JP2009169619A - Image-processing device for vehicle and image-processing method - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To precisely identify a three-dimensional object without an influence of image distortion, even if image conversion is made, in an image-processing device which processes images around a vehicle. <P>SOLUTION: Edges are extracted in top view images which are converted from images photographed at some time intervals, and the straight line parameters (a tilt θ, an intercept b) of each edge element are casted on the θ-b plane which is a casting space, and the movement of noteworthy areas having high casting frequency values is observed in time sequence. The noteworthy areas Gr1 and Gr2, which move in time sequence, are determined to belong to a three-dimensional object, and the noteworthy areas Gh1 and Gh2, which do not move, are determined to belong to a plane such as a road surface. <P>COPYRIGHT: (C)2009,JPO&INPIT

Description

  The present invention relates to a vehicle image processing apparatus and an image processing method for detecting a three-dimensional object by identifying it as a road surface from an image around the vehicle.

Conventionally, a vehicle driving support device that detects a three-dimensional object such as another vehicle or a road facility from an image around the vehicle captured by an in-vehicle camera and displays it on a display or the like is disclosed in, for example, Japanese Patent Application Laid-Open No. 2007-172501. It is disclosed.
In this vehicle driving support device, a plurality of images obtained by capturing horizontal side views including a road surface at different positions are converted into top view images in which the viewpoint positions are set above the vehicle, and objects having the same characteristics in the overlapping region Image processing for detecting a three-dimensional object is performed based on the correspondence between the two different feature points.
JP 2007-172501 A

  However, in the image processing in the above-described conventional apparatus, since a three-dimensional object is detected from the correspondence between two different feature points on the top view image, the image captured from the side view by the camera is converted to the top view image. Due to the influence of the image distortion caused by this, the correspondence between the two feature points cannot be taken, and there is a risk that the detection of the three-dimensional object will fail.

  In view of the above problems, the present invention provides a vehicle image processing apparatus and an image processing method that can accurately detect a three-dimensional object without being affected by image distortion even when image conversion is performed. For the purpose.

  The present invention acquires images around the vehicle in chronological order, generates a bird's-eye view image looking down from the viewpoint position above the host vehicle, extracts feature points, and votes the parameters to the voting space, Based on the time-series state of the point cloud as the voting result, it is determined whether or not the feature point belongs to a three-dimensional object.

  According to the present invention, since a three-dimensional object can be discriminated from, for example, movement of a point cloud in a voting space, even if image distortion occurs due to image conversion from a side view captured image to a top view or the like, it is not affected. A three-dimensional object can be detected with high accuracy.

Next, embodiments of the present invention will be described by way of examples.
FIG. 1 is a block diagram showing the configuration of the first embodiment.
The image processing apparatus 1 includes an image memory 3 connected to the camera 2, a converted image creation unit 4, an edge extraction unit 5, an edge parameter calculation unit 6, a parameter voting unit 7, and a region of interest extraction sequentially connected to the image memory 3. Part 8 and plane / solid discriminating part 9. The converted image creation unit 4, edge extraction unit 5, edge parameter calculation unit 6, parameter voting unit 7, attention area extraction unit 8, and plane / solid determination unit 9 are also connected to the work memory unit 10.

  For example, the camera 2 includes a CCD image pickup device and is installed in the vehicle V so as to be able to pick up an image of the periphery of the vehicle including the road surface, and outputs the picked-up image data to the image memory 3 of the image processing apparatus 1. The installation position of the camera 2 depends on the surrounding area where the surrounding environment is to be grasped. For example, when the area around the parking frame in the parking lot is the target, as shown in FIG. Is set as the imaging range S, but in addition, a plurality of units may be provided as appropriate in appropriate parts such as the front part of the vehicle, the rear part of the vehicle, or each corner part according to the grasping object. 2A is a side view of the vehicle, and FIG. 2B is a top view.

The image memory 3 is composed of a RAM (Random Access Memory), and captures and stores image data from the camera 2 at predetermined time intervals.
The converted image creation unit 4 converts the side view image based on the image data stored in the image memory 3 into a top view image in which the viewpoint position is set above the vehicle and the vehicle periphery is looked down from the viewpoint.
In addition, when each direction around the vehicle is imaged with a plurality of cameras, the converted image creation unit 4 also performs synthesis into one image.

