JP6200780B2 - Lane recognition determination device - Google Patents

Description

  The present invention relates to a lane recognition determination apparatus that determines the reliability of whether or not an inner guide line is recognized based on image data obtained by photographing the front of a host vehicle traveling road.

  Conventionally, in traveling on an expressway or the like, a traveling environment in front of the host vehicle is detected, and a traveling lane that divides the left and right of the traveling lane is detected based on the detected traveling environment data. There are known driving support devices such as a lane departure control for automatically driving a vehicle, and a lane departure warning that warns the driver when the vehicle is predicted to depart from the recognized lane position. .

  In order for the host vehicle to travel accurately along the travel lane, it is necessary to accurately estimate the position of the travel lane. As a means for estimating the traveling lane position, the brightness of the road surface and the white line is determined based on the traveling environment in front of the host vehicle detected by an image sensor such as a camera mounted on the vehicle (hereinafter referred to as “vehicle camera”). A technique is known in which the edge of a white line is detected from the difference and the traveling lane position is estimated from the edge of the white line.

  By the way, the driving lane on the expressway is not only the left and right driving lanes that divide the driving lane, but also on curved roads, a double white line consisting of a driving lane and an inner guide line (so-called dot line) is drawn. May be. In the technology that detects the edge of the white line based on the difference in brightness from the road surface, the white line on the road has a clearer brightness difference than the white line on the outside, so for the double white line, the lane is estimated based on the inner edge of the inner guide line. There are many cases to do.

  For example, in Patent Document 1 (Japanese Patent Laid-Open No. 2012-58984), the plot position of the lane candidate point is switched from the inner edge of the traveling lane to the inner edge of the guide line at the double white line entrance (start end). A technique for estimating a lane based on a subsequent point sequence is disclosed.

JP2012-58984A

  As described above, in the technique disclosed in Patent Document 1, at the double white line entrance (start end), the detection position of the lane candidate point is switched to the inner edge of the inner guide line. The timing is not clear. Also, in the same document, when the entrance of the double white line is detected, the set time is advanced from the parameter coefficient point sequence before the set time width with reference to the time of the current calculation, and this set time is set when the prediction parameter is set later. By increasing the width and increasing the number of points in the past parameter coefficients, the lane can be estimated stably.

  However, in the technique disclosed in this document, when the entrance (starting end) of the double white line is detected, it is necessary to capture a lot of past data, so that a response delay is likely to occur. In addition, when the entrance of the double white line is detected, the lane candidate point is automatically switched to the inner lane, and this is, for example, a pool of water that exists in the eaves formed along the traveling lane and the traveling lane. If the remaining snow along the edge or the edge of the repair mark is misrecognized as the inner edge of the guide line, the subsequent lane estimation becomes unstable.

  In view of the above circumstances, the present invention clarifies the timing when the lane candidate point is switched to the inner edge of the inner guide line by detecting the entrance (starting end) of the double white line, and further, the inner guide line, the noise component, and the like. It is an object of the present invention to provide a lane recognition determination apparatus that can clearly classify a vehicle.

  The lane recognition determination apparatus according to the present invention extracts an edge point based on a luminance difference for each of a plurality of search lines extending in the horizontal direction within each of the left and right white line detection areas on an image taken in front of the host vehicle travel path. Based on the point extraction means and the point sequence of the edge points extracted by the edge point extraction means, a Hough straight line is obtained by Hough transformation, and based on the Hough straight line, a demarcation line that divides the own vehicle traveling path is detected. A white line candidate point extraction region setting unit for setting a white line candidate point extraction region, and an edge point of a region that is located on the image center side of the white line candidate point extraction region is extracted as the 0th white line candidate point. 0th white line candidate point number calculating means for calculating the number of white line candidate points, the number of the 0th white line candidate point values calculated by the 0th white line candidate point number calculating means, and the center side on the image side of the lane marking based on the distribution Inside depicted in And a guide line recognition reliability determining means for determining the recognition reliability of the conductor.

  According to the present invention, a Hough straight line is obtained by Hough transformation based on a sequence of edge points extracted by a luminance difference from an image taken in front of the own vehicle traveling road, and the own vehicle traveling road is partitioned based on the Hough straight line. A white line candidate point extraction region having a predetermined width for detecting a dividing line to be detected is set, and an edge point of an area that is out of the white line candidate point extraction region on the center side of the image is extracted as a zeroth white line candidate point. Since the number of 0 white line candidate points is calculated and the number of 0th white line candidate points and the distribution are determined, the recognition reliability of the inner guide line drawn on the center side of the image from the partition line is determined. When the recognition reliability is low, the white line shape is not estimated by the inner guide line, so the inner guide line and the noise component can be clearly distinguished, and the lane candidate point is the inner edge of the inner guide line. The timing to switch to Since it does not incorporate many points in the past parameter coefficients, it is possible to detect the entrance (starting end) of the double white line and switch the lane candidate point to the inner edge of the inner guide line. There is no response delay.

