JP2018173762A - Traffic lane information generation device, traffic lane information generation method, and program for traffic lane information generation - Google Patents

Abstract

PROBLEM TO BE SOLVED: To provide a new traffic lane information generation device that can automatically generate traffic lane network data indicating the connection mode between traffic lanes included in the roads.SOLUTION: A traffic lane information generation device comprises: an extraction part 1 that extracts lines captured in a real image obtained by picking up an image of the roads, from ortho image data corresponding to the real image; a comparison part 2 that compares the interval in a direction intersecting the roads between the lines arranged side by side in the intersecting direction with the width of traffic lanes included in the roads on the basis of the extracted lines; and a generation part 3 that generates traffic lane networks indicating the traffic lanes on the basis of a result of comparison performed by the comparison part 2.SELECTED DRAWING: Figure 2

Description

  The present application belongs to a technical field of a lane information generation device, a lane information generation method, and a lane information generation program. More specifically, the present invention belongs to a technical field of a lane information generation device and a lane information generation method for generating lane information indicating a lane (so-called lane) included in a road, and a program for the lane information generation device.

  In recent years, research on so-called automatic driving has been actively conducted on navigation devices that guide the movement of vehicles. On the other hand, in the conventional navigation device, for example, road data including link data and node data is used to search for the shortest route to the destination. As an example of the prior art for automatically creating such road data, there is a technique described in Patent Document 1 below.

  In the technique described in Patent Document 1, a road area corresponding to a road is extracted from image data corresponding to an aerial photograph obtained by photographing the ground from an aircraft, and the connection state between the roads is determined based on the extraction result. Data of a road link (also referred to as “road network”) is generated and used as the road data. In the technique described in Patent Document 1, the width (road width) and the number of lanes (lanes) of each road are acquired / recorded in advance as auxiliary data other than the image data corresponding to the aerial photograph. The road area is determined and the lane boundary line is extracted from the image data.

JP 2008-170611 A

  However, in the above-mentioned Patent Document 1, although it is described that road link data indicating a connection mode between roads is generated, lane network data indicating a connection mode between lanes included in each road is described. There is no disclosure or suggestion about the configuration to be generated. On the other hand, in order to realize the automatic driving, it is required to use the data of the lane network for each lane included in the road instead of the road.

  Therefore, the present application has been made in view of the above request, and an example of the problem is a new lane capable of automatically generating lane network data indicating a connection mode between lanes included in a road. An object is to provide an information generation device, a lane information generation method, and a program for the lane information generation device.

  In order to solve the above-described problem, the invention according to claim 1 is an extraction means for extracting line segment information indicating a line segment reflected in a real image obtained by imaging a road from real image information corresponding to the real image. And comparing the distance between the intersecting directions of the line segments arranged in the direction intersecting the road with the width as a lane included in the road based on the extracted line segment information Means and generating means for generating lane information indicating the lane based on a comparison result by the comparing means.

  In order to solve the above-described problem, the invention according to claim 7 is a lane information generation method executed in a lane information generation apparatus that generates lane information indicating a lane included in a road, and images the road. The line segment information indicating the line segment shown in the actual image is extracted from the real image information corresponding to the real image, and aligned in the direction intersecting the road based on the extracted line segment information. It includes a comparison step of comparing an interval between the line segments in the intersecting direction and a width as the lane, and a generation step of generating the lane information based on a comparison result in the comparison step.

  In order to solve the above-described problem, the invention described in claim 8 shows a computer included in a lane information generation device that generates lane information indicating a lane included in a road reflected in an actual image obtained by imaging the road. Extraction means for extracting line segment information indicating a line segment from actual image information corresponding to the actual image, based on the extracted line segment information, the line segments arranged in a direction intersecting the road It is made to function as a comparison means for comparing the interval in the intersecting direction and the width as the lane, and a generation means for generating the lane information based on a comparison result by the computer functioning as the comparison means.