The edge extraction unit 5 extracts vertical and horizontal edges as feature points of the top view image using a general Sobel filter. Here, the vertical differential value (dx) and the horizontal differential value (dy) of the edge are obtained.
The edge parameter calculation unit 6 obtains a linear equation for each pixel (hereinafter referred to as an edge element) extracted as an edge, and calculates its slope and intercept as an edge element parameter.

  Here, in the parameter calculation, the edge parameter calculation unit 6 corrects the change in the image coordinates due to the movement of the vehicle while the image data from the camera 2 is captured at the predetermined time interval. In other words, the position and orientation of the vehicle in one image are defined as absolute coordinates, and when the vehicle moves, the movement and orientation changes are taken into account, and the slope and intercept of the edge element in the next image are Calculate on the coordinates.

The parameter voting unit 7 votes the edge element parameters of each edge element in the voting space. The voting space here includes a plane having the slope and intercept of the straight line as the horizontal axis and the vertical axis.
The attention area extraction unit 8 extracts an attention area having a high voting frequency value from the voting result of the edge element parameters for one frame image.
The plane / solid discriminating unit 9 compares the regions of interest in time series, and determines whether the edge extracted by the edge extracting unit 5 belongs to a pattern drawn on a plane such as a road surface or a solid object. to decide.

The work memory unit 10 is composed of a RAM similar to the image memory 3, stores the top view image converted by the converted image creation unit 4, and calculates by the edge and edge parameter calculation unit 6 extracted by the edge extraction unit 5. The obtained edge element parameters, the voting space used by the parameter voting unit 7, the voting result thereof, the extraction result of the attention area extracting unit 8, the discrimination result of the plane / solid discriminating unit 9, and the like are temporarily stored for work of each unit. In the flowchart described below, the storage processing in the work memory unit 10 by each unit is not particularly mentioned.
The converted image creation unit 4, the edge extraction unit 5, the edge parameter calculation unit 6, the parameter voting unit 7, the attention area extraction unit 8, and the plane / solid determination unit 9 can be configured as a CPU in which the respective functions are integrated.

Next, the flow of determination processing for determining whether or not the image processing apparatus 1 is a three-dimensional object will be described with reference to the flowchart of FIG.
First, in step 100, image data from the camera 2 is fetched into the image memory 3 for each frame at a predetermined time interval and stored.
In step 101, the converted image creation unit 4 reads out the image data from the image memory 3 and converts it into a top view image.
Subsequently, in step 102, the edge extraction unit 5 extracts an edge in the top view image. Here, by using the Sobel filter, the vertical differential value (dx) and the horizontal differential value (dy) of the edge are obtained.

In step 103, the edge parameter calculation unit 6 calculates an edge element parameter for each extracted edge element.
In other words, the angle (slope) of the edge element uses the differential value of each direction of the edge,
θ = tan ⁻¹ (dy / dx)
Is required. Therefore, with respect to the linear equation y = ax + b of the edge element (pixel), the position (x, y) of the edge element is known, and since the coefficient a can be calculated from θ, the intercept b can also be calculated. An edge element parameter consisting of a set of inclination θ and intercept b is obtained for one edge element.

In step 104, the parameter voting unit 7 votes the edge element parameters of each edge element in the voting space.
Here, the edge element parameter (θ, b) of one edge element is plotted as one point on the θ-b plane as the voting space T.
For example, when a parking frame W as shown in FIG. 4A is obtained as a top view image, an edge image as shown in FIG. 4B is obtained. The edge element parameters of the edge element of the E1 portion in the edge image become the voting point group of F1 in the voting space T of (c). Similarly, the E2 and E3 portions of the edge image of (b) are the voting space T of (c). Are plotted as voting point groups of F2 and F3, respectively.
Note that the θ-b plane of the voting space T is not particularly illustrated, but for example, the horizontal axis θ is in increments of 1 ° from 0 ° to 360 °, and the vertical axis b is divided into meshes in increments of 3 pixels.

