JP6398294B2 - Traveling lane recognition device and traveling lane recognition method - Google Patents

Description

  The present invention relates to a travel lane recognition device and a travel lane recognition method.

  In the prior art described in Patent Document 1, when the road surface in the traveling direction is imaged with a single camera and the passage dividing line is a broken line, the length of the broken line is extended based on an image captured before a predetermined time. The generated composite image is generated and the travel lane is recognized. Here, an image captured before a predetermined time is corrected by an amount corresponding to the change in the position and posture of the host vehicle from that time, and the corrected past image and the current image are combined.

JP-A-2005-332105

However, only one camera may be used, and when it becomes difficult to observe the broken line, it becomes more difficult to extend the length of the broken line, especially the farther away. In recent years, a fish-eye lens that can image a wide range is often used, but there is a situation where it is difficult to observe a broken line due to distortion that appears to swell outward from the center of the image.
The subject of this invention is improving the recognition accuracy of a traveling lane when a traffic division line is a broken line.

  A traveling lane recognition apparatus according to an aspect of the present invention images front and side traveling roads in a host vehicle, and detects edges of traffic marking lines marked on a road surface from the captured front and side images. . Further, according to the detected edge continuity, it is determined whether the traffic dividing line is a broken line, and when it is determined that the traffic dividing line is a broken line, the degree of detection of the edge detected in the front image, and The dashed line is complemented according to the magnitude relationship of the degree of detection of the edge detected in the side image. And a lane is recognized by this complementation result.

  According to the present invention, when the traffic dividing line is a broken line, interpolation of the broken line is performed according to the magnitude relationship between the degree of detection of the edge detected in the front image and the degree of detection of the edge detected in the side image. Thus, the recognition accuracy of the traveling lane can be improved.

It is a schematic block diagram which shows a traveling lane recognition apparatus. It is an arrangement plan of a front camera. It is an arrangement plan of a left camera and a right camera. It is a figure which shows the image imaged with each camera. It is a figure explaining the lane model. It is a figure which shows a mode that the lane model was projected on each image. It is a figure which shows a mode that the edge was detected in each image. It is a figure which shows a mode that the lane model was projected on each image which detected the edge. It is a figure explaining evaluation value e of a lane model. It is a figure which shows the evaluation value which evaluated the lane model in each image. e _{S_F} = 0, a diagram and shows the case of _{e S_SL>} 0. _{e S_F>} e _{S_SL,} and _| e _{S_F} -e S_SL _| is a diagram showing a case of _{a> e th.} It is a diagram showing a case of _{e S_F} ≒ e _{S_SL. e S_SL>} e _{S_F,} and _| e _{S_F} -e S_SL _| is a diagram showing a case of _{a> e th.} e _{S_SL} = 0, a diagram and shows the case of _{e S_F>} 0. It is a flowchart which shows a travel lane recognition process.

Hereinafter, embodiments of the present invention will be described with reference to the drawings.
<< First Embodiment >>
"Constitution"
The configuration of the traveling lane recognition device 11 will be described with reference to FIG.
The traveling lane recognition device 11 of the present embodiment includes a front camera 12F, a left camera 12SL, a right camera 12SR, and a controller 13.
The front camera 12F captures a front traveling path in the host vehicle and acquires a front image. The left camera 12SL captures a left travel path in the host vehicle and acquires a left image. The right camera 12SR captures a right travel path in the host vehicle and acquires a right image. The front camera 12, the left camera 12SL, and the right camera 12SR each use a fisheye lens. The front camera 12F, the left camera 12SL, and the right camera 12SR each output captured image data to the controller 13. Although a fisheye lens is used here, the present invention is not limited to this, and any lens may be used as long as a predetermined region can be imaged.

The arrangement of the front camera 12F will be described with reference to FIG.
The front camera 12F is provided, for example, on the front grille. The mounting position of the front camera 12F is the height of the lens h1 from the ground, and the mounting angle is tilted by θ1 with the optical axis downward with respect to the horizontal plane. .
The arrangement of the left camera 12SL and the right camera 12SR will be described with reference to FIG.
The left camera 12SL is provided, for example, on the left door mirror, and the mounting position thereof is such that the lens position is at a height of h2 from the ground, and the mounting angle is tilted by θ2 with the optical axis downward with respect to the horizontal plane. ing. The right camera 12SR is provided, for example, on the right door mirror. The mounting position of the right camera 12SR is such that the position of the lens is at a height of h3 from the ground, and the mounting angle is inclined by θ3 with the optical axis pointing downward with respect to the horizontal plane. ing. Note that h2 = h3 and θ2 = θ3.