It is a block diagram which shows schematic structure of the lane information generation apparatus which concerns on embodiment. It is a block diagram which shows schematic structure of the lane network production | generation apparatus which concerns on an Example. It is a flowchart which shows the lane network production | generation process which concerns on an Example. It is a figure which illustrates the ortho image data etc. which concern on an Example, (a) is a figure which illustrates the said ortho image data, (b) is a figure which illustrates the said ortho image data about a road surface area, ( (c) is a figure which illustrates the data of the skeleton line which concerns on an Example, (d) is a figure which illustrates the position data of the road marking which concerns on an Example. It is a figure which illustrates the relationship etc. of the ortho image data and road link which concern on an Example, (a) is a figure which illustrates the said relationship, (b) is a figure which illustrates the data of the skeleton line which concerns on an Example It is. It is a figure for demonstrating the up-and-down line probability which concerns on an Example, (a) is a figure for demonstrating the up-and-down line probability in the case of Ichijo road, (b) is the up-and-down line probability in the case of a Nijo road. It is a figure for demonstrating. It is a figure which illustrates the long line concerning an example, (a) is a figure which illustrates the data of the skeleton line concerning an example, (b) is a figure which illustrates the outside line concerning an example, (c ) Is a diagram illustrating a long line according to the embodiment, (d) is a diagram illustrating a short line according to the embodiment, and (e) is a diagram illustrating a vertical line according to the embodiment. It is a figure which illustrates situations, such as a long line concerning an example. It is a figure which illustrates details, such as a long line concerning an example, (a) is a figure which illustrates the details, and (b) is a figure which illustrates ortho image data of a corresponding point. It is a figure which illustrates the data of the pedestrian crossing which concerns on an Example, (a) is a figure which illustrates the short line which concerns on an Example, (b) is a figure which illustrates the data of the pedestrian crossing which concerns on an Example. . It is a figure which illustrates the lane network etc. which concern on an Example, (a) is a figure which illustrates the data of the skeleton line which concerns on an Example, (b) is a figure which illustrates a corresponding lane network. It is a figure which shows the vertical network etc. which concern on an Example, (a) is a figure which illustrates the data of a corresponding skeleton line, (b) is a figure which illustrates the vertical network which concerns on an Example, (c) (E) to (e) are diagrams for explaining generation of a vertical network according to the embodiment. It is a figure which illustrates calculation of the line score which concerns on an Example, (a) is a figure which illustrates a corresponding vertical network, (b) thru | or (f) is a figure which illustrates calculation of the said line score. . It is a figure explaining the width score which concerns on an Example, (a) is a figure which shows a legal width, (b) is a figure which shows the specific example of the said width score. It is a figure which shows the line score etc. which concern on an Example, (a) is a figure which illustrates a corresponding vertical network etc., (b) is a figure which shows the specific example of the said line score. It is a figure which shows the vertical network score etc. which concern on an Example, (a) is a figure which illustrates a corresponding vertical network etc., (b) is a figure which shows the specific example of the said vertical network score. It is a figure which shows calculation of the lane line which concerns on an Example, (a) is a figure which illustrates a corresponding vertical network etc., (b) And (c) is a figure which illustrates calculation of the said lane line. (D) is a figure which illustrates the said lane line after calculation. It is a figure which illustrates the lane score which concerns on an Example, (a) is a figure which shows the 1st example of the said lane score, (b) is a figure which shows the 2nd example of the said lane score, It is a figure (I) explaining generation | occurrence | production of a lane network when there exists a defect | deletion etc. of the lane boundary line concerning an Example, (a) is a figure of the ortho image data which illustrates the defect | deletion of a lane boundary line, b) is a diagram illustrating the situation of a line when there is the deficiency, etc., (c) is a diagram illustrating the situation of the vertical network according to the embodiment when there is the deficiency, etc., (d) It is a figure which illustrates the width score which concerns on the Example when there exists the said defect | deletion etc. It is a figure (II) explaining the production | generation of a lane network when there exists a defect | deletion etc. of the lane boundary line which concerns on an Example, (a) is a figure which illustrates grouping when there exists a defect | deletion, etc. (A) is a diagram illustrating the situation of the lane line according to the embodiment when there is such a deficiency, and (c) is a diagram illustrating the situation of the temporary line according to the embodiment when there is such a deficiency. Yes, (d) is a diagram (II) illustrating the state of the lane line according to the embodiment when there is such a deficiency or the like. It is a figure which illustrates a lane score etc. when there is a loss of a lane boundary line concerning an example, (a) is the figure (I) which illustrates the example, (b) is the figure (II) which illustrates the example is there. It is a figure which illustrates a lane network in case there is a loss of a lane boundary line concerning an example, (a) is a figure which illustrates the 1st example of the lane network concerned, and (b) is the figure of the lane network concerned. It is a figure which illustrates two examples. It is a figure which illustrates the lane network in the intersection which concerns on an Example.

  Next, the form for implementing this application is demonstrated using FIG. FIG. 1 is a block diagram illustrating a schematic configuration of the lane information generation device according to the embodiment.

  As illustrated in FIG. 1, the lane information generation device S according to the embodiment includes an extraction unit 1, a comparison unit 2, and a generation unit 3.