Schematic configuration diagram of a vehicle driving support device including a lane recognition determination device Flow chart showing a double white line shift start determination processing routine (part 1) Flowchart showing a double white line shift start determination processing routine (part 2) Flowchart showing white line shape estimation processing routine Explanatory drawing of the frame image that captured the travel lane Explanatory drawing of the lane candidate point sequence which detected the inner edge of the traveling lane coordinate-transformed in real space Explanatory drawing of a frame image in which a band-like water pool exists inside the traveling lane Explanatory drawing of the lane candidate point sequence which detected the inner edge of the driving | running | working lane coordinate-transformed on real space, and the edge of a puddle Illustration of a frame image of a double white line Explanatory drawing of the lane candidate point sequence which detected the inner edge of the driving | running | working lane coordinate-converted on real space, and the inner edge of an inner guide line Explanatory drawing which shows transition of the brightness | luminance and the differential value of a brightness | luminance in the white line detection area in a horizontal search line (A) is an explanatory diagram of the candidate point sequence for the double white line, (b) is an explanatory diagram of the candidate point sequence when there is a puddle inside the traveling lane, and (c) is a belt-like residual snow inside the traveling lane. Explanatory drawing of candidate point sequence when it exists Explanatory drawing of coordinate conversion when moving the previous representative position to the current representative position Explanatory drawing which shows the estimated position of the edge point accompanying the movement of the own vehicle Explanatory drawing which shows the candidate point sequence and rear white line candidate point which were coordinate-transformed in real space Explanatory drawing showing the calculation method of Hough transform Explanatory drawing showing Hough space (A) is explanatory drawing which shows a Hough straight line and white line candidate point extraction area | region, (b) is explanatory drawing which shows the extracted white line candidate point (A) is explanatory drawing which shows the edge point inside a white line candidate point extraction area | region immediately before determined as a double white line entrance, (b) is a description which shows a white line candidate point when it determines with a double white line entrance. Figure Explanatory drawing which shows the white line approximate curve and white line search area | region recognized from the white line candidate point and the back white line candidate point.

  Hereinafter, an embodiment of the present invention will be described with reference to the drawings. In FIG. 1, reference numeral 1 denotes a vehicle such as an automobile (own vehicle), and a driving support device 2 that steers the vehicle (steering control, brake control, etc.) based on environmental information ahead of the own vehicle. It is installed. The driving support device 2 includes, for example, an in-vehicle camera 3 as an in-vehicle image sensor, a stereo image recognition device 4, a control unit 5 having a function as driving support means, and the like.

  The host vehicle 1 is connected to a vehicle speed sensor 11 that detects the host vehicle speed, a yaw rate sensor 12 that detects the yaw rate, a main switch 13 that performs ON / OFF switching of each function of the driving support control, and the steering wheel 14. A steering angle sensor 15 that detects the rudder angle provided on the steering shaft 14a, an accelerator opening sensor 16 that detects an accelerator pedal depression amount (accelerator opening) by the driver, and the like are provided. Further, a power steering motor 17 that assists the driver's steering operation is connected to the steering shaft 14a.

  The in-vehicle camera 3 is composed of a pair of left and right color cameras including a main camera having a solid-state imaging device such as a CCD or CMOS and a sub camera as a stereo optical system. Each of these cameras is attached to the ceiling of the passenger compartment behind the windshield with a certain distance, takes a stereo image of an object outside the vehicle from different viewpoints, and transmits the image data to the stereo image recognition device 4. In the following description, an image captured by the main camera among images captured in stereo is referred to as a reference image, and an image captured by the sub camera is referred to as a comparative image.

  First, the stereo image recognition device 4 divides the reference image into small regions of 4 × 4 pixels, compares the luminance or color pattern of each small region with the comparison image, finds the corresponding region, and creates the entire reference image. Find the distance distribution across. Further, the stereo image recognition device 4 examines the luminance difference between each pixel on the reference image and an adjacent pixel, extracts those whose luminance difference exceeds a threshold value as an edge point, and extracts the extracted pixels. By giving distance information to (edge point), a distance image (edge point distribution image having distance information) is generated. Then, the stereo image recognition device 4 performs a well-known grouping process on the generated distance image, and compares it with a pre-stored three-dimensional frame (window) so that a white line, a side wall, Recognizes three-dimensional objects.

  Here, the white lines to be recognized in the present embodiment are, for example, drawn on the inner sides of the left and right white lines (partition lines) 21L and 21R that divide the own vehicle traveling lane and the lane lines 21L and 21R. This is a general term for lines drawn on the road to indicate the vehicle lane, such as a so-called double white line, composed of the inner guide line 22 (see FIGS. 5 and 9). The form of each white line is not limited to a solid line or a broken line, and further includes a yellow line or the like. In the white line shape recognition of the present embodiment, even if the white line existing on the road is a double white line, the shape along the inner edge of each white line is estimated by approximating each of the left and right single approximate lines. It shall be.