Images taken by each camera are shown in FIG.
In both cases, using a fish-eye lens causes barrel-shaped distortion that appears to bulge outward from the center of the image. The closer the subject is, the larger the image is. The farther the subject is, the smaller the image is. .
Here, a front image, a left image, and a right image when the host vehicle is traveling in the right lane on a one-way two-lane road are shown. A traffic marking line (white line) is marked on the road surface. The left traffic dividing line LL in the left lane is marked with a solid line. The right traffic division line LS in the left lane (= left side in the right lane) is indicated by a broken line, and the right traffic division line LR in the right lane is indicated by a solid line. In the front image, the traffic division lines LL, LS, LR are shown, in the left image, the traffic division lines LL, LS are shown, and in the right image, the traffic division lines LS, LR are shown. Yes.

The controller 13 is composed of, for example, a microcomputer, and includes a lane model projection unit 21, an edge detection unit 22, a broken line determination unit 23, a model evaluation unit 24, a broken line complement unit 25, and a lane recognition unit 26. .
The lane model projection unit 21 projects a predetermined lane model on each of the front image, the left image, and the right image according to each parameter representing the position and orientation of the host vehicle in the lane. In order to estimate the optimum parameters for the lane model, a plurality of vehicle models in which each parameter is changed from the previous estimation result are prepared. Each parameter of the previous estimation result is output from the lane recognition unit 26.

The lane model will be described with reference to FIG.
(A) in the figure is an overhead view, and (b) in the figure is a side view. The lane model is given as a straight line model with four parameters. That is, the lane width wt, the lateral position xt, the yaw angle θt, and the pitch angle φt. The lane width wt is a distance in the lane width direction from the left traffic division line to the right traffic division line. The lateral position xt is a distance in the lane width direction from the center position of the lane width wt to the center position of the vehicle width at the front end of the vehicle body, for example. The yaw angle θt is an angle formed by the lane center line and the longitudinal direction of the vehicle body when the vehicle body is viewed in a plane. The pitch angle φt is an angle formed by the horizontal plane and the longitudinal direction of the vehicle when the vehicle is viewed from the side.

FIG. 6 shows a lane model projected on each image.
Here, the lane model projected on each image is represented by a thick dotted line for convenience. In the front image, a lane model MS _F corresponding to the right edge of the right traffic lane in the right lane and a lane model MR _F corresponding to the left edge of the right traffic lane in the right lane are projected. doing. In the left image, the vehicle model MS1 _SL corresponding to the right edge is projected on the right lane marking (= left side in the right lane) in the left lane. In the right image, the lane model MR1 _SR corresponding to the left edge of the right lane marking in the right lane is projected.

The edge detection unit 22 detects edges in the front image, the left image, and the right image. For edge detection, for example, a Sobel filter or the like is used, and the output value is binarized to obtain an edge image.
FIG. 7 shows how edges are detected in each image.
Here, the edge detected in each image is represented by a thick solid line for convenience. In the front image, the right edge EL _F constituting the left traffic division line in the left lane, the right edge ES _F constituting the right traffic road in the left lane (= left side in the right lane), and the right edge in the right lane and detects the left edge ER _F constituting the traffic separation line. In the left image, the right edge EL _SL constituting the left traffic division line in the left lane, the left edge ES1 _SL constituting the right traffic division line (= left side in the right lane) in the left lane, and the right edge ES2 _SL is detected. The right image, and detects the left edge ER1 _SR, and the right edge ER2 _SR constituting the right traffic lane marking in the right lane.

The broken line determination unit 23 determines whether or not the passage division line is a broken line.
Specifically, the lane model is projected on the edge image, and the number of pixels of the edge that rides (matches) on the lane model is counted. The length of the lane model is L, the ratio of the number of edge pixels m to the lane model length L is r (= m / L), and it is determined whether or not the ratio r is less than a predetermined threshold r1. To do. The threshold r1 is set to an optimum value from experiments or the like. Here, when the ratio r is greater than or equal to the threshold value r1, it is determined that the traffic line is not a broken line, that is, a solid line, and when the ratio r is less than the threshold value r1, it is determined that the traffic line is a broken line.

FIG. 8 shows a state in which the lane model is projected on each image where the edge is detected.
Although the lane model substantially coincides with the edge detected in each image, the lane model is drawn slightly shifted from the edge here for convenience. In the front image, in the left traffic dividing line in the right lane, there is no edge ES _F in the region A, and the ratio r of the number m of pixels of the edge ES _F to the length L of the lane model MS _F is low. It is determined that the traffic dividing line is a broken line. On the other hand, the right side of the traffic separation in the right lane, all edges ER _F overlaps the lane model MR _F, since the lane model MR _F of the length L against the ratio r of the pixel number m of edges ER _F increases, traffic separation line Is determined to be a solid line.

