JPH1120546A - Readward/sideway monitering method for vehicle and device thereof - Google Patents

Abstract

PROBLEM TO BE SOLVED: To detect a vehicle approaching from the rear part and the side part and to warn a driver of the approach by monitering the degree of approach of a succeeding vehicle or a vehicle running on adjacent lane against one's own vehicle, based on an optical flow and a point of infinity. SOLUTION: An optical flow forming circuit 13 forms an optical flow by connecting a feature point to a correspondence point, and sends it to a degree of danger judging circuit 14. The degree of danger judging circuit 14 judges the degree of approach (i.e., the degree of danger) of the speed of a vehicle running on the adjacent lane and a succeeding vehicle, according to whether the optical flow is going from a point of infinity (FOE) in the direction of radiating, or in the direction of converging into the FOE, and sends the judged result to a warning part 15. Namely, as the optical flow is largely going in the direction of radiating from the FOE, the degree of danger is judged to be higher.

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention detects a vehicle approaching from the rear and side when the vehicle is running, using images captured by the video camera installed on the vehicle such as an automobile. The present invention relates to a method and an apparatus for monitoring a vehicle rear side for giving a warning to a driver.

[0002]

2. Description of the Related Art When a vehicle changes lanes on a road having two or more lanes on one side, if a vehicle running at a faster speed than the own vehicle is overlooked on an adjacent lane to be changed, a large accident may occur. The danger is great. In addition, when there is a following vehicle traveling in the same lane as the own vehicle, if a sudden brake is applied when approaching suddenly, there is a risk of a collision, and it is necessary to recognize a nearby vehicle.

As a conventional example for recognizing a vehicle traveling in an adjacent lane and a following vehicle, Japanese Patent Application Laid-Open No. 1-189289 (vehicle information display device) has been proposed. Japanese Patent Application Laid-Open No. Hei 1-189289 discloses a technique in which a rear side and a side of a vehicle are photographed by a camera, and the photographed images are displayed on a monitor so that a vehicle traveling in an adjacent lane and a following vehicle can be recognized. ing.

However, there is a risk that the driver may overlook the vehicle displayed on the monitor, and it is difficult to determine whether the speed of the vehicle traveling in the adjacent lane and the speed of the following vehicle are higher than the speed of the own vehicle only by looking at the monitor. It was difficult to judge immediately.

Conventionally, a point at an infinity point or a vanishing point (generally, an FOE (Focus)) in which each point of a captured image obtained during traveling of the vehicle by a video camera attached to the rear side of the vehicle converges as a whole. of the feature points in the captured image (hereinafter referred to as optical flow),
According to whether the optical flow is directed in a direction diverging from the FOE or in a direction converging on the FOE, the approaching speed of the vehicle traveling in the adjacent lane and the speed of the following vehicle to the own vehicle is determined. A rear side monitoring method for vehicles has been proposed (Japanese Patent Application No. 5-196189).
issue).

This method of monitoring the rear side of the vehicle will be briefly described with reference to FIG. FIG. 11 is a temporally continuous image obtained by a camera mounted rearward behind the host vehicle. FIG. 11 shows a case where a vehicle running in the adjacent lane in which the host vehicle is running passes the host vehicle.

First, a corresponding point corresponding to the feature point A in FIG. 11A is searched for on a straight line connecting the FOE and the feature point A in FIG. In this example, point A ′ in FIG. 11B is detected as a corresponding point. The vector connecting the point A and the point A 'is an optical flow. In this example, since the optical flow is diverging from the FOE, the vehicle traveling in the adjacent lane has a large approach to the own vehicle. Is determined.

[0008]

However, FIG.
In the case shown in (1), since the corresponding point cannot be obtained correctly, an erroneous optical flow may be obtained. To explain this, assuming that a point on the white line is set as a feature point B in FIG. 12A, it is necessary to detect a corresponding point B ′ corresponding to the feature point B from FIG.