  In the estimation of the white line shape, the stereo image recognition device 4, for example, as shown in FIGS. 5, 7, and 9, white line detection areas Al, Ar (hereinafter, Al and Ar are collectively referred to as A), and within each white line detection area A, the horizontal search line i with a one-pixel pitch is distant from the lower side (the front side of the host vehicle). The pixels are sequentially searched from the inner side in the vehicle width direction (center side on the image) to the outer side while offsetting by one pixel pitch in the (j direction) to examine the luminance change. Then, when there is an edge point on the horizontal search line in each white line detection area A where the luminance changes more than a predetermined value from dark to bright, the stereo image recognition device 4 determines the first edge point as the first edge point. The edge point P1 is detected, and the next edge point is detected as the second edge point P2.

  In addition, the stereo image recognition device 4 determines whether the white line that divides the vehicle traveling path is only a traveling lane (hereinafter referred to as a single white line) based on the detected point sequence of the first and second edge points P1 and P2. To determine whether it is a double white line.

  If it is determined that the white line is a single white line, the stereo image recognition device 4 will be described based on the Hough straight lines H1, Hr (hereinafter, H1, Hr are collectively referred to as H) based on the sequence of the first edge points P1. The first edge point P1 existing in the white line candidate point extraction areas Adl, Adr (hereinafter, Adl, Adr will be collectively referred to as Ad) determined based on the Hough straight line H is calculated as a white line. Extracted as candidate points Ps. Although details will be described later, as shown in FIG. 19A, the white line candidate point extraction region Ad has a predetermined width α1 on the inner side and a predetermined width α2 on the outer side across the Hough straight line H. The predetermined width α1 is set to a value shorter than the distance from the inner edge of the partition lines 21L and 21R to the inner edge of the inner guide line 22 and longer than the line width of the partition lines 21L and 21R shown in FIG. Yes.

  On the other hand, when it is determined that the white line is a double white line, the stereo image recognition device 4 uses the white line candidate point extraction region Ad defined based on the Hough straight line H calculated based on the detected second edge point P2. Also, the first edge point P1 existing inside is extracted as the white line candidate point Ps. As described above, the predetermined width α1 inside the white line candidate point extraction region Ad is set to be shorter than the distance from the inner edge of the partition lines 21L and 21R to the inner edge of the inner guide line 22, so that the double white line In this case, the first edge point P1 exists inside the white line candidate point extraction area Ad.

  Further, the stereo image recognition device 4 is located behind the vehicle within a set distance (for example, 20 to 30 [m]) based on the coordinates of the white line candidate point Ps extracted before the previous frame and the amount of movement of the vehicle. The white line candidate point Ps_pre is estimated, and the shape of the white line is approximated based on the extracted white line candidate point Ps and the point sequence of the estimated rear white line candidate point Ps_pre.

  The control unit 5 receives driving environment information in front of the host vehicle 1 recognized by the stereo image recognition device 4. Further, the vehicle speed V detected by the vehicle speed sensor 11 and the yaw rate γ detected by the yaw rate sensor 12 are input to the control unit 5 as travel information of the host vehicle 1, and the main switch is input as operation input information by the driver. The ON / OFF signal from 13, the steering angle θst detected by the steering angle sensor 15, the accelerator opening θacc detected by the accelerator opening sensor 15, etc. are input.

  For example, when an instruction to execute lane departure prevention control or lane keeping support control, which is one of the functions of the driving support control, is given through the operation of the main switch 13 by the driver, the control unit 5 displays the stereo image. A departure determination line is set based on the travel lane recognized by the recognition device 4 (approximate line of the white line shape), and the own vehicle travel path is estimated. In the lane departure prevention control, when it is determined that the host vehicle tends to depart from the departure determination line, steering (steering control or brake control) is performed to return the host vehicle to the lane center side, and lane maintenance support is performed. In the control, a target travel position is set in the travel lane, and steering (steering control or brake control) is performed so that the host vehicle travels along the target travel position.

  Specifically, the white line shape recognition processing executed by the stereo image recognition device 4 described above is performed according to the double white line transition start determination processing routine shown in FIGS. 2 and 3 and the white line shape estimation processing routine shown in FIG. Done.

  The flowcharts shown in FIGS. 2 and 3 are executed for each frame. First, in step S1, an edge point P is detected from an image (reference image) taken by the in-vehicle camera 3. That is, as shown in FIGS. 5, 7, and 9, the stereo image recognition device 4 is located on the inner side in the vehicle width direction (center in the width direction of the image) for each horizontal search line in the white line detection area A set on the image. Investigate the change in luminance from the outside to the outside (i direction in the figure), the luminance of the outer pixel is relatively higher than the luminance of the inner pixel, and the luminance differential value indicating the amount of change is positive An edge point PH (white line start point) that is equal to or greater than the set threshold value is searched. Subsequently, an edge point PL (white line end point) that is relatively low with respect to the luminance of the inner pixel and that has a luminance differential value that indicates the amount of change is equal to or less than a set threshold value on the negative side is searched for ( FIG. 11).

