FIG. 13;

FIG. 15 is a view of another modified example for explaining detection of a movement amount of a feature amount.

[Explanation of symbols]

 Reference Signs List 11 camera 12 steering angle sensor 13 turn signal sensor 14 vehicle speed sensor 15 controller 16 alarm device 17, 18 A / D converter 19 counter 20 driver C vehicle Cs own vehicle amount (target moving body) Co other vehicle (overtaking vehicle, other movement) Body) VP vanishing point RW White line search area PW White line (default position) RS Overtaking vehicle search area RS 'Traveling lane area BR band area

Claims (9)

[Claims] 1. An image pickup means for picking up an image of a situation around a moving object to be issued a collision prevention alarm, and the image pickup means searches a limited detection area of each of a plurality of frames of images taken in time series. Detecting means for detecting another moving body, and predicting a degree of possibility of collision between the two moving bodies from a relationship between a moving state of the target moving body and a detected moving state of the other moving body. An apparatus for preventing collision of a moving body, comprising: a predicting unit; and an alarm generating unit that issues an alarm to the target moving unit in accordance with a prediction result of the predicting unit. 2. The collision preventing apparatus according to claim 1, wherein the image pickup unit, the detection unit, the prediction unit, and the alarm generation unit are all provided in the target mobile unit. Prevention device. 3. The collision preventing apparatus according to claim 2, wherein both of the moving bodies are vehicles. 4. The collision prevention device according to claim 3, wherein the detection unit determines position information of an edge existing on a road on which the target vehicle is traveling in the image and a vanishing point that is an intersection thereof. Means for determining the limited detection area based on the position information, means for detecting the amount of movement of the detection area between the time-series images, and whether to approach the target vehicle based on the amount of movement. Or a means for detecting only another vehicle running at the same speed as the vehicle. 5. The collision prevention device according to claim 4, wherein the prediction unit calculates a time required for the area of the vehicle to reach the end of the image on the image based on the detected movement amount of the another vehicle. And means for predicting a time required for the other vehicle to catch up with the target vehicle using the calculated value. 6. The collision prevention device according to claim 5, wherein the warning generation unit determines a possibility that the another vehicle will collide with the target vehicle based on the predicted time;
Means for detecting a change in the traveling direction of the target vehicle, and a state in which it is determined that there is a possibility of the collision and the target vehicle is changing the traveling direction to the lane direction in which the other vehicle is traveling; Means for issuing an alarm when detected. 7. The moving body according to claim 6, wherein the detecting means is means for detecting a lighting state of a turn signal or a steering angle state of a steering wheel by a driver of the target vehicle. Anti-collision device. 8. A time-series image of the surroundings of a moving object to be issued a collision prevention alarm, and searches for a limited detection area of each of the plurality of imaged frames to detect another moving object. Then, the degree of the possibility of collision between the two moving bodies is predicted from the relationship between the moving state of the target moving body and the detected moving state of the another moving body. A method for preventing a collision of a moving object, wherein a warning is issued to the moving object. 9. A computer-readable recording medium on which a program for preventing collision of a moving object is recorded, wherein imaging means for imaging a situation around the moving object for which an alarm for preventing collision is issued is provided. Means for inputting images of a plurality of frames taken in a sequence, means for searching for a limited detection area of each of the images and detecting other moving objects, and detecting the moving state of the target moving object and Means for predicting the possibility of collision between the two moving bodies from the relationship with the moving state of the other moving bodies, and issuing an alarm to the target moving body via an alarm in accordance with the prediction result. And a program for executing the means.