At this time, as shown in FIG.
A point (corresponding point) having a correlation with the mask 1 formed on the feature point B in (A) is searched by the mask 2 from a straight line connecting the FOE and the feature point B in FIG. Since there are many points having the same luminance distribution as in the mask 1, it is difficult to detect the corresponding point, and there is a high possibility that an erroneous point will be set as the corresponding point. In fact, in the example of FIG. 13, the corresponding point B ′ corresponding to the feature point B is detected on the opposite side of the FOE with respect to the feature point B, but the true corresponding point is more FOE than the feature point B.
Should be detected on the side white line.

When an erroneous corresponding point is detected for a feature point and an erroneous optical flow is obtained,
It is determined that a vehicle running behind or adjacent to the own vehicle is approaching the vehicle, but it is judged that the vehicle is approaching, or, on the contrary, there is a vehicle approaching, but the vehicle is overlooked. And the reliability of the rear side monitoring is significantly reduced.

The present invention has been made in consideration of the above points, and a vehicle rear side monitoring method and a vehicle rear side monitoring method capable of performing highly reliable rear side monitoring by obtaining an accurate optical flow. It is intended to propose a side monitoring device.

[0012]

According to a first aspect of the present invention, there is provided a method for monitoring a rear side of a vehicle, comprising the steps of:
In the vehicle rear side monitoring method of detecting the movement of the corresponding point in the image of two frames before and after the predetermined time as an optical flow, and monitoring the relative relationship of the vehicle running in the rear lane or the adjacent lane to the own vehicle, A feature point is detected from a captured image at a certain point in time, and feature points that are linearly arranged through the infinite point (FOE) of the captured image at the certain point in time are deleted from the feature points. In images that are temporally consecutive, a point on a straight line connecting the remaining feature point without being deleted and the point at infinity of the captured image as a candidate point,
Detect corresponding points corresponding to the remaining feature points without being deleted,
A vector connecting the feature point and the corresponding point is detected as an optical flow, and based on the optical flow and the point at infinity, the degree of approach of a vehicle traveling behind or adjacent to the own vehicle is monitored.

According to a fifth aspect of the present invention, there is provided a vehicle rear side monitoring apparatus according to the present invention, wherein a rear view is imaged from an own vehicle running by an image pickup means, as shown in FIG. In the rear side monitoring device 1 for a vehicle, which detects the movement of the corresponding point in the two frames of images before and after the predetermined time as an optical flow, and monitors the relative relationship between the own vehicle and a vehicle traveling in a rear lane or an adjacent lane. A feature point selecting means 5 for detecting a feature point from a captured image at a certain time, and deleting a feature point linearly arranged through an infinity point (FOE) of the captured image at the certain time from the feature points; In an image temporally before and after an image at a certain time, a point on a straight line connecting a feature point remaining without being deleted by the feature point selection means 5 and an infinity point of the captured image as a candidate point, For feature points that remain without being deleted A corresponding point detecting means 12 for detecting the corresponding points, optical flow forming means 1 for detecting a vector connecting the feature points and the corresponding points as optical flow
Risk determining means 14 for determining the degree of danger by obtaining the degree of approach of a vehicle traveling in a rear lane or an adjacent lane to the own vehicle based on the optical flow and the point at infinity
And was prepared.

In the above configuration, the corresponding point detecting means 12
In the above, the characteristic points for which the detection of the corresponding points becomes difficult, that is, the characteristic points in which a plurality of similar corresponding candidate points exist on a straight line connecting the characteristic points and the point at infinity, Since they are deleted in advance, the corresponding point detecting means 12
Can accurately and quickly detect a corresponding point corresponding to a feature point. As a result, the optical flow forming unit 13 can obtain an accurate optical flow, and the risk determination unit 14 can perform a reliable determination process.

According to a second aspect of the present invention, when a feature point is deleted, a first mask image including peripheral pixels including the feature point is formed when the feature point is deleted. Forming a second mask image on a straight line adjacent to the first mask image and passing through the feature point and the point at infinity, comparing the correlation between the first mask image and the second mask image, When the correlation between the first mask image and the second mask image is strong, the feature points are deleted.