  When one edge point PH (white line start point) is detected in the white line detection area A, the edge point PH is set as the first edge point P1. Alternatively, when two or more edge points PH are detected in the white line detection area A, for example, the first edge point PH is set as the first edge point P1, and the next edge point PH is set as the second edge point P2. To do. Furthermore, the stereo image recognition device 4 converts the coordinates of each edge point PH into coordinates in the real space by giving distance information on the distance image to each detected edge point PH (FIGS. 6 and 8). (See FIG. 10). Note that the processing in step S1 corresponds to the edge point extracting means of the present invention.

  Thereafter, when proceeding to step S2, the rear white line candidate point Ps_pre is estimated based on the coordinates of the white line candidate point Ps extracted before the previous frame and the movement amount of the host vehicle 1. Here, the movement amount of the host vehicle 1 can be obtained by using, for example, the host vehicle speed V and the yaw angle θ (an angle calculated from the yaw rate (dθ / dt)). For example, as shown in FIG. 13, assuming that the frame rate is Δt, the movement amounts Δx and Δz that the host vehicle 1 moves in the host vehicle 1 ′ direction, that is, the X axis direction and the Z axis direction in one frame are as follows: It can be obtained from the equations (1) and (2).

Δx = V · Δt · sin θ (1)
Δz = V · Δt · cos θ (2)
Therefore, for example, as shown in FIG. 14, the coordinates of the white line candidate point Ps_old detected in the previous frame are (xold, zold), and the white line candidate point Ps_old is the white line candidate point (back white line candidate point) Ps_pre of the current frame. When it is estimated that it has moved, the coordinates (xpre, zpre) can be obtained from the following equations (3) and (4). That is, the coordinates (xpre, zpre) are obtained by subtracting the vehicle movement amounts Δx, Δz from the coordinates (xold, zold) and then converting the coordinates to the vehicle fixed coordinate system (X ′, Z ′) in the current frame. It can be obtained by doing.

xpre = (xold−Δx) · cos θ− (zold−Δz) · sin θ (3)
zpre = (xold−Δx) · sin θ + (zold−Δz) · cos θ (4)
Thereby, for example, as shown in FIG. 15, the rear white line candidate point Ps_pre within the set distance behind the host vehicle 1 can be estimated.

  Thereafter, when the process proceeds to step S3, a Hough straight line H is obtained by Hough transformation using the point sequence of the first edge points P1. Here, an example of how to obtain the Hough straight line H will be described. First, as shown in FIG. 16, with respect to each of the first edge points P1 constituting the point sequence, the inclination of the straight line h passing through the first edge point P1 is set to a predetermined angle Δθ from 0 ° to 180 °. Based on the following equation (5), an offset amount ρ, which is the distance (vertical length) from the origin O to the straight line h at each angle θ, is obtained.

ρ = x · cos θ + z · sin θ (5)
Then, the relationship between each angle θ obtained for each edge point P1 and the offset amount ρ is voted (projected) as a frequency on a corresponding portion on the Hough plane (θ, ρ) shown in FIG. Thereafter, a combination of the angle θ and the offset amount ρ with the largest frequency on the Hough plane (θ, ρ) is extracted, and the straight line (5) is defined using the angle θ and the offset amount ρ ( Find the Hough straight line H). The Hough straight line H obtained this time is read as a Hough straight line Hold in the subsequent step S4 when the next routine is executed.

  Next, when proceeding to Step S4, the Hough straight line Hold obtained based on the previous frame image is read, and at Step S5, a white line candidate point extraction region Ad having a predetermined width is set with reference to the Hough straight line Hold. As shown in FIGS. 18A and 19A, in the white line candidate point extraction region Ad, predetermined widths α1 and α2 are set inside and outside the Hough straight line H. As described above, the inner predetermined width α1 is shorter than the distance between the inner edge of the double white line partition lines 21L and 21R shown in FIG. 9 and the inner edge of the inner guide line 22, and the partition line 21L. It is set to a value longer than the line width (or 21R). It should be noted that the double white line dividing lines 21L and 21R, the width of the inner guide line 22, and the interval between the inner edge of the partition lines 21L and 21R and the outer edge of the inner guide line 22 are defined in advance.

  On the other hand, the outer predetermined width α2 is set to a value longer than the line width of the dividing line 21L (or 21R), and is set to the same width as the predetermined width α1 in the present embodiment. Therefore, in the case of the double white line, as shown in FIG. 19A, the first edge point P1 for detecting the inner edge of the inner guide line 22 is. It is scattered inside the white line candidate point extraction area Ad. 18 and 19I show the left lane marking 21L, the white line candidate point extraction region Ad is similarly set for the right lane marking 21R. The processing in steps S3 to S5 corresponds to the white line candidate point extraction region setting means of the present invention.