  In this configuration, the extracting unit 1 extracts line segment information indicating a line segment shown in the actual image obtained by imaging the road from the actual image information corresponding to the actual image.

  And the comparison means 2 is based on the line segment information extracted by the extraction means 1 and the interval between the line segments arranged in the direction intersecting the road and the width as the lane included in the road. And compare.

  Accordingly, the generation unit 3 generates lane information indicating the lane based on the comparison result by the comparison unit 2.

  As described above, according to the operation of the lane information generation device S according to the embodiment, the line segment indicating the line segment shown in the actual image from the actual image information corresponding to the actual image obtained by imaging the road including the lane. Extract information. And the space | interval of the said crossing direction of the line segment located in the direction which cross | intersects a road and the width | variety as a lane are compared based on line segment information, and lane information is produced | generated based on the comparison result. Therefore, since the lane information is generated based on the comparison between the width of the direction intersecting the road between the line segments reflected in the real image and the width, the lane information indicating the lane included in the road is based on the real image information. It can be generated automatically and accurately.

  Next, specific examples corresponding to the above-described embodiment will be described with reference to FIGS. In addition, the Example demonstrated below is an Example at the time of applying this application to the automatic production | generation of the lane (lane) network information which shows the connection state etc. for every lane contained in a road.

  2 is a block diagram showing a schematic configuration of the lane network generation device according to the embodiment, FIG. 3 is a flowchart showing a lane network generation processing according to the embodiment, and FIG. 4 is ortho image data according to the embodiment. FIG. 5 is a diagram illustrating the relationship between the ortho image data and the road link according to the embodiment. 6 is a diagram for explaining the probability of up and down lines according to the embodiment, FIG. 7 is a diagram illustrating long lines and the like according to the embodiment, and FIG. 8 illustrates the situation of long lines and the like according to the embodiment. FIG. 9 is a diagram illustrating details of long lines and the like according to the embodiment, and FIG. 10 is a diagram illustrating data of a pedestrian crossing according to the embodiment. 11 is a diagram illustrating a lane network according to the embodiment, FIG. 12 is a diagram illustrating a vertical network according to the embodiment, and FIG. 13 illustrates calculation of a line score according to the embodiment. FIG. 14 is a diagram for explaining the width score according to the embodiment. FIG. 15 is a diagram showing line scores and the like according to the embodiment, FIG. 16 is a diagram showing vertical network scores and the like according to the embodiment, and FIG. 17 shows calculation of lane lines and the like according to the embodiment. FIG. 18 is a diagram illustrating a lane score according to the embodiment, and FIG. 19 is a diagram (I) illustrating generation of a lane network when there is a lane boundary line defect or the like according to the embodiment. . FIG. 20 is a diagram (II) illustrating generation of a lane network when there is a lane boundary line defect or the like according to the embodiment, and FIG. 21 is a diagram when there is a lane boundary line defect or the like according to the embodiment. 22 is a diagram illustrating a lane score and the like. FIG. 22 is a diagram illustrating a lane network when a lane boundary line is missing according to the embodiment. FIG. 23 is a diagram illustrating a lane network in an intersection according to the embodiment. It is a figure to do. At this time, in FIG. 2, the same member numbers as the respective constituent members in the lane information generating apparatus S for the respective constituent members in the examples corresponding to the respective constituent members in the lane information generating apparatus S according to the embodiment shown in FIG. 1. Is used.

  As shown in FIG. 2, the lane network generation device SV according to the embodiment includes a processing unit 10 including a CPU, a RAM (Random Access Memory), a ROM (Read Only Memory), and the like, and an external network such as the Internet. An interface 11 for controlling the exchange of data between them, an operation unit 12 including a keyboard and a mouse, a display 13 including a liquid crystal display, and a recording including an HDD (Hard Disc Drive) or an SSD (Solid State Drive). And a portion 14. The processing unit 10 includes an extraction unit 1 as an example of the extraction unit 1 according to the embodiment, a detection unit 2 as an example of the comparison unit 2 according to the embodiment, and a generation as an example of the generation unit 3 according to the embodiment. Part 3. At this time, each of the extraction unit 1, the detection unit 2, and the generation unit 3 may be realized by a hardware logic circuit such as a CPU constituting the processing unit 10, or a lane network generation process according to an embodiment described later. The CPU or the like may read out and execute a program corresponding to the above, and may be realized as software. Moreover, as shown by a broken line in FIG. 2, the extraction unit 1, the detection unit 2, and the generation unit 3 constitute an example of the lane information generation device S according to the embodiment. Further, the extraction unit 1 corresponds to an example of the “road marking extraction unit” according to the present application, the detection unit 2 corresponds to an example of the “length detection unit” according to the present application, and the generation unit 3 It corresponds to an example of “means”, an example of “virtual line segment information generating means”, and an example of “connection information generating means”.