According to a sixth aspect of the present invention, there is provided a rear side monitoring apparatus for a vehicle, as shown in FIG. Mask image forming means 7 for forming a first mask image, and forming a second mask image on a straight line adjacent to the first mask image and passing through the feature point and the point at infinity; An inter-comparison means 10 for comparing the correlation between the mask image and the second mask image, and a feature point deletion means 11 for deleting a feature point when the first mask image and the second mask image have a strong correlation. I prepared for it.

In the above configuration, the corresponding point detecting means 12
In this case, it is possible to easily and reliably delete a feature point that makes it difficult to detect a corresponding point.

According to a third aspect of the present invention, when a feature point is deleted, an edge image of a captured image at a certain point in time is formed, and an edge in the edge image is formed. Set the above predetermined point as a feature point,
Forming a first mask image including peripheral pixels including the feature point, and forming a second mask image on a straight line that is adjacent to the first mask image and that passes through the feature point and the point at infinity; The correlation between the first mask image and the second mask image is compared, and when the correlation between the first mask image and the second mask image is strong, the feature point is deleted.

In the vehicle rear side monitoring apparatus according to a seventh aspect of the present invention, as shown in FIG. 1, the feature point selecting means 5 according to the fifth aspect is arranged such that the feature point selecting means 5 is adapted to detect an edge of a captured image at a certain point in time. Edge image forming means 6 for forming an image, feature point setting means 8 for setting a predetermined point on an edge in the edge image as a feature point, and forming a first mask image including peripheral pixels including the feature point And a mask image forming means 7 for forming a second mask image on a straight line adjacent to the first mask image and passing through the feature point and the point at infinity.
An inter-comparison means 10 for comparing the correlation between the first mask image and the second mask image;
And a feature point deleting unit 11 for deleting a feature point when the mask image has a high correlation.

In the above configuration, by using the points on the edge image as the feature points, the feature points sufficiently reflecting the features of the image can be efficiently narrowed down from the picked-up image. Can be effectively reduced, and a more accurate optical flow can be detected.

According to a fourth aspect of the present invention, there is provided a method for monitoring a rear side of a vehicle, comprising the steps of: deleting a feature point together with a feature point to be deleted; Straight lines are now deleted all at once.

According to a further aspect of the present invention, there is provided a rear side monitoring apparatus for a vehicle, wherein the characteristic point deleting means according to claim 6 or 7 includes the characteristic point to be deleted together with the characteristic point to be deleted. The line portion of the detected edge image is collectively deleted.

In the above configuration, instead of setting feature points one by one and performing feature point selection processing on those feature points, when one feature point to be deleted is detected, the feature point is determined. The feature points that are linearly aligned with the points are collectively deleted, so that the feature points that will be deleted in the subsequent feature point selection process can be deleted without performing those processes. As a result, the number of times of the feature point setting process and the feature point deletion process can be significantly reduced.

[0024]

DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS An embodiment of the present invention will be described below with reference to the drawings.

(1) Configuration In FIG. 1, reference numeral 1 denotes a rear side monitoring device for a vehicle as a whole. An image signal S1 picked up by a video camera 2 mounted rearward behind the host vehicle is stored in a frame memory 3 and a FOE. It is supplied to the detection circuit 4. The FOE detection circuit 4 detects a point at infinity (that is, FO) from a temporally continuous image.
E) is detected.

The image signal S1 of one frame past read from the frame memory 3 is input to the high-pass filter (HPF) 6 and the mask image forming circuit 7 of the feature point selection unit 5. The high-pass filter 6 functions as an edge image forming unit for forming an edge image of the image signal S1,
An edge image signal S2 is formed by passing only the high-frequency component of the image signal S1 (in other words, differentiating the image signal S1), and this is sent to the feature point setting circuit 8.
That is, an edge image as shown in FIG. 3 is formed from a raw image as shown in FIG.

The feature point setting circuit 8 sets a feature point P1 on the edge as shown in FIG. 3 based on the edge image signal S2, and sends the coordinate information S3 of the set feature point P1 to the inclination detection circuit 9. . The tilt detection circuit 9 is provided with coordinate information S3 of the feature point P1 and FOE coordinate information S4 from the FOE detection circuit 4.
As shown in FIG. 3, the inclination θ of the straight line L1 connecting the FOE and the feature point P1 is obtained, and this is sent to the mask image forming circuit 7 as inclination information S5.