In step 105, the attention area extraction unit 8 extracts a high-frequency value area having a high density of edge element parameter voting points in the voting space T as the attention area. That is, a mesh region having a plot number higher than a predetermined threshold (for example, 100 points) is extracted as a region of interest.
In step 106, the plane / solid determination unit 9 checks whether the extraction result of the attention area for the previous top view image is stored in the work memory unit 10.
When the previous extraction result of the attention area is not stored in the work memory unit 10, the process returns to step 100, new image data is taken from the camera 2, and the above-described processing is repeated.

If there is a previous extraction result of the attention area, in step 107, the plane / solid determination unit 9 determines whether the edge corresponding to the attention area belongs to a solid object or a road surface.
Specifically, the previous attention area in the voting space T is compared with the current attention area to check the relative positional relationship.
For example, if it is a paint of a parking frame drawn on a plane such as a road surface, even if the vehicle moves with time, as shown in FIG. The attention areas Gh1, Gh2, etc. hardly move in the voting space T. In the figure, a white circle point group represents the previous attention area point group, and a black circle point group represents the current attention area point group.

On the other hand, in the case of a three-dimensional object, since the appearance direction and position of the edge change when the three-dimensional object moves or the vehicle moves and the relative relationship changes, as indicated by an arrow in FIG. Furthermore, the positions in the voting space such as the attention areas Gr1 and Gr2 move.
Therefore, even when based on an image captured by one camera 2, an edge belongs to either a three-dimensional object or a plane depending on the position change state of the attention area in the voting space T in time series (that is, the previous time and the current time). Can be judged.

  Thereafter, in step 108, together with the top view image stored in the work memory unit 10 or the like, the determination result as to whether or not the object applied to the edge is a three-dimensional object is output to a subsequent device such as a driving support device, and one determination process is performed. End. Note that the previous view image or the current image is appropriately selected as the top view image to be output in accordance with the purpose of the subsequent apparatus.

In the present embodiment, steps 100 and 101 in the flowchart of FIG. 3 constitute the image acquisition means in the invention, and step 102 corresponds to the feature point extraction means.
Steps 103 and 104 constitute a voting means, and in particular, step 103 corresponds to an edge element parameter calculation means. Steps 105 to 107 constitute discrimination means.

  The present embodiment is configured as described above, and creates a top view image based on each side view image captured with the camera 2 shifted in time, extracts an edge from the top view image, and votes the parameters of each edge element. Since voting on the space T and determining whether the edge belongs to the three-dimensional object based on the time-series state of the voting point group in the voting space, the top view image is converted from the side view image. 3D objects can be extracted with high accuracy without being affected by distortion caused by image conversion.

In particular, when the edge element parameters are set to the slope θ and intercept b of the straight line, one voting point is generated from one edge element, so that the voting amount can be greatly reduced as compared with Hough transform or the like. The processing speed can be increased.
Further, since the determination as to whether or not the object is a three-dimensional object is performed based on the state of the attention area in the voting space T, which has a particularly high voting frequency value, the processing is quick in this respect as well.
Also, the slope θ and the intercept b of the edge element parameters in the two images are calculated on the absolute coordinates corrected for the change in the image coordinates due to the movement of the vehicle. When the edge corresponding to the area belongs to the three-dimensional object and the attention area has not moved, it can be easily determined that the edge corresponding to the area belongs to a plane pattern such as a road surface.

Next, a second embodiment will be described.
FIG. 6 is a block diagram showing the configuration of the second embodiment.
The image processing apparatus 1A of the present embodiment is different from the first embodiment in the edge parameter calculation unit 6A, the parameter voting unit 7A, and the plane / solid determination unit 9A, and a grouping calculation unit instead of the attention area extraction unit. 11 is different.
The edge parameter calculation unit 6A switches between taking the intercept of the straight line on the x-axis and the y-axis depending on the inclination θ of the edge element.

