JP2012003604A - Mobile body detector and mobile body detection method - Google Patents

Abstract

PROBLEM TO BE SOLVED: To provide a mobile body detector for more reliably detecting a mobile body which moves in every direction.SOLUTION: A mobile body detector includes: a motion vector generation part for taking in a camera image photographed by a camera mounted in a vehicle and generating motion vectors at a plurality of points P of an image; a self vehicle movement parameter estimation part for estimating rotary components (Rx, Ry, Rz) of a self vehicle movement parameter as being equal to inclination from a point P to a vanishing point when correcting the inclination of the motion vectors at the points P is corrected by the rotary components of the self vehicle movement parameter; and a mobile body determination part for correcting the inclination of the motion vector at an arbitrary point Q in the image by using the rotary component of the self vehicle movement parameter, comparing the inclination of the motion vector after correction with inclination of a line joining the arbitrary point Q and the vanishing point, and detecting a mobile body having a different direction from the self vehicle moving direction when the correspondence is small.

Description

  The present invention relates to a moving body detection apparatus and a moving body detection method for capturing an image around a vehicle with a camera mounted on the vehicle and detecting a moving body by analyzing a motion vector.

  2. Description of the Related Art Conventionally, there has been known a moving body detection device in which a plurality of cameras are mounted on a vehicle, the front, rear, or side of the host vehicle is photographed, and a moving body is detected using an optical flow or a motion vector of the photographed image. ing.

  For example, Patent Document 1 discloses a moving body detection apparatus that can determine a stationary object and a moving object even when the camera position moves like an in-vehicle camera. In patent document 1, the motion vector of the observation object in the observed image is extracted, the intersection point on the extension line of the motion vector is calculated, the intersection point (FOE: Focus of Expansion) is obtained, and the temporal position change of the FOE is calculated. A moving body is determined.

  Patent Document 2 discloses an image processing apparatus that removes the influence of road surface unevenness and steering when observing front and rear with an in-vehicle camera and detecting an obstacle. In Patent Document 2, since the distance in the vicinity of the FOE can be regarded as infinite, the translation vector of the camera movement becomes 0 and only the rotation component remains in the motion vector in this area, that is, from the motion vector of the stationary object in the vicinity of the FOE. The shaking amount (rotational component) is obtained and the shaking is corrected.

  In Patent Document 3, on the implicit assumption that a stationary object is dominant in the image, the motion vector predicted from the own vehicle movement parameter is compared with the observed motion vector, and repeatedly matched so as to match. Thus, a moving image analysis method is disclosed in which a subject vehicle motion parameter is obtained, and a subject having a large difference between a motion vector predicted from the subject vehicle motion parameter and a measured motion vector is determined as a moving object.

  Patent Document 4 discloses a moving body detection method and apparatus for calculating an optical flow of a moving image taken from a moving camera and estimating the own vehicle movement parameters (rotation component R, translation component T) by analyzing the optical flow. Is disclosed. In the example of Patent Document 4, when an index (translational component Tz / distance z) that is the reverse distance is defined from the analysis result, the index range is given by setting the assumed range in advance far away, and the measured index value is outside the setting It is determined to be a moving object.

  Patent Document 5 discloses a monitoring device that obtains own vehicle movement information from an external sensor, estimates a background flow using a spatial model, compares it with a measured optical flow, and determines a region that does not match as a moving body. Yes.

  However, since the camera itself moves in a vehicle-mounted application, a moving image is displayed on the screen even if it is a stationary object. For this reason, it is difficult to reliably determine whether the object is a stationary object or a moving object. In Patent Documents 1 to 5, there is a scene dependency, and even if a moving body can be detected in a specific scene, the moving body may not be detected in another scene.

  For example, in Patent Document 1, when a moving body moves on a radial line toward the FOE, the FOE is constant and cannot be detected as a moving body because no temporal change occurs. Further, when the host vehicle rotation is large, the FOE itself cannot be defined and the moving body cannot be detected. In Patent Document 2, if there is no suitable subject in the vicinity of the FOE in the image, the shake cannot be corrected.