  In the above configuration, the aerial photograph data to be subjected to the lane network generation process according to the embodiment is recorded in advance in the recording unit 14 as ortho image data generated from the aerial photograph data taken by the aircraft, for example. Or acquired from the outside via the interface 11. On the other hand, when an instruction operation is performed on the lane network generation device SV in the operation unit 12, the operation unit 12 generates an operation signal corresponding to the instruction operation and outputs the operation signal to the processing unit 10. Thus, the processing unit 10 performs lane network generation processing according to the embodiment using the ortho image data while displaying necessary information on the display 13 by the extraction unit 1, the detection unit 2, and the generation unit 3. Here, the lane network according to the embodiment is a network that shows, for each lane, the state of a lane included in a road indicated by a conventional road link. The “lane condition” in this case is, for example, how many (how many) lanes are included in one road, the change in the number of lanes on one road, or the lanes on other roads. And how they are connected. The lane network according to the embodiment is one of the targets for use in automatic driving, which is currently being researched.

  Next, the lane network generation processing according to the embodiment will be specifically described with reference to FIGS.

  The lane network generation process according to the embodiment is started by a start operation in the operation unit 12, for example. As shown in the flowchart corresponding to FIG. 3, when the lane network generation process according to the embodiment is started, the ortho image data that is the target of the lane network generation process is, for example, from the recording unit 14 or the interface 11. (Step S1). At this time, the ortho image data is, for example, image data of an aerial photograph as illustrated in FIG. 4A, and a structure BR such as a pedestrian bridge across the road is displayed together with an image of the road R intersecting at the intersection CR. ing. In addition to the passenger car C and the bus B existing on the road R, a separation zone P and the like are also shown. When such ortho image data is input, the extraction unit 1 of the processing unit 10 next extracts a road surface area corresponding to the road R and the intersection CR from the ortho image data by, for example, segmentation processing or mask processing ( Step S2). As a result of this step S2, for example, only the road surface area as shown in FIG. 4B is extracted from the ortho image data. Next, the extraction part 1 extracts the road marking drawn on the surface of the road R and the intersection CR from the ortho image data which became only the road surface area (step S3). The road sign (so-called road surface paint) extracted in step S3 is a sign that is usually drawn using white paint at the center of each lane, such as an arrow sign indicating the direction of travel, a destination sign, and a crossing. A sidewalk sign or a character sign such as “bus priority” is included. The road marking data in this case includes, for example, orientation data, scale data, similarity value data, and the like in addition to the position data of the road marking indicated by white dots in FIG.

  Next, the extraction unit 1 extracts a skeleton line composed of line segments constituting the road R and the like from the ortho image data of only the road surface area by a process such as so-called polyline conversion (step S4). The process of step S4 is a process of thinning or subdividing the ortho image data of only the road surface area. As a result of this step S4, for example, skeleton line data as shown in FIG. 4C corresponding to the ortho image data of only the road surface area shown in FIG. 4B is generated. This skeleton line data is so-called vector data.

  Next, as illustrated in FIG. 5, the extraction unit 1 uses the road link RL data (see FIG. 5A) corresponding to the road R shown in the ortho image data and the skeleton line extracted in step S <b> 4. Each included line (line segment; see FIG. 5 (b)) is compared, and a line around the road link RL and a line parallel to the link direction and a line perpendicular to the link direction are extracted from the skeleton line. It is linked to the data of the road link RL that has become (step S5). At this time, in the case illustrated in FIG. 5B, for example, the road marking indicating the pedestrian crossing before the intersection is composed of a line segment parallel to the road link RL, and a line perpendicular to the road link RL is, for example, a temporary stop line. The corresponding line is mentioned. Next, the extraction unit 1 associates each line with the road link RL of the up line and the down line (step S6). At this time, the extraction unit 1 considers the difficulty of allocating to any road link due to the relative distance, and for all lines, the vertical line probability in relation to the road link RL (the line corresponding to the up line) Or the probability corresponding to the downlink) (step S6).

  Here, the above vertical line probability is, for example, in the case of a single road shown in FIG. 6A (that is, a road corresponding to a single road link RL consisting of an up line and a down line in principle) from the road line RL. The farther away the line, the higher the probability of being an up line or down line, but the closer to the center of the road R corresponding to the road link RL, the more likely that the line is in the up line and the down line. The probability of being present is equal (see FIG. 6A). On the other hand, for example, in the case of a two-row road shown in FIG. 6B (that is, a road corresponding to two road links RL each corresponding to a plurality of lanes and having opposite link directions in principle), The probability that the outside of the road line RL is an up line or a down line is one hundred percent, but in the area between the two road lines RL, the probability of being an up line or down line differs depending on the distance. Comes (see FIG. 6B).