Here, as a basic concept, as shown in FIG. 7, the inclination of the edge belonging to the range α in the horizontal direction + −45 ° and the inclination of the edge belonging to the range β in the vertical direction + −45 °, depending on the direction from the screen center Q. And classify. The slope was calculated linear equation y = a _{x x-b} x from the edge element parameter (θ, b _x) as sections y-axis in the case of the range α as .theta.1, ranging as slope .theta.2 beta for linear equation _x = a _{y y-b} y from the edge element parameters in the x-axis as the intercept (θ, _{b y)} is calculated.
According to this, as shown in FIG. 8, the area surrounded by the straight line passing through + -45 ° corners (corners) of the screen K is the intercept range of voting space T, the intercept b _x, b _y the relevant It will be limited to the area.
Note that the edge element parameters are calculated on absolute coordinates obtained by correcting changes in image coordinates due to vehicle movement, as in the edge parameter calculation unit of the first embodiment.

The parameter voting unit 7A sets two voting spaces T1 and T2 corresponding to the classification of the slope of the edge into two ranges, and votes the edge element parameters of each edge element to the corresponding voting space.
Note that, in an area where the inclination θ that divides the voting space into two is near + −45 °, the voting values may be dispersed in the two voting spaces T1 and T2, so in this embodiment, as shown in FIG. In addition, an overlapping region δ of + −5 ° is created across an angle line (+ −45 °) that divides the two ranges in the basic concept described above. That is, when the range α is enlarged and specifically, A in the horizontal direction + −50 °, and the range β is also set to B in the vertical direction + −50 °, and the edge inclination θ appears in the overlapping region δ. Causes the edge element parameters to be voted on to both voting spaces T1, T2. In this case, the intercept b _x, the range of b _y will be expanded to slightly outward from the region shown in FIG.

  The grouping calculation unit 11 extracts a high-frequency area in which the voting frequency of the edge element parameter is higher than a predetermined threshold in each of the voting spaces T1 and T2, and groups the high-frequency value areas adjacent to the surrounding areas as the same area. .

The plane / solid discriminator 9A compares each group in time series in each voting space as in the plane / solid discriminator of the first embodiment, and the edges belonging to the group are drawn on the road surface. Whether it belongs to 3D or 3D.
The plane / solid determination unit 9A further determines whether the solid object is moving or stationary by using a representative value representing the group determined to belong to the solid object.

The work memory unit 10 stores the top view image converted by the converted image creation unit 4, and the edge extracted by the edge extraction unit 5, the edge element parameter calculated by the edge parameter calculation unit 6A, and the parameter voting unit 7A. The voting space to be used, the voting result thereof, the grouping result by the grouping calculation unit 11, the representative value of the group calculated by the plane / solid discrimination unit, the discrimination result, and the like are temporarily stored for work of each unit.
The converted image creation unit 4, the edge extraction unit 5, the edge parameter calculation unit 6A, the parameter voting unit 7A, the grouping calculation unit 11, and the plane / solid determination unit 9A can be configured as a CPU in which the respective functions are integrated.
Other configurations are the same as those of the first embodiment.

In the following, the flow of discrimination processing in the image processing apparatus 1A of the second embodiment will be described with reference to the flowchart of FIG.
Steps 200 to 202 are the same as steps 100 to 102 in FIG. 3 in the first embodiment.
In step 203, the edge parameter calculation unit 6A calculates the gradient θ for each extracted edge element, and the linear equation y = a depending on whether the gradient θ belongs to the range A or B shown in FIG. Edge element parameters (θ, b _x ) or (θ, b _y ) are calculated from _x x−b _x or x = a _y y−b _y .

In step 204, the parameter voting unit 7A votes each edge element parameter calculated by the edge parameter calculating unit 6A in one or both of the two voting spaces T1 and T2 corresponding to the ranges A and B of the inclination θ.
Next, in step 205, the grouping calculation unit 11 extracts high-frequency regions of edge element parameters in each voting space, and groups adjacent high-frequency value regions in step 206.

Specifically, a maximal region in which the voting frequency of the edge element parameter is particularly high compared to the surrounding region is extracted from the entire voting spaces T1 and T2, and the voting frequency of the maximal region is determined within a predetermined distance. An area having a voting frequency of 50% or more is integrated into one group. As a result, a group having the maximum region as a nucleus is formed.
Each group is represented by one representative parameter such as its center of gravity.