  Further, in Patent Document 3, on the premise that a stationary object is dominant as a subject of an image, a movement parameter of the own vehicle (camera) is estimated from a motion vector of the entire screen. In the scene, the vehicle movement parameter cannot be obtained correctly. In Patent Document 4, although a moving body in the direction toward the FOE can be detected, there is a precondition that there is no subject within a certain distance, and therefore it cannot be applied to a camera system that monitors a short distance range near the camera. Furthermore, in Patent Document 5, wheel speed, rudder angle, gyro sensor, yaw sensor, or the like is required in addition to the camera.

JP-A-8-194822 JP 2002-112252 A Japanese Patent Laid-Open No. 5-233813 JP-A-6-203163 Japanese Patent Laid-Open No. 2004-56763

  In a conventional mobile object detection device for in-vehicle use, the camera itself moves, so even on a stationary object, a moving image is displayed on the screen, and it is highly reliable to determine whether it is a stationary object or a moving object. Difficult to do. In Patent Documents 1 to 5, there is a problem that there is a scene dependency, and even if a moving body can be detected in a specific scene, the moving body cannot be detected in another scene.

  In view of such circumstances, it is an object of the present invention to provide a moving body detection apparatus and a moving body detection method that can more reliably detect a moving body that moves in all directions.

  A moving body detection apparatus according to an embodiment of the present invention includes an image input unit that captures a camera image captured by a camera mounted on a vehicle, and an image from the image input unit, and a plurality of points P of the image. When the inclination of the motion vector at the point P is corrected by the rotation component (Rx, Ry, Rz) of the own vehicle movement parameter, the inclination is equal to the inclination from the point P to the vanishing point. A vehicle movement parameter estimation unit for estimating a rotation component (Rx, Ry, Rz) of the own vehicle movement parameter, and an inclination of the motion vector at an arbitrary point Q in the image as a rotation component (Rx , Ry, Rz), and the inclination of the corrected motion vector is compared with the inclination of the line connecting the arbitrary point Q and the vanishing point. Different Direction is detected as the moving body with, characterized by comprising a movable body and determines the moving object determination unit that moves radially toward the stationary or vanishing point when the degree of coincidence is high, the.

  The moving body detection apparatus according to an embodiment of claim 8 includes an image input unit that captures a camera image captured by a camera mounted on a vehicle, an image from the image input unit, and a motion vector of the image. A motion vector generation unit that generates a motion vector at a plurality of points P in the image and a vanishing point at which an extended motion vector obtained by extending the motion vector is concentrated. And an estimated distance weighting coefficient z0 that associates a camera viewing angle with the road surface, and a vehicle movement parameter estimation unit that estimates a rotation component (Rx, Ry, Rz) and a translation component of the vehicle movement parameter. An estimated vehicle translation component is calculated from the average of the translation components multiplied by z0, and the motion vector at an arbitrary point Q in the image is converted into the rotation component (Rx, Ry, Rz) and the translation. Min and predicted from the coefficients z0, characterized by comprising a moving object determination unit determines that the mobile, the if the measured motion match degree as compared to the vector is low.

  A moving body detection method according to an embodiment of the present invention includes a camera image captured by a camera mounted on a vehicle, processes an image from the image input unit, and performs motion vectors of a plurality of points P of the image. When the inclination of the motion vector of the point P is corrected by the rotation component (Rx, Ry, Rz) of the own vehicle movement parameter, the rotation component of the own vehicle movement parameter is assumed to be equal to the inclination from the point P to the vanishing point. (Rx, Ry, Rz) is estimated, the inclination of the motion vector at an arbitrary point Q in the image is corrected using the rotation component (Rx, Ry, Rz) of the own vehicle movement parameter, and the corrected motion When the inclination of the vector is compared with the inclination of the line connecting the arbitrary point Q and the vanishing point, when the coincidence is low, it is detected as a moving body having a direction different from the own vehicle moving direction, and the coincidence is high Stationary objects There is characterized by determining a moving body that moves radially toward the vanishing point.