  Next, the extraction unit 1 is an embodiment of the road R (see FIG. 4C, FIG. 5B, or FIG. 7A) that has been converted into a skeleton line based on the setting operation or the like in the operation unit 12. One or a plurality of sections to be subjected to the lane network generation process are set (step S7). Next, the extraction unit 1 determines whether any one of the set (see step S7) sections has been selected based on the operation in the operation unit 12 (step S8). In the determination in step S8, for example, if no section is selected for the reason that the lane network generation processing for all sections other than in the intersection CR has been completed (step S8: NO), the processing unit 10 will be described later. The process proceeds to step S15 to perform a lane network generation process in the intersection CR. On the other hand, if any one of the sections is selected in the determination in step S8 (step S8: YES), the extraction unit 1 then selects each actual line for each line included in the section selected in step S8. Classification based on the length of. Here, the actual (real) length of the lines included in the skeleton line can be detected, for example, by converting the ortho image data into a vector data after thinning (see step S4). As another method, the original ortho image data (aerial photograph) scale (scale) and pixels in the image may be converted to meters, or the distance in the pixel space without the conversion. It is also possible to calculate the value. However, in this case as well, it is considered that it is necessary to improve accuracy by processing with sub-pixels instead of simply pixel distances. More specifically, the extraction unit 1 converts each line extracted as the skeleton line shown in FIG. 7A into an outer line OL (FIG. 7) indicating an outer line of each road R (a boundary line with a region other than the road). (See (b)), a long line LL (see FIG. 7C) parallel to the road and having a length of 8.5 meters or more, and a short line parallel to the road and having a length of less than 8.5 meters. It is divided into SL (see FIG. 7D) and a vertical line VL perpendicular to the road (see FIG. 7E). At this time, the threshold for separating the long line LL and the short line SL is “8.5 meters” because the lane boundary line (broken line) of the expressway in the road structure ordinance is 8 meters. , A threshold value slightly longer than that is provided. When these outer lines OL, long lines LL, short lines SL, and vertical lines VL are viewed on the skeleton line data, for example, at the intersection CR, the long lines LL are distributed as illustrated in FIG. become. When the intersection CR portion is enlarged and compared with an actual aerial photograph (see FIG. 9B) as ortho image data, as illustrated in FIG. 9A, a line 50 corresponding to the structure BR, A line 51 corresponding to a lane boundary line, a line 52 corresponding to noise on image data, a line 53 corresponding to the separation band P, a line 54 corresponding to a temporary stop line, a line 55 corresponding to a bicycle crossing band, a bus, etc. It can be seen that a line 56 corresponding to noise due to the above, a line 57 corresponding to a road marking, a line 58 corresponding to a lane boundary line, a line 59 corresponding to a pedestrian crossing, and the like are included.

  Next, the extraction unit 1 arranges the skeleton lines at equal intervals in parallel and at a density equal to or higher than a preset density from each line (see FIG. 10A) included as the short lines SL divided in step S9. The short line SL is detected as a line 59 indicating a pedestrian crossing (step S10). This is because the line 59 indicating the pedestrian crossing is not included in the lane network according to the embodiment.

  On the premise of the processing up to step S10 described above, the processing described below is sequentially executed in the detection unit 2 and the like of the processing unit 10, whereby the lane network generation processing according to the embodiment is performed. As a result of the lane network generation processing, for example, a lane network LNW as illustrated in FIG. 11B is generated from the data of each line included in the skeleton line illustrated in FIG. 11A (step S11). To Step S15). In the following description, the section that is the target of the lane network generation process set in step S7 is a dashed rectangle shown in FIG. 12A in the area surrounded by the dashed ellipse shown in FIG. It is assumed that the section is shown.

  When step S1 to step S10 are completed, the detection unit 2 of the processing unit 10 next targets the vector data VD (skeleton line data) corresponding to the section indicated by the broken-line rectangle in FIG. Then, a lane network using the vertical network according to the embodiment is generated (step S11). At this time, the detection unit 2 and the generation unit 3 exclude the vertical line VL and the short line SL constituting the pedestrian crossing (see FIG. 10B) from the processing target.

  Here, the vertical network N according to the embodiment is a line segment connecting two adjacent lines (lines as data of the skeleton line) L as illustrated in FIG. Are line segments having nodes ND on each line L.