The mask image forming circuit 7 receives the image signal S1, the coordinate information S3 of the characteristic point P1, and the inclination information S5, and as shown in FIG. While setting the mask A, as shown in FIG. 4, a mask B and a mask C having the same size as the mask A are formed above and below the mask A at an inclination θ. Then, the pixel information in the masks A, B and C is sent to the mutual comparison circuit 10 as mask image information S6.

The mutual comparison circuit 10 determines the correlation between the masks A and B by comparing the pixels in the masks A and B based on the mask image information S6, and also compares the pixels in the masks A and C. Thus, the correlation between the mask A and the mask C is obtained, and the obtained comparison result S7 is sent to the feature point deletion circuit 11.

The feature point elimination circuit 11 compares the mask A with the mask B and the mask C to indicate that the correlation is high.
Is input, the coordinate information S3 of the feature point is deleted. On the other hand, the mask A is the mask B and the mask C
When the comparison result S7 indicating that the correlation is low is input to the corresponding point detection circuit 12 and the optical flow formation circuit 13 without deleting the coordinate information S3 of the feature point. I have. This allows the feature point selection unit 5 to delete in advance the feature points whose corresponding points are difficult to detect.

More specifically, in the case shown in FIG. 4, since the mask A has a high correlation with the masks B and C, the mutual comparison circuit 10 obtains a comparison result S7 indicating this, and the feature point is deleted. In the circuit 11, the feature point P1 is deleted.

On the other hand, as shown in FIGS.
The characteristic point P2 is set by the characteristic point setting circuit 8, and the characteristic point P2 is set.
In the case where masks B and C having the same size as the mask A are formed at the upper and lower positions of the inclination A 'with respect to the mask A according to 2, the correlation between the mask A and the masks B and C is low. The comparison result S7 indicating this in the mutual comparison circuit 10
Are obtained and sent to the corresponding point detection circuit 12 and the optical flow forming circuit 13 without deleting the feature point P2 in the feature point deletion circuit 11.

As a result, as shown in FIG. 8, the feature point deletion circuit 11 outputs an edge image from which feature points that are arranged in a straight line through the FOE are deleted as an edge image signal S8. Become.

The corresponding point detection circuit 12 includes the feature point deletion circuit 1
A corresponding point corresponding to the feature point coordinates at time t input from 1 is searched for on the image at time t + Δt input from the video camera 2. That is, as described above with respect to the figure, a corresponding point corresponding to the feature point is searched for on the straight line connecting the FOE and the feature point.

The optical flow forming circuit 13 forms an optical flow by connecting the characteristic points and the corresponding points, and sends the optical flow to the risk determining circuit 14. The risk determination circuit 14 determines whether the speed of the vehicle traveling in the adjacent lane and the speed of the following vehicle are close to the own vehicle according to whether the optical flow is flowing in the direction diverging from the FOE or in the direction converging on the FOE. (I.e., the degree of risk) is determined, and the determination result is sent to the alarm unit 15. In other words, it is determined that the greater the optical flow is in the direction diverging from the FOE, the higher the risk is.

The alarm unit 15 comprises an alarm buzzer, an alarm lamp, and the like. When a judgment result indicating danger is input from the danger judgment circuit 14, the alarm buzzer is sounded or the alarm lamp is turned on. Thus, the driver is informed that there is a vehicle approaching from the rear side.

(2) Operation In the above configuration, the rear side monitoring device 1 for a vehicle is configured as shown in FIG.
By executing the rear-side monitoring processing procedure as shown in (1), it is possible to perform accurate rear-side monitoring by obtaining an accurate optical flow.

That is, first, at step SP1, the image at the time point t is fetched from the video camera 2, and at step SP2, the image at the time point t + Δt is fetched from the frame memory 3. Then, at the next step SP3, the FOE detection circuit 4 detects the FOE.