  The moving object detection method according to the embodiment of claim 15 takes in a camera image taken by a camera mounted on a vehicle, processes an image from the image input unit, and performs motion vectors of a plurality of points P of the image. And a vanishing point at which the extended motion vector obtained by extending the motion vector is concentrated, and a rotation component (Rx, Ry, Rz) and a translation component of the own vehicle movement parameter based on the motion vector and the vanishing point Is set, an estimated distance weighting coefficient z0 associated with the camera viewing angle with the road surface is set, an estimated vehicle translation component is calculated from the average of the translation components multiplied by the coefficient z0, and an arbitrary point Q in the image is calculated. The motion vector is predicted from the rotation component (Rx, Ry, Rz), the translation component, and the coefficient z0. Characterized in that it.

  According to the moving body detection apparatus according to the embodiment of the present invention, it is possible to more reliably detect a moving body that moves in any direction.

The block diagram which shows the structure of the moving body detection apparatus which concerns on 1st Embodiment. The block diagram which shows the detailed structure of the moving body detection part in 1st Embodiment. The flowchart explaining operation | movement of the moving body detection apparatus which concerns on 1st Embodiment. Explanatory drawing which shows the example of selection of the image area by an image area selection part. Explanatory drawing which shows the coordinate system relevant to the production | generation process of a motion vector. The histogram figure for rotation component Ry selection. Explanatory drawing which shows operation | movement of z0 setting part. Explanatory drawing which shows an example of the motion vector which correct | amended the rotation component. Explanatory drawing which shows the other example of the motion vector which correct | amended the rotation component. The block diagram which shows the moving body detection apparatus concerning 2nd Embodiment.

  Hereinafter, a mobile object detection apparatus according to an embodiment of the present invention will be described with reference to the drawings.

  FIG. 1 is a block diagram showing a configuration of a moving object detection apparatus 100 according to an embodiment of the present invention. In FIG. 1, the moving body detection unit 10 processes the image captured by the camera 30 mounted on the vehicle 1 to detect the moving body, and displays the detection result on the display unit 40. The camera 30 is mounted on the vehicle 1 and captures an image around the vehicle. FIG. 1 shows an example in which cameras 30... Are mounted in front and rear of the vehicle 1.

  FIG. 2 is a block diagram illustrating a detailed configuration of the moving object detection unit 10. The moving object detection unit 10 includes a control unit 11 that controls the operation of the moving object detection unit 10, an image input unit 12 that captures a camera image, and an image that selects an image area having image information effective for the detection process of the moving object. An area selection unit 13 and a motion vector generation unit 14 that generates a motion vector of the image area selected by the image area selection unit 13 are included.

  In addition, a stationary object region candidate estimation unit 15 that estimates a stationary object region candidate from an image selected by the image selection region unit 13 (an image in which a stationary object and a moving object are mixed), and when a stationary region candidate cannot be found, A motion scalar determination unit 16 that determines that the region where the motion has occurred is a moving body, and an FOE update unit that performs an update by obtaining FOE (Focus of Expansion) when the vehicle moves without rotating. 17.

  The own vehicle movement parameter estimation unit 18 for estimating own vehicle movement parameters (including rotation components Ωu, Ωv and translational components Tx, Ty, Tz) based on the image in the region selected by the still region candidate estimation unit 15. And a z0 setting unit 19 that sets an estimated distance weighting coefficient z0 as an approximate distance in order to estimate the translation component (Tz, Tx, Ty).

  Furthermore, the mobile body determination unit 20 focusing on the FOE, the translational direction mobile body detection unit 21 that detects the mobile body in the radial direction toward the FOE, the mobile body determined by the motion scalar determination unit 16, and the mobile body determination Information on the mobile body determined by the unit 20 and the mobile body determined by the translational direction mobile body detection unit 21 are input, respectively, and the approach determination unit 22 that determines the mobile body based on the respective logical sum (OR). And have.

  FIG. 3 is a flowchart for explaining the operation of the moving object detection apparatus according to the first embodiment of the present invention. Hereinafter, the operation will be described with reference to related drawings. The moving body detection unit 10 operates under the control of the control unit 11, and the control unit 11 includes a microprocessor including a CPU, a ROM, a RAM, and the like. The control unit 11 generates a control signal in each block so as to perform the operation described below (in FIG. 2, the control signal line is indicated by a dotted line).