  As step S11, the detection unit 2 first selects one target line LX from a plurality of lines LA in the target section, for example, as shown in FIG. When the target line LX is moved in a direction perpendicular to the link direction of the link RL (see the hatched portion in FIG. 12C), the target lines LX with other lines LA, LC, LD, and LE that are in contact with each other A vertical network N is set in between. At this time, as illustrated in FIG. 12D or 12E, the detection unit 2 intersects two adjacent lines LA and LB when they are projected in parallel in the direction of the road link RL ( The vertical network N is set with the intersection of the center position of each line LA and line LB in FIG. 12D and FIG. 12E as the position of the node ND. The detection unit 2 sets the vertical network N on the assumption that the line LF (see FIG. 12C) that is shielded by the other line LC when viewed from the target line LX is noise as the vector data VD. Not applicable. Then, when the vertical network N is set for each of the lines LA to LL included in the target section indicated by the dashed rectangle in FIG. 12A, the vertical networks N1 to N13 illustrated in FIG. 13A are set. The

  Next, the detection unit 2 determines whether each line L corresponds to a lane boundary line (that is, the likelihood of a lane boundary line) from the actual lengths of the adjacent lines L themselves. "Score" is calculated for each line L. More specifically, for each line L, when the lengths of the two lines L connected to the vertical network N via the node ND are close to each other in a range of ± 0.5 meters, One point is added to the line score of each line L. About the line score of each line L by this length, it becomes a high score, so that the line L of the same length adjoins (namely, the likelihood of the lane boundary line in the said two lines L is high). For example, when attention is paid to the line LA and the line LB shown in FIG. 13B, if the actual lengths are 1.5 m and 1.0 m in order, the line scores for the line LA and the line LB are respectively set. 1 point is added. Similarly, when attention is paid to the line LB and the line LC shown in FIG. 13C, if the actual length is 1.0 meter and 0.8 meter in order, the line LB and the line LC are also described. Each line score is incremented by one point. As a result, one line score is added for the line LA, two line scores are added for the line LB, and one line score is added for the line LC. The value of the line score based on the length of each line L is hereinafter referred to as “condition 1” as appropriate.

  Next, the detection unit 2 calculates the likelihood of the lane boundary line of each line L also from the interval (width) of the adjacent lines L. More specifically, the detection unit 2 determines the width class W shown in FIG. 14 (b) defined based on the width of the lane defined in Article 5 (4) of the Road Structure Ordinance shown in FIG. 14 (a). The lane boundary line likelihood of each line L is calculated. In the case illustrated in FIG. 14B, the definition of each width class W is defined with a margin of ± 10%. Further, as the data described in FIG. 14, for example, data recorded in advance in the recording unit 14 may be used. The detection unit 2 sets the number of vertical networks N other than those corresponding to the width class W0 and the width class W9 to “n” for each of the lines L connected to the plurality of vertical networks N, and sets the same width class. When the maximum number of the vertical networks N to which it belongs is m (m> 1), the value of “m / n” is added to the line score. The line score of each line L based on the width score W becomes higher as the lines L belonging to the same width class are adjacent to each other (that is, the higher the lane boundary line in the two lines L is). For example, when attention is paid to the line LX shown in FIG. 13D, the width score W for the line LA is different from the width score W for the line LB, and there are two effective vertical networks N. 0/2 = 0.0 is added to the line score. For example, when attention is paid to the line LX shown in FIG. 13 (e), the width score W for the line LA and the width score W for the line LB are the same, and there are two effective vertical networks N. 2/2 = 1.0 is added to the line score by W. Further, for example, when attention is paid to the line LX shown in FIG. 13 (f), three types of width scores W correspond, but the number of the vertical networks N of the width scores W2 is three, and the number of effective vertical networks N is the largest. Since it is four (excluding the vertical network N of the width score W0), 3/4 = 0.75 is added to the line score by the width class W. In addition, the value of the line score by the width class W of each line L is hereinafter referred to as “condition 2” as appropriate.

  And the detection part 2 adds the line score by the said conditions 1 and the said condition 2 for every line L, and totals a line score. At this time, in the case of each of the lines LA to LL illustrated in FIG. 13A and FIG. 15A, the total results are as shown in FIG.