In step 207, the plane / solid determination unit 9 </ b> A checks the grouping result in each voting space for the previous top view image, that is, whether the group is stored in the work memory unit 10.
When the previous group is not stored, the process returns to step 200, new image data is taken from the camera 2, and the above-described processing is repeated.

  If there is a previous group, the process proceeds to step 208, and the plane / solid determination unit 9A determines the previous group and the current group in each of the voting spaces T1 and T2 as in the comparison of the attention area in the first embodiment. In comparison, if the representative parameter (centroid) is moving, the edge corresponding to the group belongs to the solid object, and if the position of the representative parameter does not change, the edge corresponding to the group belongs to the plane. Determine.

In the next step 209, the plane / solid discrimination unit 9A calculates a straight line (inclination and intercept) as a representative value of the group of edges determined to belong to the solid object. The representative value of the group is obtained by, for example, line fitting by the least square method using the position (x, y) of the edge element of the entire group.
In step 210, the plane / solid determination unit 9 </ b> A checks whether or not the group representative values for the top view images for the past two times are stored in the work memory unit 10.
When the representative values for the past two times are not stored in the work memory unit 10, the process returns to step 200, new image data is taken from the camera 2, and the above-described processing is repeated.

When there are representative values for the past two times, in step 211, based on the movement state on the absolute coordinates of the three straight lines (representative values) including the amount calculated this time by the plane / solid determination unit 9A. Determine whether an object is moving or stationary.
Three straight lines that are representative values obtained from a stationary object are shown in FIG. 10, for example, and three straight lines obtained from a moving object are shown in FIG. Each of these shows changes in the time series of image capturing. In FIG. 10, the straight line moves in the order of L1, L2, and L3 over time, and in FIG. 11, the straight line moves in the order of L4, L5, and L6. is doing.
Focusing on the intersection of each straight line, each straight line applied to the stationary object in FIG. 10 intersects at substantially the same position P0, while each straight line applied to the moving object in FIG. 11 changes its intersection from P1 to P2. ing. Therefore, it can be determined whether the three-dimensional object is moving or stationary depending on whether or not the intersection of the straight lines is changed.

  Finally, in step 212, the plane / solid discriminating unit 9A moves along with, for example, the latest (this time) top view image stored in the work memory unit 10 as to whether or not the object applied to the edge is a solid object, and the solid object moves. The determination result of whether it is stationary or not is output to the subsequent apparatus, and one determination process is completed.

In the present embodiment, steps 200 and 201 in the flowchart of FIG. 9 constitute the image acquisition means in the invention, and step 202 corresponds to the feature point extraction means.
Steps 203 and 204 constitute voting means, and in particular, step 203 corresponds to edge element parameter calculation means.
Steps 205 to 211 constitute discrimination means. In particular, steps 205 and 206 correspond to grouping means, and step 209 corresponds to representative value calculation means.

  The present embodiment is configured as described above. Edges are extracted from a plurality of top view images based on time-series imaging using the camera 2, and parameters of each edge element are voted on the voting spaces T1 and T2 to vote. Since it is determined whether the edge belongs to the three-dimensional object based on the state of the time-series voting result in the space, the top view image is an image converted from the side view image captured by the camera 2. However, it is possible to accurately extract a three-dimensional object without being affected by distortion caused by image conversion.

In particular, when the edge element parameters are set to the slope θ and intercept b of the straight line, one voting point is generated from one edge element, so that the voting amount can be greatly reduced as compared with Hough transform or the like. The processing speed can be increased.
Furthermore, since the determination as to whether or not the object is a three-dimensional object is made based on the state of the high frequency value region in the voting spaces T1 and T2, the processing is quick also in this respect.
In addition, whether or not the position of the region of interest changes by calculating the slope θ and the intercept b of the edge element parameters in the two time-series images on the absolute coordinates corrected for the change of the image coordinates due to the movement of the vehicle. Based on the above, it can be easily determined whether the corresponding edge belongs to a solid object or a road surface pattern.