  First, in step S1, a camera image captured by the camera 30 is acquired by the image input unit 12. In step S2, the image area selection unit 13 selects an image area having image information effective for the subsequent processing. A feature point extraction method can be used as a method for determining effective image information. For example, as shown in FIG. 4, the ground spreading infinitely in the image converges to a certain point, and this point is called a vanishing line, and an area including the vanishing line is selected. When the vehicle moves without rotating (for example, moves in the direction of arrow A), a plurality of motion vectors in the image are extended, and a point where the extended motion vectors are concentrated is a vanishing point (FOE).

  In the next step S3, the motion vector generation unit 14 generates a motion vector of the region selected by the image region selection unit 13. As the motion vector generation process, a gradient method, a block matching method, or the like can be used.

  For example, as shown in FIG. 5, an xyz coordinate system in which the optical axis of the camera 30 is the z-axis, the horizontal axis to the road surface is the x-axis, and the vertical axis to the road surface is the y-axis is shown. . When Oc (0, 0, 0) is the camera center and the focal length is f, the projection plane of z = f (f is the focal length) is denoted by 200, and the point P ′ of the world coordinates is a point of this projection plane 200 Projected onto P. A coordinate system fixed to the ground of the world coordinate system is indicated by Ow.

The motion vector V (u, v) on the screen P (x, y) accompanying the movement of the vehicle can be expressed by the following well-known expressions (1) and (2).

Here, the own vehicle movement parameters are assumed to be rotational components Ωu, Ωv and translational components Tx, Ty, Tz. Tx, Ty, and Tz represent translational speeds in the x-axis, y-axis, and z-axis directions, respectively. Furthermore, Ωu and Ωv are represented by Rx, Ry, and Rz according to the following equations (3) and (4). Rx, Ry, and Rz represent rotational components around the x-axis, y-axis, and z-axis.

As is well known, when the inclination is obtained by correcting the rotational component of the motion vector V (u, v) at the point P (x, y), it becomes the same as the inclination from P (x, y) to the FOE. That is, the following equation (5) is established.

The image selected by the image selection region unit 13 (an image in which a stationary object and a moving object are mixed) is input to the stationary object region estimation unit 15, and in step S4, the stationary object region is estimated. In other words, Rx and Rz are often very small for in-vehicle cameras.
Rx = Rz = 0
(6) and (7) are obtained by approximation.

  From the expressions (5), (6), and (7), Ry is obtained for each pixel P... From the point P (x, y), the motion vector V (v, u), and FOE (x0, y0). Ry is common to stationary objects. The plurality of Ry are obtained for a plurality of image points, a histogram of Ry as shown in FIG. 6 is generated, and a region having a common Ry near the peak of the histogram is selected as a stationary object region candidate.

However, when the vehicle speed is high, Rz is generated. At this time,
Rx = 0
(8) and (9) are obtained by approximating

Ry and Rz can be obtained from a plurality of nearby motion vectors. At this time, a histogram of Ry and Rz is generated, and a region having common Ry and Rz in the vicinity of the peak of the histogram is selected as a stationary object region candidate.

  When a still area candidate cannot be found (NO in step S5), it is regarded as a full screen motion or a full screen still, and the motion scalar determination unit 16 determines a region where the motion has occurred as a moving body. That is, the motion scalar determination unit 16 determines the magnitude of the motion (scalar amount) in step S6, and determines that it is a motion region when the scalar amount exceeds a preset threshold value in step S7 (when YES). . The vector quantity is expressed by the direction and the magnitude, and only the magnitude is called a scalar or a scalar quantity. Even when the area of the stationary object is small on the screen, if the peak of the Ry histogram of the stationary object area is found, that area can be determined as the stationary object area.

  When the vehicle moves without rotating in step S8, the FOE update unit 17 obtains the FOE as an extension line intersection of the motion vectors and performs an update. In-vehicle cameras generate translational motions only in the direction of the vehicle center axis, so the vanishing point (FOE) seems to be unchanged, but the optical axis due to a large number of passengers and load, or due to secular changes in the mounting position of the camera 30, etc. Therefore, the optical axis shift of the camera 30 appears as a vanishing point (FOE) shift. Therefore, the FOE is updated in step S8.

