JP2016162323A - Travelling section line recognition device - Google Patents

Abstract

PROBLEM TO BE SOLVED: To provide a travelling section line recognition device that can suppress wrong recognition of a travelling section line in a connection part between a straight part of a vehicle lane and a curve part thereof.SOLUTION: A travelling section line recognition device comprises: extraction means that extracts a section line candidate from a photograph image by an on-vehicle camera 10; degree-of-confidence calculation means that calculates a degree of confidence which is a travelling section line of the section line candidate on the basis of an amount of characteristic of the travelling section line including a vehicle lane width; selection means that selects the section line candidate serving as a recognition object from the section line candidates on the basis of the calculated degree of confidence; learning means that learns a width of the vehicle lane on the basis of the selected section line candidate to acquire a learning value, the learning value of the width of the vehicle lane is used by the degree-of-confidence calculation means for calculation of the degree of confidence; determination means that determines transition from a straight line part of the vehicle lane to a curve part thereof, and transition from the curve part to the straight line part; and change means that, when the transition from the straight line part to the curve part is determined, expands the learning value, and when the transition from the curve part to the straight line part is determined, narrows down the learning value of the vehicle lane width.SELECTED DRAWING: Figure 2

Description

  The present invention relates to a travel lane marking recognition device that recognizes a road lane marking from an image taken by an in-vehicle camera.

  The in-vehicle image processing apparatus described in Patent Literature 1 learns the lane width based on the distance between the left and right white lines recognized in the past, and the distance between the left and right white line candidates among the plurality of white line candidates extracted from the image is The left and right white line candidates closest to the learned lane width are selected and recognized as the left and right white lines.

JP-A-9-33216

  Generally, the lane width is designed in the curved portion of the lane rather than the straight portion. Therefore, when a white line candidate is selected using the learned value of the lane width in the straight line part at the connecting part from the straight line part to the curved line part, the white line candidate closer to the learned value of the lane width in the straight line part than the original white line There is a risk of misunderstanding. In addition, when a white line candidate is selected using the learned value of the lane width in the curved part at the connecting part from the curved part to the straight line part, the white line candidate closer to the learned value of the lane width in the curved part is white line than the original white line. There is a risk of misunderstanding.

  In view of the above circumstances, it is a main object of the present invention to provide a travel lane marking recognition device that can suppress erroneous recognition of a travel lane marking at a connection portion between a straight line portion and a curved portion of a lane.

  In order to solve the above problems, the present invention provides an extraction means for extracting a lane line candidate that is a candidate for a lane line that divides a road lane from an image taken by a camera mounted on the vehicle, Based on the feature amount of the travel lane line including the width, calculated by the confidence level calculation means for calculating the confidence level that is the travel lane line of the lane line candidate extracted by the extraction means, and calculated by the confidence level calculation means Based on the certainty factor, the selection means for selecting the lane line candidates to be recognized from the lane line candidates extracted by the extraction means, and the lane line candidates selected by the selection means Learning means for learning the width of the lane and acquiring a learning value, and the certainty factor calculating means is a travel lane marking recognition device that calculates the certainty factor using the learned value, The car Determining means for determining the transition from the straight line part to the curved part and the transition from the curved part to the straight line part of the lane, and when the judging means determines the transition from the straight line part to the curved part of the lane, The learning value used by the certainty factor calculating means is widened, and the learning value used by the certainty factor calculating means is narrowed when the determination means determines the transition from the curved portion to the straight line portion of the lane. Changing means.

  According to the present invention, a lane line candidate that is a candidate for a lane line for travel is extracted from an image captured by a camera. Then, using the learned value of the lane width, the certainty factor that is the travel lane line of the extracted lane line candidate is calculated based on the feature amount of the travel lane line including the lane width. Based on the calculated certainty factor, a lane line candidate to be recognized is selected from the extracted lane line candidates, and the selected lane line candidate is recognized as a travel lane line. Further, the lane width is learned based on the selected lane line candidate, and the learned value is acquired.