  Thereafter, when proceeding to step S6, the first and second edge points P1 and P2 existing in the white line candidate point extraction area Ad are extracted as the first white line candidate point Ps1 (see FIG. 18B), and then, In step S7, the number n1 of first white line candidate points Ps1 (hereinafter referred to as first white line candidate point number n1) is calculated.

  Then, the process proceeds to step S8, where the edge point P1 existing on the inner side in the vehicle width direction from the white line candidate point extraction area Ad is extracted as the 0th white line candidate point Ps0 (see FIG. 18B), and then in step S9. Then, the number n0 of the 0th white line candidate points Ps0 (hereinafter referred to as the 0th white line candidate point number n0) is calculated. The processes in steps S8 and S9 correspond to the 0th white line candidate point number calculating means of the present invention.

  Thereafter, the process proceeds to Step S10, and in Steps S10 to S33, the double white line transition start timing, that is, the timing to start construction of a lane model formula described later is determined based on the 0th white line candidate point Ps0. The processing in steps S10 to S33 corresponds to the guide wire recognition reliability determination means of the present invention.

  In step S5 described above, the white line candidate point extraction region Ad is set based on the Hough straight line Hold set based on the previous frame. In this way, for example, as shown in FIG. 7, even if a noise component 25 such as a puddle, remaining snow, or a repair trace exists inside the lane line 21L along the lane line 21L, the white line candidate point Since the lane model formula is obtained based on Ps1, it is difficult to be influenced by the noise component 25. On the other hand, even if a double white line inner guide line 22 that is not a noise component 25 is detected, the white line candidate point that constructs the lane model formula detects the inner edge of the inner guide line 22 as the 0th white line candidate. There is an inconvenience of not shifting to the point Ps0.

  In addition, even if a curve entrance is detected far away (upper part of the screen on the frame) and a double white line is drawn at the entrance of the curve, the division lines 21L and 21R and the inner guide line 22 appear at the far curve entrance. Because it looks like a single white line, the inner guide line 22 is likely to be misrecognized as the lane markings 21L and 21R. In that case, when the white line candidate point that constructs the lane model formula detects the inner edge of the inner guide line 22 Inconvenience that it is not possible to determine the timing of the transition to the 0th white line candidate point Ps0.

  Therefore, when the 0th white line candidate point Ps0 is detected in step S10 to step S31, whether or not it is the detection of the inner edge of the inner guide line 22, in other words, the inner guide line 22 is recognized. The reliability of whether or not it is done is determined. If the reliability that the inner edge of the inner guide line 22 has been detected is high, the white line candidate point for constructing the lane model formula is switched to the 0th white line candidate point Ps0.

  That is, first, in step S10, it is checked whether or not a component that seems to be the entrance (start end) of the double white line (what seems to be the inner guide line) has been detected in the calculation based on the previous frame image. Whether or not a component that seems to be the entrance (starting end) of a double white line is detected is determined, for example, by whether or not the 0th white line candidate point Ps0 has been detected during the previous calculation.

  If a component that seems to be an entrance of a double white line is detected, the process proceeds to step S11, and the count value (hereinafter simply referred to as “count value”) Cn of the double white line reliability determination counter is incremented (Cn ← Cn + 1), the process proceeds to step S13. If no double white line-like component is detected, the process branches to step S12, decrements the count value Cn (Cn ← Cn-1), and proceeds to step S13. The count value Cn of the double white line reliability determination counter is a value representing the recognition reliability for determining whether or not the component that seems to be the inner guide line is the inner guide line 22.

  In step S13, the currently calculated 0th white line candidate point number n0 is compared with a predetermined threshold value SLn0. If n0> SLn0, the process proceeds to step S14, the count value Cn is incremented (Cn ← Cn + 1), and step S16. Proceed to If n0 ≦ SLn0, the process branches to step S15, decrements the count value Cn (Cn ← Cn−1), and proceeds to step S16. This threshold value SLn0 is a value for determining whether or not the inner edge of the inner guide line 22 has been detected in the current calculation.

  Proceeding to step S16, the distance ΔPs0 (see FIG. 12) from the first 0th white line candidate point Ps0 detected in the image of one frame to the last 0th white line candidate point Ps0 is compared with the set threshold value SLΔPs0. It is checked whether or not the distribution of the 0th white line candidate point Ps0 is sufficient to determine that the inner edge of the inner guide line 22 is detected. If ΔPs0> SLΔPs0, it is determined that the distribution of the 0th white line candidate point Ps0 is sufficient, the process proceeds to step S17, the count value Cn is incremented (Cn ← Cn + 1), and the process proceeds to step S19. If ΔPs0 ≦ SLΔPs0, the process branches to step S18, the count value Cn is decremented (Cn ← Cn−1), and the process proceeds to step S19.