That is, in the stationary object region candidate, when the rotation component = 0, and the motion vector of the point P (x, y) is (un, vn), the point on the extension line of the motion vector starting from P Is (x0, y0), the following equation (10) is obtained.

When the n points P are expressed in a matrix, equation (11) is obtained.

  Then, the vanishing point (x0, y0) is obtained from the plurality of points P using the least square method, and is updated as a new vanishing point. Therefore, although vanishing point update is limited when the vehicle is performing translational movement, vanishing point fluctuations do not occur frequently, so there is no practical problem. Since the attachment angle of the camera 30 can be obtained from the vanishing point, the vanishing point update data is sent to the z0 setting unit 19 described later and used as calibration data.

If the optical axis of the camera 30 is installed at a depression angle θ downward with respect to the road surface, the translation speeds Tz and Ty in the z-axis and y-axis directions when the host vehicle translation speed T is
Tz = Tcosθ, Ty = Tsinθ
It becomes. The vanishing point (x0, y0) is
x0 = fTx / Tz, y0 = fTy / Tz
Can be expressed as
y0 = f · tanθ
Thus, the camera depression angle can be obtained from the vanishing point. Similarly, regarding the angle φ between the vehicle center line and the optical axis when considered in the x-axis direction,
x0 = f · tanφ
Can be obtained.

In step S9, the own vehicle movement parameter estimation unit 18 uses the n P (xn, yn) motion vectors V (un, vn) in the stationary object region selected by the stationary region candidate estimation unit 15. The vehicle movement parameters Rx, Ry, Rz are estimated. That is, the following equation is obtained for n P (xn, yn) from the equations (1), (2), (3), and (4).

  From the equation (12), the rotational components Rx, Ry, Rz of the own vehicle movement parameter are estimated by the least square method using singular value decomposition or the like. The translation component obtained from the equations (1) to (4) using the estimated rotation component can be obtained only in the form of values Tx / z, Ty / z, Tx / z multiplied by the reciprocal of the distance z to the subject. In step S10, the z0 setting unit 19 sets an estimated distance weighting coefficient z0 as an approximate distance in order to estimate the translational speeds Tz, Tx, Ty in the x-axis, y-axis, and z-axis directions.

  That is, as shown in FIG. 7, the coefficient z0 is set in association with the camera viewing angle (downward depression angle θ) corresponding to the road surface distance from the camera 30. In the example of FIG. 7, z0 = 1 is set for an image having a viewing angle corresponding to a road surface distance of 1 to 1.5 m. Similarly, z0 = 1.5 for a viewing angle image corresponding to a road surface distance of 1.5 to 2 m, z0 = 2 for a viewing angle image corresponding to a road surface distance of 2 to 3 m, and a road surface distance of 3 to 4 m. Z0 = 3 for a viewing angle image and z0 = 4 for a viewing angle image corresponding to a road surface distance of 4 m to infinity m. If the angle θ is known, the distance from the subject vehicle to the subject M can be known.

  The camera viewing angle and each pixel on the camera image correspond to 1: 1. Therefore, the estimated distance weighting coefficient z0 associated with the camera viewing angle with the road surface may be obtained in advance for each pixel and set in an LUT (look-up table) or the like.

  Although the procedure for discretely setting z0 in advance has been described with reference to FIG. 7, it may be set continuously in association with the camera viewing angle as z0 = 1 / z. In this case, since there is a limit to the resolution of the camera, z0 = fixed value above a certain z.

The own vehicle movement parameter estimation unit 18 sets the distance z0 in association with the camera viewing angle corresponding to the road surface distance from the camera in this way,
Tz = z0 (.Tz / z), Tx = z0 (.Tx / z), Ty = z0 (.Ty / z)
And the estimated translational components Tzm, Tym, and Txm are calculated by averaging a plurality of points.

On the other hand, the movement detection unit 20 focusing on FOE performs the following processing in step S11. That is, for an arbitrary point Q (x, y) in the image, the own vehicle movement parameters Rx, Ry, Rz are substituted into the equations (1) to (4), and the inclination after motion correction, the equation (13) is obtained. .