The left image in the left traffic lane marking in the right lane, all edges _{ES2 SL} overlaps the lane model _{MS SL,} the proportion r edge _{ES2 SL} number of pixels m becomes high with respect to the length L of the lane model _{MS SL} Based on the determination in the front image, it is determined that the traffic dividing line is a broken line. The right image, the right side of the traffic lane marking in the right lane, all edges _{ER1 SR} overlaps the lane model _{MR SR,} the proportion r edge _{ER1 SR} number of pixels m becomes high with respect to the length L of the lane model _{MR SR} It is determined that the traffic division line is a solid line.

As described above, when it is determined that the traffic dividing line is a broken line in any one of the front image, the left image, and the right image, the corresponding side traffic dividing line is also displayed in the remaining images. It determines with it being a broken line.
Here, the broken line determination is performed according to the ratio r of the number m of edge pixels to the length L of the lane model. However, the present invention is not limited to this. For example, the number of continuous pixels is counted and the length is determined. The broken line determination may be performed accordingly. In short, it is only necessary to evaluate the continuity of the edge and determine whether or not it is a broken line.

The model evaluation unit 24 calculates the degree of coincidence between the edge and the lane model as an evaluation value e _i of the lane model as shown in the following equation. The subscript i is the lane model number, and e _i is the i-th lane model. For example, in the front image, the right traffic line in the left lane, the left traffic line in the right lane, in the left image, the right traffic line in the left lane, in the right image, the left traffic line in the right lane When a lane model is configured for a line, i = 1 to 4. L is the length of the lane model. The subscript j is an element of a set constituting the lane model. dx _j and dy _j are the gradients of the lane model at the pixel [x _j , y _j ]. gx _j and gy _j are the luminance gradients of the image in the pixel [x _j , y _j ].

The lane model evaluation value e will be described with reference to FIG.
Here, a lane model MS _SL projected on the left image will be described as an example.
In the above equation, in pixel [xj, yj], the degree of coincidence between the direction in which the lane model extends and the direction in which the luminance in the image changes is obtained. Mathematically, the inner product of the direction in which the lane model extends and the direction in which the luminance changes is obtained for each pixel. Therefore, it is possible to robustly evaluate the degree of coincidence with the edge as compared with the method of simply evaluating the overlapping of the edges.

Evaluation values obtained by evaluating the lane model in each image are shown in FIG.
In the front image, the evaluation value of the lane model MS _F is calculated as e _{S_F} in the left traffic lane in the right lane, and the evaluation value of the lane model MR _F is calculated as e _{R_F} in the right traffic lane in the right lane. To do. In the left image, the evaluation value of the lane model MS _SL is calculated as e _{S_SL} on the left traffic lane (right side in the left lane) in the right lane. In the right image, the evaluation value of the lane model MR _SR is calculated as e _{R_SR} on the right traffic lane in the right lane.

  The broken line complementing unit 25 performs the dashed line complementation when it is determined that the traffic dividing line is a broken line. Here, when the detection level of the edge detected in the side image is larger than the detection level of the edge detected in the front image, the front image is complemented according to the detection level of the edge detected in the side image. I do. Further, when the edge detection level detected in the front image is larger than the edge detection level detected in the side image, the side image is complemented according to the edge detection level detected in the front image. Do.

The edge detection degree is the evaluation value e of the lane model. That is, when the evaluation value e _S of the lane model projected on the side image is larger than the evaluation value e _F of the lane model projected on the front image, the evaluation value e _S of the lane model projected on the side image is obtained. Accordingly, the evaluation value e _F of the lane model projected on the front image is increased and corrected. Thereby, for example, if the edge of the side image matches the lane model, the edge of the front image is evaluated as if it matches the lane model, so that the side image is complemented. Also, from the evaluation value e _S of the lane model projected laterally image, when is larger evaluation value e _F of the lane model projected forward image, depending on the evaluation value e _F of the lane model projected forward image Thus, the evaluation value e _S of the lane model projected on the side image is increased and corrected. Thereby, for example, if the edge of the front image matches the lane model, the edge of the side image is evaluated so as to match the lane model, so that the front image is complemented.

As shown below, the magnitude relationship between the evaluation value e _F of the lane model projected on the front image and the evaluation value e _S of the lane model projected on the side image can be divided into five cases. Here, using the evaluation value e _{S_F} of the lane model MS _F projected on the front image and the evaluation value e _{S_SL} of the lane model MS _SL projected on the left image, the left lane _marking in the right lane is taken as an example. explain.
1. When e _{S_F} = 0 and e _{S_SL} > 0, the evaluation value e _{S_F} of the lane model MS _F projected on the front image is 0, and the evaluation value e _{S_SL} of the lane model MS _SL projected on the left image is 0 FIG. 11 shows a large state.
Here, it is a scene in which the broken line cannot be observed with the front camera 12F and the broken line can be observed only with the left camera 12SL. As described above, when no edge is detected in the front image, the dashed line is not complemented.