  In step S19, the width f (average value) of the component regarded as the inner guide line from which the 0th white line candidate point Ps0 is detected is detected, and the width f is between the predetermined lower limit width fS and the predetermined upper limit width fL. Find out if it fits. As shown to Fig.12 (a), the width f of the white line guide wire 22 is prescribed | regulated previously at about 10-15 [cm]. Therefore, by setting the predetermined lower limit width fS and the predetermined upper limit width fL to values including this specified value, is the 0th white line candidate point Ps0 detected from the inner edge of the inner guide line 22 from the width f? It can be determined whether or not. As shown in FIG. 11, the width f is the average value in the horizontal direction of the edge point PH (white line start point) detected by the change in luminance and the average value in the horizontal direction of the edge point PL (white line end point). It is a difference.

  If fS <f <fL, the process proceeds to step S20, the count value Cn is incremented (Cn ← Cn + 1), and the process proceeds to step S22. If f≤fS or f≥fL, the process branches to step S21, the count value Cn is decremented (Cn ← Cn-1), and the process proceeds to step S22.

  As a result, the width f as shown in FIG. 12 (b) is larger than the predetermined upper limit width fL (f ≧ fL), the water pool and the repaired trace, and the partition line 21L as shown in FIG. If the extending width f is smaller than the predetermined upper limit width fL (f ≦ fS), it is determined that the strip-like remaining snow or the like is the noise component 25, and the count value Cn is decremented. In FIG. 12, only the left lane marking 21L is shown for ease of explanation, but the left lane marking 21L is replaced with the right lane marking lane 21R, and applied to the right lane marking lane 21R. be able to.

  Then, when proceeding to step S22, the horizontal average value of the first white line candidate point Ps1 that has detected the inner edge of the division line 21L and the horizontal average value of the edge point PL that has detected the outer edge of the component that seems to be the inner guide line. The gap m (average value) is detected from the difference between the two, and it is checked whether or not the gap m is within the predetermined lower limit gap mS and the predetermined upper limit gap ml. As shown in FIG. 12A, the gap m between the inner edge of the partition line 21L and the outer edge of the inner guide line 22 is defined in advance to about 10 to 15 [cm]. Therefore, by setting the predetermined lower limit gap mS and the predetermined upper limit gap ml to values including this specified value, the current 0th white line candidate point Ps0 is obtained by detecting the inner edge of the inner guide line 22 from this gap m. It can be determined whether or not. As shown in FIG. 12A, the gap m is m when the horizontal position coordinate of the 0th white line candidate point Ps0 is x0 and the horizontal position coordinate of the first white line candidate point Ps1 is x1. = X0- (f + x1)

  If mS <m <mL, the process proceeds to step S23, the count value Cn is incremented (Cn ← Cn + 1), and the process proceeds to step S25. If m ≦ m S or m L, the process branches to step S24, the count value Cn is decremented (Cn ← Cn−1), and the process proceeds to step S25. Accordingly, tentatively, the width f of the puddle or repair trace shown in FIG. 12 (b) and the width f of the remaining snow shown in FIG. 12 (c) fall within the predetermined lower limit value fS and the predetermined upper limit value fL. (FS <f <fL), if the gap m does not fall within the predetermined lower limit gap mS and the predetermined upper limit gap ml, it is determined that there is a possibility of the noise component 25, and the count value Cn is decremented.

  After that, when proceeding to step S25, first, a histogram of the width f at each 0th white line candidate point Ps0 of the component that seems to be the inner guide line is generated, and the number (peak number) fpek of the width f indicating a predetermined peak width or more is obtained. A ratio (peak ratio) of the peak number fpek to the zeroth white line candidate point number n0 is obtained (fpek / n0), and the peak ratio fpek / n0 is compared with a predetermined peak determination threshold SLfp.

  If fpek / n0> SLfp, the process proceeds to step S26, the count value Cn is incremented (Cn ← Cn + 1), and the process proceeds to step S28. If fpek / n0 ≦ SLfp, the process branches to step S27, the count value Cn is decremented (Cn ← Cn−1), and the process proceeds to step S28. Since the inner guide line 22 has a substantially constant width f, the peak ratio fpek / n0 is relatively high. On the other hand, for example, the band-shaped residual snow shown in FIG. 12C has a non-constant width f. Therefore, even if the width f (average value) is the predetermined lower limit width fS and the predetermined upper limit width in step S19 described above. Even if it falls within fL, if the peak ratio fpek / n0 does not exceed the peak determination threshold SLfp, it is determined that there is a possibility of the noise component 25, and the count value Cn is decremented.