Next, in step SP4, the feature point selection unit 5 deletes the feature points that are difficult to deal with. Then, in the subsequent step SP5, the risk determination circuit 14 sets the FOE based on the optical flows obtained by the corresponding point detecting circuit 12 and the optical flow forming circuit 13.
Is detected, and the risk is calculated in the following step SP6.

Here, step SP4, which is a feature of the present invention, is described.
Will be described in more detail with reference to FIG. This feature point deletion processing SP4 is performed in the feature point selection unit 5 of FIG. When entering the feature point deletion processing SP4, the feature point selection unit 5 first obtains a raw image from the frame memory 3 in step SP10. Subsequent step SP11
Then, the characteristic point is extracted by performing a differentiation process with the high-pass filter 6 and selecting a predetermined characteristic point from the edge image by the characteristic point setting circuit 8.

In step SP12, the inclination detecting circuit 9 calculates the inclination θ of a straight line connecting the characteristic point and the FOE.
In a succeeding step SP13, a mask A is created at the position of the feature point by the mask image forming circuit 7, and a step SP1
In step 4, a mask B and a mask C are created at upper and lower positions in the direction of the inclination θ with respect to the mask A. By the way, this step S
The processing from P10 to step SP14 is the same as the conventional processing.

Next, in step SP15, it is determined by the mutual comparison circuit 10 whether or not the correlation between the mask A and the mask B has been obtained.
The process proceeds to 16 to determine whether or not the correlation between the mask A and the mask C is obtained. Here, step SP15 and step S
If an affirmative result is obtained in P16, step S
The process moves to P17, and if a negative result is obtained in any one or both, the process moves to step SP18 without performing the process in step SP17.

In step SP17, the feature point deletion circuit 1
1 deletes the feature point currently being processed. As a result, similar feature points such as feature points linearly arranged in line through the FOE on the edge image are deleted, so that an erroneous corresponding point can be prevented from being detected by the corresponding point detection circuit 12 in FIG. Become like

In step SP18, it is determined whether or not processing of all the feature points on the edge image has been completed.
Returning to P12, for the next feature point, steps SP12-
The processing up to step SP17 is repeated. On the other hand, if a positive result is obtained in step SP18, the process proceeds to step SP19 and the feature point deletion processing routine SP
Then, the process proceeds to step SP5 in FIG.

(3) Effects According to the above arrangement, after the characteristic points linearly arranged through the FOE are deleted from the captured image at a certain point in time, an image which is temporally adjacent to this image is deleted. By detecting the corresponding points in the above, the corresponding points corresponding to the feature points can be detected accurately and promptly. Thus, if the vector connecting the feature point and the corresponding point is detected as an optical flow, an accurate optical flow can be obtained, and the vehicular rear side monitoring device 1 capable of performing highly reliable rear side monitoring can be provided. realizable.

(4) Other Embodiments In the above embodiment, an edge image is formed by the high-pass filter 6 as an edge image forming means.
The case where the point on the edge image is used as a feature point has been described. However, the present invention is not limited to this. For example, when detecting a feature point, a texture image in which a texture component of the image signal S1 is extracted is formed. A point on the image may be set as a feature point.

In the above embodiment, as described with reference to FIG. 10, a case has been described in which feature points are set one by one and feature point selection processing is performed on those feature points. However, the present invention is not limited to this. When one feature point to be deleted is detected, the feature points linearly aligned with that feature point may be collectively deleted. In this way, the feature points that will be deleted in the subsequent feature point selection process can be collectively deleted without performing those processes. As a result, the number of times of the feature point setting process and the feature point deletion process can be significantly reduced.

In the above embodiment, the mask A
The mask B and the mask C having the same size as the mask A are formed above and below the inclination θ, but the present invention is not limited to this.
Alternatively, the mask C may be shifted left and right by several pixels to obtain the correlation with the mask A. In this way, even if the inclination of the edge portion to be deleted is slightly different from the straight line passing through the feature point and the FOE, the edge portion can be deleted.