2. When e _{S_F} > e _{S_SL} and | e _{S_F} −e _{S_SL} |> e _th The evaluation value e _{S_F} of the lane model MS _F projected on the front image is obtained from the evaluation value e _{S_SL} of the lane model MS _SL projected on the left image. It is large, and the difference _| shown in FIG. 12 the state is greater than the threshold value _{e th} a predetermined | e _{S_F -e S_SL.}
Here, the broken line can be observed by both the front camera 12F and the left camera 12SL, and more broken lines can be observed by the front camera 12F than by the left camera 12SL. Thus, when the detection level of the edge detected in the front image is larger than the detection level of the edge detected in the left image, the lane model MS _F projected on the front image is represented by the following equation. In accordance with the evaluation value e _{S_F} , the lane model MS _SL projected on the left image is treated as a passing division line by increasing and correcting the evaluation value e _{S_SL} of the lane model MS _SL projected on the left image.
e _{S_SL} ← e _{S_SL} + (α · e _{S_F} )

  α is calculated according to the following equation.

  The vectors g ′ and d ′ in the above formula are normalized so that the norm is 1 with respect to the gradient vector of the lane model and the luminance gradient in the image. L ′ is the length of an edge that rides (matches) on the lane model. In the above formula, [0, 1] is used to evaluate how much the luminance gradient of the pixel on the broken line in the left image matches the direction in which the lane model extends, regardless of the luminance intensity. Yes. In other words, the role of α assumes that the broken line observed with the front camera 12F and the broken line observed with the left camera 12SL are straight lines, and the lane models that are on the same straight line when complementing the low evaluation value curve The evaluation value is higher, and as a result, the higher the weight is on the same straight line, the higher the weight.

3. When e _{S_F} ≒ e _{S_SL} (when | e _{S_F−} e _{S_SL} | ≦ e _th )
The evaluation value e _{S_F} of the lane model MS _F projected on the front image is substantially the same as the evaluation value e _{S_SL} of the lane model MS _SL projected on the left image, that is, the difference | e _{S_F} −e _{S_SL} | FIG. 13 shows a state of e _th or less.
Here, a broken line can be observed by both the front camera 12F and the left camera 12SL, and the observed amounts are the same. In this way, when there is no superiority or inferiority in edge detection between the front image and the left image, the dashed line is not complemented.

4). When e _{S_SL} > e _{S_F} and | e _{S_F} −e _{S_SL} |> e _th The evaluation value e _{S_SL} of the lane model MS _SL projected on the left image is based on the evaluation value e _{S_F} of the lane model MS _F projected on the front image It is large, and the difference _| showing how is greater than the threshold value _{e th} a predetermined in Figure 14 | e _{S_F -e S_SL.}
Here, the broken line can be observed by both the front camera 12F and the left camera 12SL, and more broken lines can be observed by the left camera 12SL than by the front camera 12F. Thus, when the detection degree of the edge detected in the left image is larger than the detection degree of the edge detected in the front image, the lane model MS _SL projected on the left image as shown in the following equation. depending on the evaluation value e _{S_SL,} by increasing correcting the evaluation value e _{S_F} lane model MS _F projected forward image, dealing with lane model MS _F projected forward image as traffic separation line. α is the same as described above.
e _{S_F} ← e _{S_F} + (α · e _{S_SL} )

5. When e _{S_SL} = 0 and e _{S_F} > 0, the evaluation value e _{S_SL} of the lane model MS _SL projected on the left image is 0, and the evaluation value e _{S_F} of the lane model MS _F projected on the front image is 0 FIG.
Here, it is a scene in which the broken line cannot be observed with the left camera 12SL and the broken line can be observed only with the front camera 12F. As described above, when no edge is detected in the left image, interpolation of the broken line is not performed.
The above is the process of the broken line complement unit 25.

  The lane recognition unit 26 estimates the four parameters of the lane model using the evaluation value when the broken line is complemented. As an estimation method, the degree of coincidence between a plurality of lane models with different edges and parameters is calculated as an evaluation value of each lane model, and an optimized vehicle model is calculated. For example, a Particle Filter is used. However, the present invention is not limited to this, and any method may be used as long as a curve that optimally approximates the detection result of the detected edge and broken line is estimated to recognize the traffic dividing line. The parameters of the lane model estimated here are used for projection of the lane model in the next calculation according to the calculation cycle.