The slope of the straight line from Q (x, y) to FOE (x0, y0) is expressed by equation (14).

  As a result of comparing the slopes (13) and (14) in step S11, if the degree of coincidence is low (NO in step S12), the rotation amount is corrected as shown in FIG. 8 (with rotation components Ωu and Ωv). Since the corrected motion vector extension line does not pass through the FOE, it is determined that the moving direction is different from the moving direction of the vehicle.

  Since the camera shake component appears as the own vehicle rotation components Rx and Rz, as shown in the equation (13), by determining after correcting the rotation amount, the shake accompanying the own vehicle movement on the gravel road or steps is Even if a vehicle rotation component (mainly Rx, Rz) is generated, it is corrected, so that it is not affected by the vehicle shake.

  If the degree of coincidence is high by comparing the slopes (13) and (14) (when YES in step S12), the extension line of the motion vector with the rotation amount corrected passes through the FOE as shown in FIG. In this case, not only a stationary object but also a moving body that moves radially from the FOE may be included. Therefore, the translational direction moving body detection unit 21 performs the following operation.

  Information not detected as a moving body by the moving body determination unit 20 focusing on FOE and the estimated own vehicle translation component Tzm obtained by the own vehicle movement parameter estimation unit 18 are input to the translation direction moving body detection unit 21. . In step S13, the translational direction moving body detection unit 21 obtains Tz / z of the point Q (x, y) from the equations (1) to (4) and the rotation components Rx, Ry, Rz, and z0 (Tz / z) The average estimated translation component Tzm is compared, and if the degree of coincidence exceeds a preset threshold value in step S14, it is determined as a moving body. That is, if there is something that moves faster than the surroundings, it is determined as a moving body that comes toward the vehicle.

  As can be seen from the equations (1) to (4), when the distance is long (z is large), the translation component Tz / z is small, so the motion vector is small, and the short distance (z is small). Since the translation component Tz / z sometimes becomes large, the motion vector becomes large. Therefore, it is difficult to detect a moving object at a long distance, and it is easy to erroneously detect a stationary object at a short distance. However, by introducing the estimated distance weighting coefficient z0, it becomes easy to detect a moving object at a long distance, and it is difficult to erroneously detect a stationary object at a short distance.

  In step S15, the approach determination unit 22 determines the moving body determined by the motion scalar determination unit 16 (determination result in step S7) and the moving body determined by the mobile body determination unit 20 focusing on FOE (determination in step S12). Result) and information on the moving body (determination result in step S14) determined by the translational direction moving body detection unit 21 are input, and the moving body is determined by taking the respective logical sums (OR). Also, since the direction of the motion vector is known for those that are determined to be moving objects, the approaching direction of the moving object is determined, and the moving object that has a direction approaching the host vehicle is preferentially output as having a high degree of danger. If there is a moving body approaching the host vehicle, a warning is displayed on the display unit 40.

  As described above, in the first embodiment, a moving body in the direction toward the FOE can be detected, and a moving body moving in any direction can be detected. In addition, stable moving object detection can be performed even when the number of passengers and the load is large, and stable moving object detection can be performed without being affected by own vehicle vibration (camera shake) such as a gravel road.

  Next, a moving body detection apparatus according to a second embodiment will be described with reference to FIG. In the second embodiment, the moving body determination unit 20 focused on FOE is replaced with a motion vector prediction unit 23, and the translational direction moving body detection unit 21 is replaced with a moving body determination unit 24.

  Similar to the first embodiment, the own vehicle movement parameter estimation unit 18 outputs rotation components Rx, Ry, Rz and average translation components Tzm, Txm, Tym. The motion vector prediction unit 23 uses Tzm / z0, Txm / z0, and Tym / z0 instead of Tz / z, Tx / z, and Ty / z for the motion vector of the point Q (x, y), and (1) to The prediction vectors u ′ and v ′ are calculated based on the equation (4).

  The movement detection unit 24 compares the measured motion vectors u and v with the predicted motion vectors u ′ and v ′, and determines that the object is a moving object when the degree of coincidence is low. In the case of an in-vehicle camera, Tx and Ty are often very small, so Tx / z, Ty / z and Txm, Tym may be omitted.