In the above embodiment, the time t +
When detecting a feature point from the image of Δt, an edge image obtained by differentiating the raw image is used, and when searching for a corresponding point corresponding to this feature point, the raw image at time point t is used as it is. However, the present invention is not limited to this, and when searching for a corresponding point, an edge image may be formed from the image at the time point t, and the search may be performed based on the edge image. This makes it possible to perform a search using an image having a small amount of information as compared with a case where a search is performed using a raw image, so that a corresponding point can be detected in a much shorter time. become able to.

[0050]

As described above, according to the first and fifth aspects of the present invention, feature points that are arranged linearly from a captured image at a certain point in time through the infinite point (FOE) of the captured image are obtained. After the deletion, the corresponding point is detected in the image temporally before and after this image, so that the corresponding point corresponding to the feature point can be detected accurately and promptly. Thus, if a vector connecting a feature point and its corresponding point is detected as an optical flow, an accurate optical flow can be obtained, and a reliable rear side monitoring method and vehicle capable of performing highly reliable rear side monitoring can be obtained. A rear-side monitoring device can be realized.

According to the second and sixth aspects of the present invention, when a feature point is deleted, a first mask image including peripheral pixels including the feature point is formed, and the first mask image is added to the first mask image. A second mask image is formed on a straight line that is adjacent and passes through the feature point and the point at infinity, the correlation between the first mask image and the second mask image is compared, and the first mask image and the second mask image are compared. When the correlation of the mask image is strong, the feature points are deleted, so that the feature points that make it difficult to detect the corresponding points can be easily and reliably deleted. The device can be realized.

According to the third and seventh aspects of the present invention, when deleting a feature point, an edge image of a captured image at a certain point in time is formed, and a predetermined point on the edge in the edge image is determined as a feature point. To form a first mask image including peripheral pixels including the feature point, and a second mask image on a straight line adjacent to the first mask image and passing through the feature point and the point at infinity. And comparing the correlation between the first mask image and the second mask image. When the correlation between the first mask image and the second mask image is strong, the feature point is deleted. As a result, it is possible to efficiently narrow down feature points sufficiently reflecting the features of the image from the captured image, thereby effectively reducing the number of calculations as a whole and detecting a more accurate optical flow for a vehicle. It becomes possible to realize a lateral monitoring method and device.

According to the fourth and eighth aspects of the present invention, the feature point to be deleted and the straight line portion of the edge image in which the feature point to be deleted is detected are collectively deleted. This makes it possible to realize a vehicle rear side monitoring method and apparatus that can significantly reduce the number of times of the setting process and the feature point deletion process.

[Brief description of the drawings]

FIG. 1 is a block diagram showing a configuration of a vehicle rear side monitoring device according to an embodiment.

FIG. 2 is a schematic diagram for explaining mask image formation.

FIG. 3 is a schematic diagram for explaining edge image formation.

4 is a schematic diagram for explaining mask image formation for a feature point P1 in FIG. 3;

FIG. 5 is a schematic diagram for explaining mask image formation.

FIG. 6 is a schematic diagram for explaining edge image formation.

FIG. 7 is a schematic diagram for explaining mask image formation for a feature point P2 in FIG. 6;

FIG. 8 is a schematic diagram illustrating an edge image remaining after the feature point deletion processing.

FIG. 9 is a flowchart for explaining the operation of the vehicle rear side monitoring device according to the embodiment;

FIG. 10 is a flowchart illustrating a feature point deletion processing routine performed by a feature point selection unit.

FIG. 11 is a schematic diagram used for describing detection of an optical flow.

FIG. 12 is a schematic diagram illustrating a case where an erroneous optical flow is detected.

FIG. 13 is a schematic diagram illustrating a case where an erroneous optical flow is detected.