Next, a traveling lane recognition process that is calculated every predetermined time (for example, 10 msec) by the controller 13 will be described with reference to FIG.
First, in step S101, a front image captured by the front camera 12F, a left image captured by the left camera 12SL, and a right image captured by the right camera 12SR are acquired.
The subsequent step S102 corresponds to the processing in the edge detection unit 22, and detects edges in the front image, the left image, and the right image.

The subsequent step S103 corresponds to the processing in the lane model projection unit 21, and for each of the front image, the left image, and the right image according to each parameter representing the position and orientation in the lane of the host vehicle, A predetermined lane model is projected.
The subsequent step S104 corresponds to the processing in the model evaluation unit 24, and calculates the degree of coincidence between the edge and the lane model as an evaluation value ei of the lane model as shown in the following equation.
Subsequent step S105 corresponds to the process in the broken line determination unit 23, and determines whether or not the passage line is a broken line.

In the following step S106, corresponding to the processing in the broken line complementing unit 25, when it is determined that the traffic lane marking is a broken line, the lane model is recognized as the traffic lane marking, thereby complementing the broken line.
The subsequent step S107 corresponds to the processing in the lane recognition unit 26, and estimates the four parameters of the lane model using the evaluation value when the broken line is complemented, and then returns to the predetermined main program.
The above is the traveling lane recognition process.

<Action>
Next, the operation of the first embodiment will be described.
In this embodiment, the front and side traveling roads in the vehicle are individually imaged (step S101), and the edge of the traffic marking line marked on the road surface is detected from the captured front image and side image (step S101). S102). Then, in accordance with the continuity of the edge, it is determined whether or not the passage division line is a broken line (step S105). Further, a predetermined lane model is projected on the front image and the side image (step S103), and when the lane marking is a broken line, the lane model is recognized as the lane marking, thereby Complement is performed (step S106).

At this time, the dashed line is complemented in accordance with the magnitude relationship between the edge detection degree detected in the front image and the edge detection degree detected in the side image. In other words, the front image and the side image supplement the side where more broken lines can be observed and the side where only fewer broken lines can be observed when the edge detection level is superior or inferior.
First, when the detection level of the edge detected in the side image is larger than the detection level of the edge detected in the front image, the front image is complemented according to the detection level of the edge detected in the side image. Do. On the other hand, when the detection level of the edge detected in the front image is larger than the detection level of the edge detected in the side image, the side image is complemented according to the detection level of the edge detected in the front image. . Thus, the recognition accuracy of a traveling lane can be improved by using two cameras with different viewpoints and using a configuration that complements each other.

Here, the edge detection degree is an evaluation value e of the lane model representing the degree of coincidence between the edge and the lane model. That is, the longer the detected edge, the higher the degree of overlapping with the lane model, and the higher the evaluation value e of the lane model.
Therefore, when the evaluation value e _S of the lane model projected on the side image is larger than the evaluation value e _F of the lane model projected on the front image, the evaluation value e _S of the lane model projected on the side image is obtained. Accordingly, the lane model projected on the front image is treated as a passing division line by increasing and correcting the evaluation value e _F of the lane model projected on the front image. Specifically, the value (α · e _S ) obtained by multiplying α by the evaluation value e _S is added to the evaluation value e _F to perform increase correction. This α is a value obtained by evaluating how much the lane model and the edge in the front image coincide with each other using indices such as a model gradient and a luminance gradient. In this way, by considering the degree of coincidence between the lane model and the edge in the front image, the recognition accuracy of the traveling lane can be improved.

On the other hand, when the evaluation value e _F of the lane model projected on the front image is larger than the evaluation value e _S of the lane model projected on the side image, the evaluation value e _F of the lane model projected on the front image is used. Thus, the lane model projected on the side image is treated as a traffic line by increasing and correcting the evaluation value e _S of the lane model projected on the side image. Specifically, a value (α · e _F ) obtained by multiplying α by the evaluation value e _F is added to the evaluation value e _S to perform increase correction. This α is a value obtained by evaluating how much the lane model and the edge in the side image coincide with each other using an index called a model gradient and a luminance gradient. Thus, the recognition accuracy of the traveling lane can be improved by considering the degree of coincidence between the lane model and the edge in the side image.

  It should be noted that when either one of the front image and the side image cannot observe the broken line edge at all, the broken line is not complemented. In this embodiment, when an edge of a broken line can be observed to some extent, it is complemented, and when it is not observed at all, the complement is not performed. Thereby, it can suppress that the complementation accuracy of a broken line falls on a curved road etc. However, the present invention is not limited to this, and even if the edge of the broken line cannot be observed at all in either the front image or the side image, the broken line may be complemented.