  Furthermore, when the transition from the straight line portion to the curved portion of the lane is determined, the learning value used for calculating the certainty factor is widened. Therefore, when moving from a straight line part to a curved part with a lane width that is generally wider than the straight line part, the confidence value of the lane line candidate is calculated using a learning value that is wider than the straight line part. False recognition of lane markings can be suppressed. In addition, when it is determined that the lane is shifted from the curved portion to the straight portion, the learning value used for calculating the certainty factor is narrowed. Therefore, when transitioning from a curved portion to a straight portion having a lane width that is generally narrower than the curved portion, the certainty factor of the lane line candidate is calculated using the learned value of the lane width narrower than the curved portion. Therefore, misrecognition of the travel lane marking can be suppressed. Therefore, it is possible to suppress erroneous recognition of travel lane markings at the connection portion between the straight line portion and the curved portion of the lane. As a result, unnecessary departure warning and steering control can be suppressed.

The figure which shows the mounting position of a vehicle-mounted camera and sensors. The block diagram which shows the function of the white line recognition apparatus which concerns on this embodiment. The figure which shows the white line reliability regarding a lane width. The figure which shows the aspect which misrecognizes a white line at the exit of a curve. The flowchart which shows the process sequence which recognizes a white line.

  Hereinafter, an embodiment that embodies a lane marking recognition device will be described with reference to the drawings. First, with reference to FIG. 1 and 2, the traveling lane marking recognition apparatus according to the present embodiment will be described. The travel lane marking recognition device according to the present embodiment is an in-vehicle device that is configured by the ECU 20 and recognizes a white line (travel lane marking) that divides a road lane based on a front image taken by the in-vehicle camera 10. In the present embodiment, white lines, yellow lines, and the like drawn to indicate lane divisions on the road are all referred to as white lines.

  The in-vehicle camera 10 includes at least one such as a CCD image sensor or a CMOS sensor. As shown in FIG. 1, the in-vehicle camera 10 is installed, for example, in the vicinity of the upper end of the windshield of the vehicle 50, and photographs an area that spreads in a predetermined angle range toward the front of the vehicle 50. That is, the in-vehicle camera 10 captures the surrounding environment including the road ahead of the vehicle 50.

  The vehicle speed sensor 11 is mounted on the vehicle 50 and is a sensor that detects the speed of the vehicle 50. The yaw rate sensor 12 is mounted on the vehicle 50 and is a sensor that detects the yaw rate of the vehicle 50.

  The driving support device 40 is a device including a departure warning device that warns of a lane departure based on the recognition result of the white line recognized by the ECU 20 and a driving support device that performs driving support. The departure warning device is configured as a human machine interface such as a display, a speaker, and a vibrator, and outputs a warning to the driver when the vehicle 50 departs from the lane. Further, the driving support device is configured as a steering actuator or a braking actuator, and performs steering control and brake control of the vehicle 50.

  The ECU 20 is a computer that includes a CPU, a RAM, a ROM, an I / O, a storage device, and the like. When the CPU executes various programs stored in the ROM, the white line candidate extraction unit 21, the curvature calculation unit 22, the curve portion determination unit 23, the learning value change unit 24, the lane width learning unit 25, and the white line certainty calculation The functions of the unit 26, the candidate selection unit 27, and the recognition unit 28 are realized.

  The white line candidate extraction unit 21 (extraction means) applies a sobel filter or the like to the front image captured by the in-vehicle camera 10 to extract an up-edge point and a down-down point where the luminance value increases significantly in the horizontal direction. . Then, the white line candidate extraction unit 21 calculates an edge line by applying Hough transform or the like to the extracted edge point, and determines a line specified from the edge line including the up edge point and the edge line including the dow edge point as a white line Extracted as candidates (division line candidates).

  The white line certainty calculation unit 26 (confidence calculation means) calculates a certainty factor (white line likelihood) that is a white line of the white line candidate extracted by the white line candidate extraction unit 21 based on the white line feature amount. The white line feature amounts include lane width consistency, recognition distance, white line contrast with the road surface, and the like. The white line certainty degree calculation unit 26 calculates the certainty degree that is a white line higher as the degree of white line candidates having the feature amount is higher for each feature amount. And the white line reliability calculation part 26 integrates the reliability calculated about each feature-value, and calculates an integrated reliability.