  It should be noted that the present invention is not limited to the embodiments described above, and various modifications can be made without departing from the scope of the claims.

DESCRIPTION OF SYMBOLS 100 ... Moving body detection apparatus 10 ... Moving body detection part 11 ... Control part 12 ... Image input part 13 ... Image area selection part 14 ... Motion vector generation part 15 ... Still area candidate estimation part 16 ... Motion scalar determination part 17 ... FOE update Unit 18: Own vehicle movement parameter estimation unit 19 ... z0 setting unit 20 ... moving object determination unit (focus on FOE)
DESCRIPTION OF SYMBOLS 21 ... Translation direction moving body detection part 22 ... Approach determination part 23 ... Motion vector prediction part 24 ... Moving body determination part 30 ... Camera 40 ... Display part

Claims (15)

An image input unit that captures a camera image taken by a camera mounted on the vehicle;
A motion vector generation unit that processes an image from the image input unit and generates motion vectors of a plurality of points P of the image;
When the inclination of the motion vector of the point P is corrected by the rotation component (Rx, Ry, Rz) of the own vehicle movement parameter, the rotation component (Rx, Ry) of the own vehicle movement parameter is assumed to be equal to the inclination from the point P to the vanishing point. , Rz), the own vehicle movement parameter estimation unit,
The inclination of the motion vector at an arbitrary point Q in the image is corrected using the rotation components (Rx, Ry, Rz) of the own vehicle movement parameter, the inclination of the corrected motion vector, and the arbitrary point Q Compare the inclination with the line connecting the vanishing points, detect when the coincidence is low as a moving body with a direction different from the moving direction of the host vehicle, and move radially toward the stationary object or vanishing point when the coincidence is high A moving body determination unit that determines a moving body to perform;
A moving body detection apparatus comprising:   2. The mobile object detection apparatus according to claim 1, wherein when the rotation component is 0, the own vehicle movement parameter estimation unit updates an intersection of extension lines of motion vectors of the plurality of points P as the vanishing point. . A stationary region candidate estimation unit that estimates a stationary object region based on motion vectors of a plurality of points P of the image,
The own vehicle movement parameter estimation unit estimates a rotation component (Rx, Ry, Rz) of the own vehicle movement parameter based on a motion vector of a plurality of points P of the estimated stationary object region image and the vanishing point. The moving body detection device according to claim 1.   The stationary region candidate estimation unit obtains at least Ry of rotation components Ry and Rz depending on the steering angle of the vehicle from the motion vectors of the plurality of points P of the estimated stationary object region image and the vanishing point, and the rotation of each point 4. The moving object detection apparatus according to claim 3, wherein the vicinity of the peak of the histogram relating to the component Ry (or Ry and Rz) is selected as a candidate for a stationary object region.   Input image information that has not been detected as a moving object by the moving object detection unit, and translate the vehicle movement parameter from the motion vector and rotation components (Rx, Ry, Rz) of an arbitrary point Q in the stationary object region candidate The component Tz / z is obtained, the estimated distance weighting coefficient z0 associated with the camera viewing angle with the road surface is set, and the translation component z0 (Tz / z) multiplied by the coefficient z0 is compared with the estimated own vehicle translation component Tzm. 4. A moving body detecting apparatus according to claim 3, further comprising a translational direction moving body detecting unit that detects the moving body when the degree of coincidence is low.   The translation-direction moving body detection unit obtains a translation component Tz / z of the own vehicle movement parameter from the motion vectors and rotation components (Rx, Ry, Rz) of a plurality of points P in the stationary object region candidate, and calculates the coefficient z0. 6. The moving body detection device according to claim 5, wherein an average of the multiplied translation component z0 (Tz / z) is set as the estimated own vehicle translation component Tzm.   The translational direction moving body detection unit presets, as the estimated distance weighting coefficient z0, a correspondence relationship in which a distance on a road surface is associated with each pixel position of the camera image. Mobile object detection device. An image input unit that captures a camera image taken by a camera mounted on the vehicle;
A motion vector generation unit that processes an image from the image input unit and generates a motion vector of the image;