  In addition, even when the degree of edge detection is approximately the same in the front image and the side image, the dashed line is not complemented. Therefore, if the edges are sufficiently observed in both the front image and the side image, for example, because each line segment constituting the broken line is long or the interval (gap) between the line segments is narrow, those edges are observed. Can be recognized as a passing line. On the other hand, for example, when the line segments constituting the broken line are short and the interval (gap) between the line segments is long, the edge cannot be sufficiently observed in both the front image and the side image. Therefore, it is possible to suppress a decrease in the interpolation accuracy of the broken line.

《Correspondence relationship》
In the present embodiment, the front camera 12F, the left camera 12SL, and the right camera 12SR correspond to an “imaging unit”. The processing of the edge detection unit 22 and step S102 corresponds to an “edge detection unit”. The process of the broken line determination unit 23 and step S105 corresponds to the “broken line determination unit”. The processing of the lane model projection unit 21 and step S103 corresponds to the “lane model projection unit”. The process of the broken line complementing unit 25 and step S106 corresponds to the “broken line complementing unit”. The process of the model evaluation unit 24 and step S104 corresponds to the “model evaluation unit”. The processing of the lane recognition unit 26 and step S107 corresponds to the “lane recognition unit”.

"effect"
Next, the effect of the main part in 1st Embodiment is described.
(1) The traveling lane recognition apparatus according to the present embodiment images the front and side traveling roads of the host vehicle, and detects the edge of the traffic division line marked on the road surface from the front image and the side image. Depending on the continuity of the edge, it is determined whether or not the traffic dividing line is a broken line. Further, when it is determined that the traffic dividing line is a broken line, the broken line is complemented according to the magnitude relationship between the detection degree of the edge detected in the front image and the detection degree of the edge detected in the side image.
In this way, when the traffic dividing line is a broken line, the broken lane can be complemented according to the magnitude relationship between the detection degree of the edge detected in the front image and the detection degree of the edge detected in the side image. Recognition accuracy can be improved.

(2) The traveling lane recognition device according to the present embodiment detects the edge detected in the side image when the detection level of the edge detected in the side image is larger than the detection level of the edge detected in the front image. Depending on the degree of detection, the front image is complemented.
As described above, when the detection degree of the edge detected in the side image is relatively large, complementation is performed on the front image using the detection degree, thereby improving the interpolation accuracy of the broken line.

(3) The traveling lane recognition device according to the present embodiment detects the edge detected in the front image when the edge detection degree detected in the front image is larger than the edge detection degree detected in the side image. Depending on the degree, the side image is complemented.
As described above, when the detection degree of the edge detected in the front image is relatively large, the complementation accuracy of the broken line can be improved by performing complementation on the side image using the detection degree.

(4) The traveling lane recognition apparatus according to the present embodiment projects a predetermined lane model on the front image and the side image, and calculates the degree of coincidence between the edge and the lane model as an evaluation value of the lane model. To do. The degree of edge detection is this evaluation value.
As described above, by using the evaluation value of the lane model representing the degree of coincidence between the edge and the lane model as the edge detection degree, it is possible to improve the interpolation accuracy of the broken line.

(5) The traveling lane recognition device according to the present embodiment projects a side lane image when the evaluation value of the lane model projected on the side image is larger than the evaluation value of the lane model projected on the front image. In accordance with the evaluation value of the lane model, the evaluation value of the lane model projected on the front image is increased and corrected, so that the lane model projected on the front image is treated as a passing division line.
As described above, when the evaluation value of the lane model projected on the side image is relatively large, the complementation accuracy of the broken line can be improved by complementing the front image using the evaluation value.

(6) The traveling lane recognition device according to the present embodiment, when the evaluation value of the lane model projected on the front image is larger than the evaluation value of the lane model projected on the side image, the lane projected on the front image. In accordance with the evaluation value of the model, the evaluation value of the lane model projected on the side image is increased and corrected, so that the lane model projected on the side image is treated as a passing division line.
As described above, when the evaluation value of the lane model projected on the front image is relatively large, complementation is performed on the side image using the evaluation value, so that the interpolation accuracy of the broken line can be improved.

(7) The traveling lane recognition method according to the present embodiment images the front and side traveling roads in the host vehicle, and detects the edge of the traffic marking line marked on the road surface from the front image and the side image. Depending on the continuity of the edge, it is determined whether or not the traffic dividing line is a broken line. Further, when it is determined that the traffic dividing line is a broken line, the broken line is complemented according to the magnitude relationship between the detection degree of the edge detected in the front image and the detection degree of the edge detected in the side image.
In this way, when the traffic dividing line is a broken line, the broken lane can be complemented according to the magnitude relationship between the detection degree of the edge detected in the front image and the detection degree of the edge detected in the side image. Recognition accuracy can be improved.