  The candidate selecting unit 27 (selecting means) is a recognition target, that is, a control target by the driving support device 40, from the white line candidates extracted by the white line candidate extracting unit 21 based on the certainty factor calculated by the white line certainty factor calculating unit 26. Select white line candidates. Specifically, the candidate selection unit 27 selects the white line candidate having the highest certainty among the white line candidates having the integrated certainty level higher than the threshold on the right side and the left side of the vehicle 50.

  The recognition unit 28 recognizes the white line candidate selected by the candidate selection unit 27 as a white line, calculates a white line parameter, and outputs the calculated white line parameter to the driving support device 40. The white line parameters are lane position, lane inclination, lane curvature, lane width, pitch angle, and the like.

  The lane width learning unit 25 (learning means) learns the lane width based on the white line candidate selected as a recognition target by the candidate selection unit 27 and acquires a learning value. Specifically, the lane width learning unit 25 calculates a new learning value by performing predetermined weighting on the lane width recognized by the recognition unit 28 and the learning value of the lane width stored in the storage device, A new learning value is stored in the storage device.

  The white line certainty calculation unit 26 described above uses the learned value of the lane width acquired by the lane width learning unit 25 to calculate the certainty factor using the consistency of the lane width as the feature amount of the white line. As shown in FIG. 3, the white line certainty calculation unit 26 sets the lane width consistency so that the distance between the white line candidates on the right and left sides of the vehicle 50 increases as the learning value of the lane width increases. The certainty factor is calculated.

  Here, as shown in FIG. 4, the lane width of the curved portion (curve portion) of the lane is generally designed to be wider than the lane width of the straight portion of the lane. For example, in Japan, Article 17 of the Road Structure Ordinance stipulates that “in the curved part of the roadway, the lane should be appropriately widened according to the design vehicle and the curved radius of the curved part”.

  Therefore, using the learned value of the lane width learned at the straight line part of the lane, the certainty of the white line candidate of the curved part of the lane is calculated, or the learned value of the lane width learned at the curved part of the lane is used to calculate the lane If the certainty factor of the white line candidate in the straight line portion is calculated, the white line may be erroneously recognized.

  For example, as shown in FIG. 4, the lane width of the curved portion is designed to be W + α and the lane width of the straight portion is W, and the exit side of the curved portion, that is, the front side of the connecting portion between the straight portion and the curved portion. In some cases, a bicycle-only road having a width α may be provided in parallel with the lane having a width W in the direction of travel of the vehicle 50. In this case, the width W + α obtained by adding the lane width W of the straight line portion and the width α of the bicycle exclusive road is closer to the learning value of the lane width of the curved portion than the lane width W of the straight line portion. Therefore, using the learned value of the lane width learned at the curve portion, the white line candidate corresponding to the white line Lα that defines the right side of the bicycle exclusive road is integrated rather than the white line candidate corresponding to the white line Lr on the right side of the lane. There is a risk of increased confidence. As a result, the white line Ll on the left side of the lane and the white line Lα on the right side of the bicycle exclusive road may be recognized as the white lines on the left and right sides of the lane.

  Therefore, the learning value of the lane width is changed in the vicinity of the connecting portion between the straight line portion and the curved portion of the lane, that is, in the vicinity of the entrance / exit of the curved portion. Hereinafter, the change of the learning value of the lane width will be described.

  The curvature calculation unit 22 calculates the curvature of the lane ahead of the vehicle 50 from the change in the vehicle speed detected by the vehicle speed sensor 11 and the yaw rate of the vehicle 50 detected by the yaw rate sensor 12. Generally, before the vehicle enters the curved portion, the steering wheel of the vehicle is operated so as to match the curved portion, and the speed is reduced. Further, before the vehicle escapes from the curved portion, the vehicle handle is operated so as to match the straight portion, and the speed is increased. Therefore, the curvature of the lane in front of the vehicle 50 can be calculated from the change in the yaw rate of the vehicle 50 and the vehicle speed. Further, the curvature calculation unit 22 calculates a curvature change rate from the curvature of the lane calculated this time and the curvature of the lane calculated last time.