  In voting, since the axis for taking the intercept is switched according to the inclination of the element of the edge and the voting space is also switched, the voting spaces T1 and T2 can be limited to a finite area, which is necessary for calculation. The capacity of the work memory unit 10 can also be reduced.

  Further, the grouping calculation unit 11 groups adjacent regions with high voting frequency values in the voting space, and based on the group state, that is, when the group is moving in time series, an edge corresponding to the group is displayed. When a group belongs to a solid object and the group is not moving, it is determined that the edge corresponding to the group belongs to a plane, so it is resistant to image fluctuations and noise associated with image conversion, and is a stable solid. Object discrimination is possible.

Grouping, in particular, extracts the maximum area of the voting frequency value and collects the surrounding high frequency value areas around the maximal area, so even if it appears as a small object in the image, it is not excluded as noise in the voting space. As a result, the discrimination accuracy of the three-dimensional object is improved.
In determining the three-dimensional object, each group is represented by one parameter, so that it is possible to prevent a plurality of similar high-frequency value regions from being extracted and complicating the calculation.

  Furthermore, in this embodiment, the representative value of the group related to the three-dimensional object is calculated as a straight line, and when the intersection of the straight line in time series changes, it is determined that the three-dimensional object related to the group is a moving object. In addition to whether the target object is a three-dimensional object, detailed information can be obtained up to whether it is moving or stationary.

  In the grouping operation unit 11 of the second embodiment, a high-frequency area whose voting frequency value is higher than a predetermined threshold is extracted, and adjacent high-frequency value areas within a predetermined distance range are grouped as the same area. However, as a grouping, clustering including hierarchical methods for classifying data hierarchically and division optimization methods can also be used. In this case, the voting space is divided into meshes in advance. Instead, the values of the slope θ and the intercept b can be voted as they are for grouping.

Further, the calculation of the representative value of the group in the plane / solid discrimination unit 9A is performed by straight line fitting by the least square method. However, the present invention is not limited to this, and when the group parameter is small, a simple average or Median is used. Or may be obtained by robust estimation such as RANSAC (RANdom SAmple Consensus) or LmedS (Least Median of Squares) if the data contains a lot of noise components. Good.
Furthermore, the representative value was calculated for the group related to the three-dimensional object. However, if the calculation is performed for the group related to the plane, it can be similarly distinguished whether it is moving or stationary, so the parking frame drawn on the road surface The detection accuracy is improved further.

In each of the above-described embodiments, the image data from the camera 2 is taken into the image memory 3 at a predetermined time interval, and the converted image creation unit 4 converts the image data into the top view image at the taken interval. These image data may be sequentially taken into the image memory 3, and the converted image creation unit 4 may read out the image data from the image memory 3 at the predetermined time interval and convert it into a top view image.
Further, although the Sobel filter is used for edge extraction to determine the inclination θ of the edge element from the vertical and horizontal components (dx, dy), the present invention is not limited to this. For example, higher-order local autocorrelation ( Since the spatial filter used in HLAC has the same effect as edge extraction in various directions, the necessary edge inclination can be calculated by the combination thereof.

Each embodiment has been described with respect to an example in which a side view image is converted to a top view image. However, the present invention is also applicable to a case where a solid object and a plane are discriminated on a side view image that is captured by the camera 2, for example. It can be applied as it is.
Or, conversely, an image conversion function may be built in the camera, and the converted image data may be sent directly to the work memory unit 10. In this case, the image memory 3 and the converted image creation unit 4 are unnecessary on the image processing apparatus side.

It is a block diagram which shows the structure of a 1st Example. It is explanatory drawing which shows arrangement | positioning of a camera. It is a flowchart which shows the flow of a solid-object discrimination | determination process. It is a figure which shows the relationship of the parameter vote to an edge image and voting space. It is a figure which shows the change mode of the time series of the attention area in voting space. It is a block diagram which shows the structure of a 2nd Example. It is a figure which shows the classification | category point of the inclination of the edge element for voting space switching. It is a conceptual diagram which shows the range of the parameter intercept when it switches by inclination. It is a flowchart which shows the flow of a solid-object discrimination | determination process. It is a figure which shows the time-sequential change of the straight line representing the group of the voting points concerning a stationary object. It is a figure which shows the time-sequential change of the straight line representing the group of the vote points concerning a moving object.