<< Second Embodiment >>
"Constitution"
This embodiment performs complementation more simply.
The apparatus configuration is the same as that of the first embodiment described above.
Here, the process in the broken line complement part 25 is demonstrated. Other processes are the same as those in the first embodiment described above, and detailed descriptions of common parts are omitted.
As shown below, the magnitude relationship between the evaluation value e _F of the lane model projected on the front image and the evaluation value e _S of the lane model projected on the side image is divided into three cases. Here, using the evaluation value e _{S_F} of the lane model MS _F projected on the front image and the evaluation value e _{S_SL} of the lane model MS _SL projected on the left image, the left lane _marking in the right lane is taken as an example. explain.

1. When e _{S_F} > e _{S_SL} The evaluation value e _{S_F} of the lane model MS _F projected on the front image is larger than the evaluation value e _{S_SL} of the lane model MS _SL projected on the left image (FIG. 12). e _{S_SL} may be substantially 0 (FIG. 15). That is, this is a scene in which more broken lines can be observed with the front camera 12F than with the left camera 12SL. Thus, when the detection level of the edge detected in the front image is larger than the detection level of the edge detected in the left image, the lane model MS _F projected on the front image is represented by the following equation. In accordance with the evaluation value e _{S_F} , the lane model MS _SL projected on the left image is treated as a passing division line by increasing and correcting the evaluation value e _{S_SL} of the lane model MS _SL projected on the left image.
e _{S_SL} ← e _{S_SL} + e _{S_F}

2. When e _{S_F} ≒ e _{S_SL} (when | e _{S_F−} e _{S_SL} | ≦ e _th )
The evaluation value e _{S_F} of the lane model MS _F projected on the front image is substantially the same as the evaluation value e _{S_SL} of the lane model MS _SL projected on the left image, that is, the difference | e _{S_F} −e _{S_SL} | This is a case where it is equal to or less than _eth (FIG. 13). That is, the broken line can be observed by both the front camera 12F and the left camera 12SL, and the amount of observation is the same. In this way, when there is no superiority or inferiority in edge detection between the front image and the left image, the dashed line is not complemented.

3. When e _{S_SL} > e _{S_F} The evaluation value e _{S_SL} of the lane model MS _SL projected on the left image is larger than the evaluation value e _{S_F} of the lane model MS _F projected on the front image (FIG. 14). e _{S_F} may be substantially zero (FIG. 11). That is, this is a scene in which more broken lines can be observed with the left camera 12SL than with the front camera 12F. Thus, when the detection degree of the edge detected in the left image is larger than the detection degree of the edge detected in the front image, the lane model MS _SL projected on the left image as shown in the following equation. depending on the evaluation value e _{S_SL,} by increasing correcting the evaluation value e _{S_F} lane model MS _F projected forward image, dealing with lane model MS _F projected forward image as traffic separation line.
e _{S_F} ← e _{S_F} + e _{S_SL}
The above is the configuration of this embodiment.

<Action>
Next, the operation of the second embodiment will be described.
Than the evaluation value e _F of the lane model projected forward image, when the direction of evaluation value e _S of the lane model projected laterally image is large, according to the evaluation value e _S of the lane model projected laterally image The lane model projected on the front image is treated as a passing lane marking by increasing and correcting the evaluation value e _F of the lane model projected on the front image. Specifically, the evaluation value e _S is added to the evaluation value e _F to perform increase correction. That is, the calculation of α is omitted as compared with the first embodiment described above. Therefore, the dashed line can be complemented more easily in the front image.

On the other hand, when the evaluation value e _F of the lane model projected on the front image is larger than the evaluation value e _S of the lane model projected on the side image, the evaluation value e _F of the lane model projected on the front image is used. Thus, the lane model projected on the side image is treated as a traffic line by increasing and correcting the evaluation value e _S of the lane model projected on the side image. Specifically, the evaluation value e _F is added to the evaluation value e _S to perform increase correction. That is, the calculation of α is omitted as compared with the first embodiment described above. Therefore, it is possible to more easily complement the broken line in the side image.
In the present embodiment, other parts common to the first embodiment described above are assumed to have the same operational effects, and detailed description thereof is omitted.

  Although the present invention has been described with reference to a limited number of embodiments, the scope of rights is not limited thereto, and modifications of the embodiments based on the above disclosure are obvious to those skilled in the art. Moreover, each embodiment can be adopted in any combination.