  Next, in step S28, a difference (luminance difference) AVps between the average luminance AVps1 of each first white line candidate point Ps1 and the average luminance AVps0 of each zeroth white line candidate point Ps0 is obtained (AVps ← AVps1-AVps0). The brightness difference AVps is compared with a predetermined brightness difference determination threshold SLps. The brightness of the 0th white line candidate point Ps0 that detected the inner edge of the inner guide line 22 is brighter than the brightness of the first white line candidate point Ps1 that detected the inner edge of the partition line 21L (21R). When detected, the luminance difference AVps is a small value. Further, when the brightness of the 0th white line candidate point Ps0 is high, it indicates that the inner guide line 22 appears clearly. Therefore, when the brightness difference AVps shows a small value, the lane model formula detects a single white line. It is preferable that the first white line candidate point Ps1 is switched to the 0th white line candidate point Ps0 for detecting the inner guide line 22 of the double white line.

  If AVps <SLps, the process proceeds to step S29, the count value Cn is incremented (Cn ← Cn + 1), and the process proceeds to step S31. If AVps ≧ SLps, the process branches to step S30, the count value Cn is decremented (Cn ← Cn−1), and the process proceeds to step S31.

  In step S31, it is determined whether or not the count value Cn exceeds the reliability determination threshold Cno. If Cn> Cno, it is determined that the detected component is highly reliable that it is the inner edge of the inner guide line 22, In step S32, the double white line start flag Fps0 is set (Fps0 ← 1), and the routine is exited. On the other hand, when Cn ≦ Cno, it is determined that the reliability of the detected component as the inner edge of the inner guide line 22 is low, the process branches to step S33, and the double white line start flag Fps0 is cleared (Fps0 ← 0) Exit the routine. Therefore, by setting the double white line reliability determination threshold Cno to a predetermined value, desired determination accuracy can be obtained, and the double white line start timing can be appropriately set.

  Thus, according to the present embodiment, when detecting the entrance (starting end) of the double white line and switching the lane candidate point to the inner edge of the inner guide line, the reliability is determined as the double white line reliability determination. Since the determination is made based on the count value Cn of the counter, the inner guide line and the noise component can be clearly distinguished, and the inner guide line of the double white line is recognized with high accuracy without causing erroneous recognition. be able to. Further, since the double white line start flag Fps0 is not set unless the count value Cn exceeds the threshold value Cno for determining the reliability, the double white line start timing can be clarified.

  This double white line start flag Fps0 is read by the white line shape estimation processing routine shown in FIG. In this routine, first, the value of the double white line start flag Fps0 is read in step S41, and it is checked in step S42 whether the double white line start flag Fps0 is 0. If Fps0 = 0, the process proceeds to step S43. A lane model formula is set based on the first white line candidate point Ps1 and the rear white line candidate point Ps_pre at which the inner edges of 21L and 21R are detected, and the routine is exited. If Fps0 = 1, the process proceeds to step S44, where the lane model formula is set based on the 0th white line candidate point Ps0 and the rear white line candidate point Ps_pre that detected the inner edge of the inner guide line 22, and the routine is exited.

This lane model formula uses the least square method based on the point sequence of the first white line candidate point Ps1 and the rear white line candidate point Ps_pre when Fps0 = 0, that is, when the double white line is not detected. A curve approximation formula (X = aZ ² + bZ + c) for the lane markings 21L and 21R is calculated, and the shape of the white line is estimated using this curve approximation formula. Further, when Fps0 = 1, that is, when a double white line is detected, as shown in FIG. 20, the least square method is used based on the point sequence of the 0th white line candidate point Ps0 and the back white line candidate point Ps_pre. Is used to calculate a curve approximation formula (X = aZ ² + bZ + c) of the inner guide line 22, and the shape of the white line is estimated by this curve approximation formula. Here, the coefficient a is a parameter related to the curvature of the lane, and the coefficient b is an angle related to the inclination (yaw angle) of the host vehicle 1 with respect to a virtual line passing through the center of the host vehicle 1 (more precisely, between the main camera and the sub camera). The parameter, coefficient c, is a parameter indicating the horizontal position, and is an offset amount (horizontal position from the origin) in the X-axis direction with the center of the own vehicle amount 1 (intermediate between the main camera and the sub camera) as the origin. Is a negative value (−) and the right side is a positive value (+), the coefficient c of the left lane model formula is a negative value (−), and the coefficient c of the right lane model formula is a positive value (+). .

  The approximate curve obtained based on the lane model formula is read when the driving support is executed, and the left and right departure judgment lines are set based on the approximate curve to estimate the own vehicle travel path. In the lane departure prevention control, when it is determined that the host vehicle 1 tends to depart from the departure determination line, steering (steering control or brake control) is performed to return the host vehicle 1 to the lane center side. In the lane keeping support control, a target travel position is set in the travel lane based on the approximate curve, and steering (steering control and brake control) is performed so that the host vehicle travels along the target travel position.