Explanation of symbols

DESCRIPTION OF SYMBOLS 1, 1A image processing apparatus 2 Camera 3 Image memory 4 Conversion image creation part 5 Edge extraction part 6, 6A Edge parameter calculation part 7, 7A Parameter voting part 8 Attention area extraction part 9, 9A Plane / stereoscopic discrimination part 10 Work memory part 11 Grouping calculation unit θ Inclination of edge element b intercept Gh1, Gh2, Gr1, Gr2 Region of interest L1, L2, L3, L4, L5, L6 Straight line (group representative value)
T Voting space W Parking frame

Claims (12)

Image acquisition means for acquiring images around a plurality of vehicles based on time-series imaging;
From the captured image obtained by the image acquisition means, an overhead image generation means for generating an overhead image looking down from the viewpoint position above the host vehicle;
Feature point extraction means for extracting feature points from the image obtained by the image acquisition means from the overhead view image obtained by the overhead view image generation means;
Voting means for voting feature point parameters to a voting space;
An image processing apparatus for a vehicle, comprising: a determination unit that determines whether the feature point belongs to a three-dimensional object based on a state of a voting point group in the voting space. The feature points extracted by the feature point extraction means are edges in the image,
The voting means has an edge element parameter calculation means that calculates an intercept between a slope and an axis forming a linear equation of the edge element and uses the parameter as the parameter,
The vehicular image processing apparatus according to claim 1, wherein the determination unit determines whether the feature point belongs to a three-dimensional object based on a state of a region having a high vote frequency value in a voting space. The edge element parameter calculation means defines the position and orientation of the vehicle in one image as absolute coordinates, and when the vehicle moves between the acquisition points of the plurality of images, corrects the movement to obtain absolute coordinates. The vehicular image processing apparatus according to claim 2, wherein the inclination and the intercept of the edge element are calculated. The edge element parameter calculation means switches the axis taking the intercept according to the inclination of the edge element,
4. The vehicular image processing apparatus according to claim 3, wherein the voting means switches the voting space in accordance with an inclination of an edge element. The discriminating means, when an area with a high voting frequency value is moving in time series, an edge corresponding to the area belongs to a solid object, and when the area with a high voting frequency value is not moving, the area 5. The vehicular image processing apparatus according to claim 3, wherein an edge corresponding to is belonging to a plane. The discriminating means comprises grouping means for grouping adjacent areas having high voting frequency values in a voting space, and determining whether the feature points belong to a three-dimensional object based on a group state. The vehicular image processing apparatus according to claim 3 or 4. The vehicular image processing apparatus according to claim 6, wherein the grouping unit extracts a maximum area of voting frequency values and performs grouping using the maximum area as a nucleus. When the group is moving in time series, the determination means is such that an edge corresponding to the group belongs to a solid object, and an edge corresponding to the group belongs to a plane when the group is not moving. The vehicular image processing device according to claim 6 or 7, wherein the vehicular image processing device is discriminated. The vehicular image processing apparatus according to claim 8, wherein each group is represented by one parameter in the determination. The discriminating means further includes representative value calculating means for calculating a representative value of the group, and based on the state of the representative value in time series, the three-dimensional object or plane related to the group is a moving object or a stationary object. The vehicular image processing apparatus according to claim 6, wherein the vehicular image processing apparatus determines whether it is present. The representative value calculating means calculates the representative value of the group as a straight line;
11. The vehicular image processing apparatus according to claim 10, wherein when the intersection of the straight lines in time series changes, the determination unit determines that a three-dimensional object or a plane related to the group is a moving object. In an image processing apparatus method for processing an image around a vehicle imaged by a camera,
A plurality of images around the vehicle are acquired in time series,
From the acquired image, generate a bird's eye view looking down from the viewpoint position above the host vehicle,
Extract feature points from the generated overhead image and vote the parameters to the voting space.
An image processing method comprising: determining whether or not the feature point belongs to a three-dimensional object based on a time-series state of voting point groups in the voting space.

Images