DESCRIPTION OF SYMBOLS 11 Traveling lane recognition apparatus 12F Front camera 12SL Left camera 12SR Right camera 13 Controller 21 Lane model projection part 22 Edge detection part 23 Broken line determination part 24 Model evaluation part 25 Broken line complement part 26 Lane recognition part

Claims (7)

An imaging unit for imaging the front and side traveling paths of the host vehicle;
An edge detection unit for detecting an edge of a traffic marking line marked on a road surface in a front image and a side image captured by the imaging unit;
A broken line determination unit that determines whether or not the passage division line is a broken line according to the continuity of the edges detected by the edge detection unit;
When the broken line determination unit determines that the traffic dividing line is a broken line, depending on the magnitude relationship between the degree of detection of the edge detected in the front image and the degree of detection of the edge detected in the side image , A dashed line complement part for performing dashed line completion,
A lane recognition unit for recognizing a lane based on the complemented result of the broken line ,
A lane model projection unit that projects a predetermined lane model on the front edge image and the side edge image in which the edge is detected;
A model evaluation unit that calculates the degree of coincidence between the edge and the lane model as an evaluation value of the lane model,
The traveling lane recognition device according to claim 1, wherein the detection degree of the edge is an evaluation value calculated by the model evaluation unit . The broken line complement part is
When the detection degree of the edge detected in the side image is larger than the detection degree of the edge detected in the front image, the front is determined according to the detection degree of the edge detected in the side image. The travel lane recognition apparatus according to claim 1, wherein the supplement is performed using an image. The broken line complement part is
When the detection level of the edge detected in the front image is larger than the detection level of the edge detected in the side image, the lateral detection is performed according to the detection level of the edge detected in the front image. The travel lane recognition apparatus according to claim 1, wherein the complement is performed using an image. The broken line complement part is
When the evaluation value of the lane model projected onto the side edge image is larger than the evaluation value of the lane model projected onto the front edge image, the evaluation value of the lane model projected onto the side edge image The lane model projected on the front edge image is treated as the traffic division line by increasing and correcting the evaluation value of the lane model projected on the front edge image according to the above . The travel lane recognition device according to any one of claims 3 to 4 . The broken line complement part is
When the evaluation value of the lane model projected onto the front edge image is greater than the evaluation value of the lane model projected onto the side edge image, the evaluation value of the lane model projected onto the front edge image Correspondingly, by increasing correcting the evaluation value of the lane model projected onto the lateral edge image, according to claim 1, characterized in that handle the lane model projected onto the lateral edge image as the traffic separation line The traveling lane recognition device according to any one of? An imaging unit for imaging the front and side traveling paths of the host vehicle;
An edge detection unit for detecting an edge of a traffic marking line marked on a road surface in a front image and a side image captured by the imaging unit;
A broken line determination unit that determines whether or not the passage division line is a broken line according to the continuity of the edges detected by the edge detection unit;
When the broken line determination unit determines that the traffic dividing line is a broken line, depending on the magnitude relationship between the degree of detection of the edge detected in the front image and the degree of detection of the edge detected in the side image , A dashed line complement part for performing dashed line completion,
A lane recognition unit for recognizing a lane based on the complemented result of the broken line,
The broken line complement part is
When the detection degree of the edge detected in the side image is larger than the detection degree of the edge detected in the front image, the front is determined according to the detection degree of the edge detected in the side image. Perform the above completion with images,
When the detection level of the edge detected in the front image is larger than the detection level of the edge detected in the side image, the lateral detection is performed according to the detection level of the edge detected in the front image. A travel lane recognition device characterized in that the supplement is performed with an image . The imaging unit provided in the traveling lane recognition device individually images the front and side traveling paths in the host vehicle,
The edge detection unit provided in the traveling lane recognition device detects an edge of a traffic marking line marked on a road surface in a captured front image and a side image,
The broken line determination unit included in the traveling lane recognition device determines whether the passage line is a broken line according to the detected continuity of the edge,
The broken line complementing unit included in the traveling lane recognition device is configured such that when the traffic lane marking is a broken line, the degree of detection of the edge detected in the front image and the degree of detection of the edge detected in the side image Depending on the relationship, the dashed line is complemented,
The lane recognition unit provided in the traveling lane recognition device recognizes a lane based on the interpolation result of the broken line ,
The lane model projection unit provided in the traveling lane recognition device projects a predetermined lane model on the front edge image and the side edge image in which the edge is detected,
The model evaluation unit provided in the traveling lane recognition device calculates the degree of coincidence between the edge and the lane model as an evaluation value of the lane model,
The traveling lane recognition method, wherein the edge detection degree is an evaluation value calculated by the model evaluation unit .

Images