  The present invention is not limited to the above-described embodiment. For example, the in-vehicle camera 3 does not have to be a stereo camera, and may be a monocular camera. When a monocular camera is used as the in-vehicle camera 3, three-dimensional information is obtained by a known motion stereo method, for example. Furthermore, the lane model formula is not limited to a quadratic formula, but may be a cubic formula.

1 ... own vehicle,
2 ... Driving assistance device,
3… In-vehicle camera,
4 ... Stereo image recognition device,
5 ... Control unit,
21L ... Left line,
21R ... Right dividing line,
22 ... Inner guide wire,
Adl, Adr ... White line candidate point extraction region,
A, Al, Ar ... Lane detection area,
AVps ... brightness difference,
AVps0, AVps1 ... average brightness,
Cn: count value,
Cno: double white line reliability determination threshold,
f ... width,
fL ... upper limit width,
fpek ... number of peaks,
fpek / n0: peak ratio,
Fps0 ... Double white line start flag,
fS ... lower limit width,
H, H1, Hr, Hold ... Hough straight line,
I ... Horizontal search line,
m ... Gap,
mL: upper limit gap,
mS ... lower limit gap,
n0: 0th white line candidate score,
n1 ... number of first white line candidate points,
P1 ... first edge point,
P2 ... second edge point,
PH: Edge point (white line start point),
PL ... Edge point (white line end point),
Ps ... White line candidate point,
Ps0 ... 0th white line candidate point,
Ps1 ... 1st white line candidate point,
SLfp: Peak judgment threshold,
SLn0, SLΔPs0 ... threshold value,
SLps ... Brightness difference judgment threshold

Claims (8)

Edge point extraction means for extracting an edge point by a luminance difference for each of a plurality of search lines extending horizontally in each of the left and right white line detection regions on the image taken in front of the host vehicle traveling path;
A white line candidate point extraction area having a predetermined width for obtaining a Hough straight line by Hough transformation based on the point sequence of edge points extracted by the edge point extracting means, and detecting a dividing line that divides the vehicle traveling path based on the Hough straight line White line candidate point extraction area setting means for setting
A white line candidate point number calculating means for extracting an edge point of an area outside the white line candidate point extraction area on the image center side as a zero white line candidate point, and calculating the number of the zero white line candidate points;
Guide line for determining the recognition reliability of the inner guide line drawn on the center side of the image with respect to the division line based on the number and distribution of the 0th white line candidate points calculated by the 0th white line candidate point number calculation means A lane recognition determination apparatus comprising: a recognition reliability determination unit. The guide line recognition reliability determination means determines the recognition reliability based on a count number, and when the count number is a set reliability determination threshold value or less, determines that the recognition reliability of the inner guide line is low, The lane recognition determination apparatus according to claim 1, wherein the white line shape is not estimated based on the 0th white line candidate point. The lane recognition determination apparatus according to claim 2, wherein the guide line recognition reliability determination means decrements the count number when the number of the zeroth white line candidate points is equal to or less than a set threshold value. The guide line recognition reliability determination means is obtained from a difference between an average value of horizontal positions of the zeroth white line candidate points and an average value of horizontal positions of edge points outside the zeroth white line candidate points. 4. The lane recognition determination apparatus according to claim 2, wherein the count number is decremented when the width deviates from between the predetermined lower limit width and the predetermined upper limit width. First white line candidate point number calculating means for extracting edge points in the white line candidate point extracting area set by the white line candidate point extracting area setting means as first white line candidate points and calculating the number of the first white line candidate points. With
The guide line recognition reliability determination unit obtains a value obtained by adding the preset inner guide line width to the average value of the horizontal position of the 0th white line candidate point and the first white line candidate point number calculation unit. The count number is decremented when a gap obtained from a difference from an average value of horizontal positions of the first white line candidate points is out of a predetermined lower limit gap and a predetermined upper limit gap. The lane recognition determination apparatus according to any one of claims 2 to 4. The guide line recognition reliability determination means decrements the count when the distance from the first 0th white line candidate point detected in the image of one frame to the last 0th white line candidate point is equal to or less than a set threshold value. The lane recognition determination apparatus according to any one of claims 2 to 5, wherein: The guide line recognition reliability determination means is obtained from a difference between an average value of horizontal positions of the zeroth white line candidate points and an average value of horizontal positions of edge points outside the zeroth white line candidate points. Generating a histogram having a width that is greater than or equal to a predetermined peak width, and decrementing the count when the ratio of the number to the number of 0th white line candidate points is equal to or less than a predetermined peak determination threshold value. The lane recognition determination apparatus according to any one of claims 2 to 6. The guide line recognition reliability determination means obtains a difference between an average luminance of each of the first white line candidate points and an average luminance of each of the zeroth white line candidate points, and when the difference is equal to or less than a predetermined luminance difference determination threshold, The lane recognition determination apparatus according to any one of claims 2 to 7, wherein the count number is decremented.

Images