  Next, the detection unit 2 calculates, for each vertical network N, a vertical network score, which is a parameter indicating the width likeness, regarding the width between the lines L. More specifically, the vertical network score is calculated for each vertical network N by adding the width score shown in FIG. 14B and the totaled line score (see FIG. 15B). At this time, the vertical network N (in the example illustrated in FIG. 16, the vertical network N1, the vertical network N6, and the vertical network N7) in which the vertical network score is negative is not used for the subsequent lane network generation process. As a result, only the vertical network N that truly corresponds to the lane can be the target of the lane network generation process. More specifically, for example, in the line LA to line LL and the vertical network N1 to vertical network N13 illustrated in FIG. 16A, focusing on the vertical network N3 connecting the line LD and the line LE, the width class W1 and the line Since the line score of the LD is “2” and the line score of the line LE is “3”, the vertical network score of the vertical network N3 is “7”. Similarly, the vertical network score calculated for each vertical network N is illustrated in FIG. 16B.

  Next, the generation unit 3 of the processing unit 10 groups the vertical networks N connected to each other in the vertical network N that has been studied so far as one group. That is, the groups G1 to G3 illustrated in FIG. 17A are formed. In each of these groups, the vertical network N2, the vertical network N5, the vertical network N10, the vertical network N11, and the vertical network N13 at both ends are connected to the line LA, the line LB, the line LC, and the line LL corresponding to the outer line OL on the road. And the vertical network N8 is connected to the line LB via the vertical network N7 having a negative vertical network score. Each vertical network N included in each of the groups G1 to G3 crosses the road as illustrated in FIG. 17A, and the number of the vertical networks N is the number of lanes themselves.

  Next, the generation unit 3 calculates lane lines corresponding to the lanes connecting the groups G1 to G3 using the vertical network N included in the groups G1 to G3. More specifically, as illustrated in FIGS. 17B and 17C, the generation unit 3 determines one target vertical network NX, and all the verticals included in the group G adjacent from the center. The direction to the network N is calculated, and the direction closest to the link direction of the parallel road links RL is defined as a lane line LL (that is, one lane) starting from the vertical network NX. In the case illustrated in FIG. 17B and FIG. 17C, the direction indicated by the broken line is not adopted as the lane line.

  The lane line LLA to the lane line LLG according to the embodiment are calculated as illustrated in FIG. 17D by the series of processes described with reference to FIGS. 12 to 17, and the lanes according to the embodiment are thereby calculated. A network is generated (step S11).

Next, the generation unit 3 calculates, for each line L and each vertical network N, a lane score corresponding to a comprehensive evaluation of the lane boundary line likelihood and the width likelihood. More specifically, in the case illustrated in FIG. 18A, for each lane line LLA to lane line LLG, the total value of the vertical network scores of the connected vertical networks N is “S” and is included in each lane. The number of lane lines LL is “LL”, the number of vertical networks N included in each lane is “LN”, the maximum number of the same width class included in each lane is “W”, and the lane score is ( S / LL) × (W / LN). For example, the lane score of the lane composed of the lane line LLA to the lane line LLC illustrated in FIG.
{(4 + 7 + 6 + 6) / 3} × (4/4) = 7.6
The lane score of the lane composed of the lane line LLD and the lane line LLE illustrated in FIG.
{(7 + 7 + 7) / 2} × (3/3) = 10.5
The lane score of the lane composed of the lane line LLF and the lane line LLG illustrated in FIG.
{(5 + 5 + 5) / 2} × (3/3) = 7.5
It becomes. As shown in FIG. 18B, the generation unit 3 assumes that the lane passing near the center of the road marking will be correct based on the position RP of the road marking extracted in step S3. Alternatively, a road marking score may be calculated and added to the lane score. That is, a circle with a radius of about 50 centimeters centered on the detection position of the road marking is assumed as the road marking circle, and when the road marking circle and the lane line LL intersect or contact each other, the lane line LL and the road As the shortest distance from the center of the marking circle is shorter, it is preferable to increase the road marking score and add it to the lane score.

  Next, the generation unit 3 has the lane boundary LL generated up to step S11, and the lane boundary line rubs or disappears at the stage of the original ortho image data. It is determined whether or not there is a width that is narrower than the width of the road in which the lane is included and is wider than the width class W7 (for example, for two lanes) (step S12).

  If it is determined in step S12 that there is no wide width (step S12: NO), the generation unit 3 proceeds to step S13 described later. On the other hand, if it is determined in step S12 that there is a wide width as described above (step S12: YES), the generation unit 3 next converts the virtual line (line segment) corresponding to the width to the vector data VD of the skeleton line. And using the vector data VD after the addition, a lane network generation process (second pattern lane network generation process) similar to step S11 is performed (step S14).