  Based on the curvature of the lane calculated by the curvature calculation unit 22, the curved line determination unit 23 shifts from the straight part of the lane to the curved part, and the transition from the curved part of the lane to the straight part. Determine. In other words, the curved line determination unit 23 determines that it is in the vicinity of the entrance of the curved part, which is a connection part between the straight line part and the curved part of the lane, and in the vicinity of the exit of the curved part.

  Specifically, the curve part determination unit 23 determines the transition from the straight line part to the curved part and the transition from the curved part to the straight line part from at least one of the curvature of the lane and the curvature change rate calculated by the curvature calculation part 22. To do. Generally, in the vicinity of the entrance of the curved portion, the curvature increases from a value close to 0, and the absolute value of the curvature change rate (curvature increase rate) increases. Further, in the vicinity of the exit of the curved portion, the curvature decreases to a value close to 0, and the absolute value (curvature decrease rate) of the curvature change rate increases. Therefore, the curve portion determination unit 23 shifts from the straight line portion to the curved portion in at least one of the case where the curvature is larger than the curvature threshold and the case where the curvature increase rate is larger than the change amount threshold. Determine. Further, the curve portion determination unit 23 shifts from the curve portion to the straight portion in at least one of the case where the curvature is smaller than the curvature threshold and the case where the curvature reduction rate is larger than the change amount threshold. Determine. The curvature calculating unit 22 and the curve part determining unit 23 constitute a determining unit.

  The learning value changing unit 24 (changing unit) sets the learning value of the lane width used by the white line certainty calculation unit 26 when the curve portion determining unit 23 determines the transition from the straight line portion to the curved portion of the lane. spread. Further, the learning value changing unit 24 narrows the learning value of the lane width used by the white line certainty calculating unit 26 when the curve determining unit 23 determines the transition from the curved line to the straight line.

  Specifically, the learning value changing unit 24 has been learned by the lane width learning unit 25 until the determination is made when the curve portion determining unit 23 determines the transition from the straight line portion of the lane to the curved portion. Correct the lane width learning value to increase. In addition, the learning value changing unit 24, when the curve portion determining unit 23 determines that the lane is shifted from the curved portion to the straight portion, the lane learned by the lane width learning unit 25 until the determination is made. The learning value of the width is corrected so as to be narrowed.

  In general, the curve portion having a larger curvature is designed to have a wider lane width. Therefore, the learning value changing unit 24 changes the change amount of the learning value of the lane width based on the curvature of the lane. Specifically, the learning value changing unit 24 determines the amount of change in the learning value of the lane width from at least one of the curvature of the lane and the curvature change rate. The learning value changing unit 24 increases the amount of change as the curvature when the transition from the straight line portion to the curved portion of the lane is determined is large, or as the absolute value of the curvature change rate when the transition is determined is large. Enlarge. Further, the learning value changing unit 24 changes as the curvature before the transition from the curved portion to the straight portion of the lane is determined is large or as the absolute value of the curvature change rate when the transition is determined is large. Increase the amount. In one curved portion, the amount of change in the learned value of the lane width in the vicinity of the entrance of the curved portion and the amount of change in the learned value of the lane width in the vicinity of the exit of the curved portion are not necessarily the same amount.

  Next, a processing procedure for recognizing a white line will be described with reference to the flowchart of FIG. This processing procedure is executed by the ECU 20 every time an image of one frame is acquired by the in-vehicle camera 10.

  First, the information of the image image | photographed with the vehicle-mounted camera 10 is acquired (S10). Subsequently, white line candidates are extracted from the image information acquired in S10 (S11).

  Subsequently, the curvature of the lane and the curvature change rate are calculated from the change in the vehicle speed detected by the vehicle speed sensor 11 and the yaw rate of the vehicle 50 detected by the yaw rate sensor 12 (S12). Subsequently, from at least one of the curvature and curvature change rate calculated in S12, it is determined whether or not the vehicle is near the entrance / exit of the curved portion of the lane (S13). That is, it is determined whether or not there is a transition from a straight line part to a curved line part or a transition from a curved part to a straight line part. When it is determined that it is near the entrance / exit of the curved portion of the lane (S13: YES), the learning value of the lane width is changed (S14).