[Explanation of symbols]

 REFERENCE SIGNS LIST 1 rear side monitoring device for vehicle 2 video camera (imaging means) 5 feature point selection unit (feature point selection means) 6 high-pass filter (edge image formation means)

Claims (8)

[Claims] 1. An image of a rear view of a running own vehicle is captured, a movement of a corresponding point in two frames preceding and following a predetermined time is detected as an optical flow, and a rear vehicle or an adjacent lane to the own vehicle is detected. In a vehicular rear side monitoring method for monitoring a relative relationship of a running vehicle, a feature point is detected from a captured image at a certain time, and an infinite point (FO) of the captured image at the certain time among the characteristic points is detected.
E), removing the feature points linearly arranged through E), and in an image temporally before and after the image at a certain time, the feature points remaining without being deleted and the infinity point of the captured image are determined. Assuming a point on the connecting straight line as a candidate point, detecting a corresponding point corresponding to the feature point remaining without being deleted, detecting a vector connecting the feature point and the corresponding point as an optical flow, A rear side monitoring method for a vehicle, wherein the degree of approach of a vehicle running in a rear lane or an adjacent lane to the own vehicle is monitored based on the infinity point. 2. The method according to claim 1, further comprising: forming a first mask image including peripheral pixels including the feature point, deleting the feature point, and adjoining the first mask image, and the feature point and the infinity. Forming a second mask image on a straight line passing through points; comparing a correlation between the first mask image and the second mask image; a correlation between the first mask image and the second mask image; 2. The method according to claim 1, wherein the characteristic point is deleted when the characteristic is strong. 3. When deleting the feature point, an edge image of the captured image at a certain point in time is formed, a predetermined point on an edge in the edge image is set as a feature point, and a periphery including the feature point is set. Forming a first mask image composed of pixels, forming a second mask image on a straight line that is adjacent to the first mask image and that passes through the feature point and the point at infinity; Comparing the correlation between the image and the second mask image, and deleting the feature point when the correlation between the first mask image and the second mask image is strong. Item 7. The method for monitoring a rear side of a vehicle according to Item 1. 4. The method according to claim 1, wherein when the feature point is deleted, a straight line portion of the edge image in which the feature point to be deleted is detected is collectively deleted together with the feature point to be deleted. The vehicle rear side monitoring method according to claim 1, 2 or 3. 5. An image of a rear view of an own vehicle traveling by an imaging means, and movement of a corresponding point in two frames preceding and succeeding a predetermined time is detected as an optical flow, and a rear vehicle or a rear vehicle with respect to the own vehicle is detected. In a vehicle rear side monitoring device that monitors a relative relationship between vehicles traveling in an adjacent lane, a feature point is detected from a captured image at a certain point, and an infinite point of the captured image at the certain point in time among the characteristic points. (FO
E) a feature point selection unit that deletes feature points that are lined up in a straight line through E); and in the image temporally preceding and succeeding the image at a certain time, the feature remaining without being deleted by the feature point selection unit. A point on a straight line connecting a point and an infinity point of the captured image as a candidate point, a corresponding point detecting means for detecting a corresponding point corresponding to the feature point remaining without being deleted, and the feature point and the correspondence Optical flow forming means for detecting a vector connecting points as an optical flow, and determining a degree of approach of a vehicle traveling in a rear vehicle or an adjacent lane with respect to the own vehicle based on the optical flow and the point at infinity to determine a degree of risk. And a danger determining means for determining a vehicle. 6. The feature point selecting means forms a first mask image including peripheral pixels including the feature point, and sets the feature point and the infinity point adjacent to the first mask image. Mask image forming means for forming a second mask image on a straight line passing therethrough; mutual comparing means for comparing the correlation between the first mask image and the second mask image; 6. The rear side monitoring device for a vehicle according to claim 5, further comprising a feature point deleting unit that deletes the feature point when the second mask image has a high correlation. 7. The feature point selecting means includes: an edge image forming means for forming an edge image of the captured image at a certain point in time; and a feature point setting means for setting a predetermined point on an edge in the edge image as a feature point. And forming a first mask image including peripheral pixels including the feature point, and forming a second mask image on a straight line adjacent to the first mask image and passing through the feature point and the point at infinity. A mask image forming means for forming; a mutual comparing means for comparing a correlation between the first mask image and the second mask image; and a strong correlation between the first mask image and the second mask image. 6. The rear side monitoring device for a vehicle according to claim 5, further comprising a feature point deleting unit that deletes the feature point. 8. The feature point deleting means collectively deletes a straight line portion of the edge image in which the feature point to be deleted is detected, together with the feature point to be deleted. The vehicle rear side monitoring device according to claim 7.