A vanishing point at which a motion vector of a plurality of points P in the image and an extended motion vector obtained by extending the motion vector are concentrated is obtained, and a rotation component of the own vehicle movement parameter based on the motion vector of the plurality of points P and the vanishing point (Rx, Ry, Rz) and the own vehicle movement parameter estimation unit for estimating the translation component;
An estimated distance weighting coefficient z0 that associates the camera viewing angle with the road surface is set, an estimated vehicle translation component is calculated from the average of the translation components multiplied by the coefficient z0, and a motion vector at an arbitrary point Q in the image is calculated. A moving object determination unit that predicts from the rotation component (Rx, Ry, Rz), the translation component, and the coefficient z0, and determines a moving object when the degree of coincidence is low compared to the actually measured motion vector;
A moving body detection apparatus comprising:   9. The moving body detection apparatus according to claim 8, wherein a correspondence relationship in which a distance on a road surface is associated with each pixel position of the camera image is set in advance as the estimated distance weighting coefficient z0. Capture the camera image taken by the camera mounted on the vehicle,
Processing an image from the image input unit to generate motion vectors of a plurality of points P of the image;
When the inclination of the motion vector of the point P is corrected by the rotation component (Rx, Ry, Rz) of the own vehicle movement parameter, the rotation component (Rx, Ry) of the own vehicle movement parameter is assumed to be equal to the inclination from the point P to the vanishing point. , Rz),
The inclination of the motion vector at an arbitrary point Q in the image is corrected using the rotation components (Rx, Ry, Rz) of the own vehicle movement parameter, the inclination of the corrected motion vector, and the arbitrary point Q Compare the inclination with the line connecting the vanishing points, detect when the coincidence is low as a moving body with a direction different from the moving direction of the host vehicle, and move radially toward the stationary object or vanishing point when the coincidence is high A moving body detection method, characterized in that the moving body is determined to be a moving body.   11. The moving object detection method according to claim 10, wherein when the rotation component is 0, an intersection of extension lines of motion vectors of the plurality of points P is updated as the vanishing point. A stationary object region is estimated based on motion vectors of a plurality of points P in the image,
From at least the motion vectors of the plurality of points P of the estimated stationary object region image and the vanishing point, obtain at least Ry of rotation components Ry and Rz depending on the steering angle of the vehicle,
11. The moving object detection method according to claim 10, wherein the vicinity of a peak of a histogram relating to the rotation component Ry (or Ry and Rz) of each point is selected as a candidate for the stationary object region. Input image information that was not detected as the moving body,
A translation component Tz / z of the own vehicle movement parameter is obtained from a motion vector of an arbitrary point Q in the stationary object region candidate and the rotation component (Rx, Ry, Rz),
When the estimated distance weighting coefficient z0 associated with the camera viewing angle with the road surface is set, the translation component z0 (Tz / z) multiplied by the coefficient z0 is compared with the estimated own vehicle translation component Tzm, and the degree of coincidence is low The moving body detecting method according to claim 12, wherein the moving body is detected as a moving body.   A translation component Tz / z of the own vehicle movement parameter is obtained from the motion vectors of a plurality of points P in the stationary object region candidate and the rotation components RX, Ry, Rz, and is multiplied by the coefficient z0 (Tz / z). The moving body detection method according to claim 13, wherein an average of the vehicle is the estimated translation component Tzm of the subject vehicle. Capture the camera image taken by the camera mounted on the vehicle,
The image from the image input unit is processed to generate motion vectors at a plurality of points P of the image, and a vanishing point at which extended motion vectors obtained by extending the motion vector are concentrated is determined. The motion vector and the vanishing point Based on the above, the rotational component (Rx, Ry, Rz) and translational component of the own vehicle movement parameter are estimated,
An estimated distance weighting coefficient z0 that associates the camera viewing angle with the road surface is set, an estimated vehicle translation component is calculated from the average of the translation components multiplied by the coefficient z0, and a motion vector at an arbitrary point Q in the image is calculated. A moving body detecting method, wherein the moving body is predicted based on the rotation components (Rx, Ry, Rz), the translation component, and the coefficient z0, and is determined to be a moving body when the degree of coincidence is low compared to the actually measured motion vector.

Images