  Specifically, when it is determined that it is in the vicinity of the entrance of the curved portion, correction is made so that the learning value of the lane width stored in the storage device is increased by an amount of change corresponding to at least one of the curvature of the lane and the curvature change rate. . Further, when it is determined that the vehicle is near the exit of the curved portion, correction is performed so that the learning value of the lane width stored in the storage device is narrowed by an amount of change corresponding to at least one of the curvature of the lane and the curvature change rate. Then, the process proceeds to processing for narrowing down the next white line candidate. On the other hand, when it is determined that it is not near the entrance / exit of the curved portion of the lane (S13: NO), the process proceeds to the next white line candidate narrowing-down process without correcting the learning value of the lane width stored in the storage device.

  Subsequently, in each of the left and right sides of the vehicle 50, the white line candidates extracted in S11 are narrowed down to white line candidates that are likely white lines. Specifically, the certainty factor that the white line candidate is a white line is calculated for each feature amount of the white line, and the certainty factor calculated for each feature amount of the white line is integrated to calculate the integrated certainty factor. Then, in each of the left and right sides of the vehicle 50, the integrated line reliability is narrowed down to white line candidates that are higher than the threshold and have the highest reliability. Here, when the certainty factor is calculated using the white line feature amount as the lane width consistency, if it is determined in S13 that it is not near the entrance / exit of the curved portion, the certainty factor is calculated using the uncorrected learning value of the lane width. If it is calculated and it is determined in S13 that it is in the vicinity of the entrance / exit of the curved portion, the certainty factor is calculated using the learning value of the lane width corrected in S14.

  Subsequently, the white line candidates narrowed down in S15 are recognized as white lines, and white line parameters are calculated (S16). Subsequently, the lane width among the white line parameters calculated in S16 is learned (S17). If it is determined in S13 that it is in the vicinity of the entrance of the curved portion of the lane, a new learning value is calculated from the learning value widened in S14 and the lane width calculated in S16. When it is determined in S13 that the vehicle is near the exit of the curved portion of the lane, a new learning value is calculated from the learning value narrowed in S14 and the lane width calculated in S16. This process is complete | finished above.

  According to this embodiment described above, the following effects are obtained.

  When the transition from the straight line portion to the curved portion of the lane is determined, the learning value of the lane width used for the certainty factor calculation is widened. As a result, when moving from a straight line part of a lane to a curved part that is generally wider than the straight line part, the confidence value of the white line candidate is calculated using a learning value that is wider than the learning value in the straight line part. Therefore, erroneous recognition of white lines can be suppressed. In addition, when it is determined that the lane is shifted from the curved line portion to the straight line portion, the learning value of the lane width used for calculating the certainty factor is narrowed. As a result, when transitioning from a curved part of a lane to a straight part where the lane width is generally narrower than the curved part, the certainty factor of the white line candidate is calculated using the learning value narrower than the learned value in the curved part. Therefore, the erroneous recognition of the travel lane marking can be suppressed. Therefore, erroneous recognition of the white line can be suppressed near the entrance / exit of the curved portion of the lane. As a result, unnecessary departure warning and steering control can be suppressed near the entrance / exit of the curved portion of the lane.

  ・ Generally, the lane width is a value corresponding to the curvature of the lane. Therefore, when changing from the straight line part to the curved part and changing from the curved part to the straight part by changing the amount of change of the learning value of the lane width according to at least one of the curvature of the lane and the curvature change rate It can be a learning value of an appropriate lane width. As a result, erroneous recognition of the white line can be appropriately suppressed near the entrance / exit of the curved portion of the lane.

  -By appropriately correcting the learned value of the lane width in the vicinity of the entrance / exit of the curved portion of the lane, it is possible to suppress erroneous recognition of the travel lane line in the vicinity of the entrance / exit of the curved portion of the lane.

  -The curvature is greatly different between the straight part and the curved part of the lane. Therefore, based on at least one of the curvature of the lane and the curvature change rate, it is possible to determine the transition from the straight line portion of the lane to the curved portion and the transition from the curved portion of the lane to the straight line portion.

  Generally, before the vehicle enters the curved portion from the straight line portion of the lane, the vehicle handle is operated to reduce the speed. Further, before the vehicle enters the straight line portion from the curved portion of the lane, the vehicle handle is operated to increase the speed. Therefore, the curvature of the lane ahead of the vehicle 50 can be calculated from changes in the yaw rate and vehicle speed of the vehicle 50.

(Other embodiments)
The curvature calculating unit 22 may calculate the curvature of the lane ahead of the vehicle 50 from the white line candidates extracted by the white line candidate extracting unit 21. When calculating the curvature of the lane from the white line candidate, it is possible to calculate the curvature of the lane far from the vehicle 50 than when calculating the curvature of the vehicle speed from the change in the yaw rate or the vehicle speed. In this case, the curvature of the lane may be calculated using a white line candidate with a relatively high certainty factor calculated based on the feature amount of the white line other than the consistency of the lane width. As shown in FIG. 4, when a bicycle exclusive road is provided in parallel with the lane from the vicinity of the exit of the curved portion to the traveling direction, both of the white line candidates corresponding to the white lines Lr and Lα are consistent in the lane width. There is a possibility that the certainty calculated based on the feature amount of the white line other than is relatively high. However, since both white line candidates are straight, there is no problem in calculating the curvature of the lane with either one.

  ・ The learning value of the lane width for the straight line part and the learned value of the lane width for the curved part are separated, and the lane width is separately learned for each of the straight line part and the curved part. When the shift to is determined, the lane width learning value for the straight line portion may be switched to the learning value for the curve portion without correcting the lane width learning value. In this case, as the learning value for the lane width for the curved portion, a plurality of learning values that are wider than the learning value for the straight portion are prepared in stages, and according to at least one of the curvature of the lane and the curvature change rate. Thus, the learning value of the lane width for the curved portion may be selected and switched. Then, when the transition from the curved portion of the lane to the straight portion is determined, the learned value of the lane width is not corrected, but the learned value of the lane width for the curved portion is changed to the learned value of the lane width for the curved portion. Switch. Even if it does in this way, the erroneous recognition of the white line in the vicinity of the entrance / exit of the curve part of a lane can be suppressed.

  10: In-vehicle camera, 20: ECU, 50: Vehicle.

Claims (6)

Extraction means for extracting lane line candidates that are candidates for travel lane lines that divide road lanes from an image taken by a camera (10) mounted on the vehicle (50);
A certainty factor calculating means for calculating a certainty factor that is the traveling lane line of the lane line candidate extracted by the extracting means, based on the feature amount of the traveling lane line including the width of the lane;
A selection unit that selects the lane line candidate to be recognized from the lane line candidates extracted by the extraction unit based on the certainty factor calculated by the certainty factor calculation unit;
Learning means for learning a width of the lane and acquiring a learning value based on the lane marking candidates selected by the selection means;
With
The certainty factor calculating means is a travel lane marking recognition device (20) that calculates the certainty factor using the learned value,
Determination means for determining the transition from the straight line portion of the lane to the curved portion and the shift from the curved portion of the lane to the straight portion;
When the determination unit determines that the lane is shifted from the straight line part to the curved line part, the learning value used by the certainty factor calculating part is widened, and the determination part changes the lane from the curved line part to the straight line part. A changing means for narrowing the learning value used by the certainty factor calculating means when the transition is determined,
A travel lane marking recognition device comprising:   The travel lane line recognition device according to claim 1, wherein the change unit changes a change amount of the learning value used by the certainty factor calculation unit based on a curvature of the lane.   The changing means corrects the learning value learned by the learning means to be widened when the judging means judges the transition from the straight line portion to the curved portion of the lane, and the judging means The travel lane marking recognition device according to claim 1 or 2, wherein when the transition from the curved portion of the lane to the straight portion is determined, correction is performed so as to narrow the learned value learned by the learning means.   4. The method according to claim 1, wherein the determination unit determines the transition from the straight line portion to the curved portion of the lane and the shift from the curved portion to the straight portion of the lane based on the curvature of the lane. The travel lane marking recognition device described.   The travel lane line recognition device according to any one of claims 1 to 4, wherein the determination unit calculates a curvature of the lane from the lane line candidates extracted by the extraction unit.   5. The travel lane marking recognition device according to claim 1, wherein the determination unit calculates a curvature of the lane from a change in a yaw rate of the vehicle and a vehicle speed.

Images