  Here, when the lane boundary line rubbing or the like as described above occurs, the corresponding line does not appear in the corresponding ortho image data as exemplified by the broken line ellipse in FIG. . Then, when the extraction unit 1 extracts the line L from such ortho image data, as a result, the lines LA to LI as illustrated in FIG. 19B are extracted. When the setting of the vertical network N similar to that in step S11 is performed by the generating unit 3, the vertical networks N1 to N14 illustrated in FIG. 19C are set as a result. At this time, the width class and the width score for the vertical networks N1 to N14 are as shown in FIG. 19D (see also FIG. 14). Next, when the grouping similar to step S11 is performed on each line L and each vertical network N illustrated in FIG. 19C by the generation unit 3, as a result, the groups G1 to G1 illustrated in FIG. G4 is formed. Then, when the same lane line LL is calculated with respect to the grouping result, the lane lines LLA to LLG illustrated in FIG. 20B are calculated as the result.

  In step S14, the vertical network N2, the vertical network N3, and the vertical network N5 that are wide in the state shown in FIG. 20A are each bisected in the length direction, and the original vertical network N2 is divided into the vertical network. N2-1 is divided into a vertical network N2-2, the original vertical network N3 is divided into a vertical network N3-1 and a vertical network N3-2, and the original vertical network N5 is further divided into a vertical network N5-1 and a vertical network N5-2. (Refer to the “recombination” column in FIG. 14B). Then, as illustrated in FIG. 20C, provisional lines LB ′, provisional lines LC ′, and provisional lines LD ′ passing through the bisectors of the vertical network N2, the vertical network N3, and the vertical network N5 are set. The lane line LL is calculated again using each line L and each vertical network N. In this case, the lane line LLA to the lane line LLI illustrated in FIG. And if the same lane score as step S11 is calculated based on this, the result will be illustrated in FIG. 21 (step S14).

  Next, the generation unit 3 performs final evaluation on the lane network generated until the determination in step S12 or until step S14, and outputs the lane network according to the embodiment to the outside through the interface 11. Or is recorded in the recording unit 14 (step S13). Specifically, in step S13, the generation unit 3 determines that the lane network generated by the processing up to step S11 is set as the lane network of the section when there is no wide range in the determination in step S12 (step S12: NO). After that, the process returns to step S8 to generate a lane network for the next section. On the other hand, when passing through step S14, the generation unit 3 compares the lane scores generated in step S14, and as shown in FIG. 22 (a), the value is higher (in the case illustrated in FIG. 22 (a)). Outputs a lane network corresponding to FIG. 20D, and then returns to step S8 to generate a lane network for the next section. In addition, you may use the visual determination by a user together with determination of said step S13.

  In the embodiment described above, if there is a wide width, a temporary line LA ′ or the like is generated and interpolated, and the lane scores for the lane lines LLA and the like calculated according to the presence or absence of the interpolation are compared. And output the final lane network. However, other than this, for example, as illustrated in FIG. 22B, the link direction of the parallel road link RL with respect to the lane line LLA and the like calculated without generating the temporary line LA ′ and the like in a state where there is a wide width. Are calculated by comparing the sum of these values according to the presence / absence of interpolation using the temporary line LA ′ or the like, and the sum of the angles is smaller (that is, the angle with the link direction of the road link RL). You may comprise so that a difference may be output as a lane network.

  Next, when no section is selected in the determination of step S8 (step S8: NO), the processing unit 10 generates a lane network in an intersection CR where a plurality of roads R intersect (step S15). Step S15 will be described in detail later. Thereafter, the processing unit 10 determines whether or not to end the lane network generation process according to the embodiment, for example, for the reason that the power of the lane network generation apparatus SV is turned off (step S16). When the lane network generation process ends in the determination of step S16 (step S16: YES), the processing unit 10 ends the lane network generation process as it is. On the other hand, when continuing the said lane network generation process in determination of step S16 (step S16: NO), the process part 10 returns to said step S1, and repeats a series of processes mentioned above.

  Next, the lane network generation process at the intersection in step S15 will be specifically described. Specifically, as the lane network generation process, the following four methods can be considered.

(I) The method of referencing the connection relationship in the road link RL, that is, the processing unit 10 assumes, as the first method, that there are groups of lane networks connected to the intersection CR illustrated in FIG. 23 from groups G1 to G4. . At this time, when there is a road link RL for a road connected to the intersection CR, a rough connection relationship between the groups G (connection between the groups G) is determined. More specifically, for example, the groups G1 to G4 illustrated in FIG. 23 determine that “the intersection CR forms a crossroad”. In addition, for example, the lane line LL included in the group G1 and the lane line LL included in the group G2 are connected at the intersection CR, and included in the lane line LL and group G4 included in the group G3 as a three-dimensional intersection or underground passage therewith It is determined that the connected lane line LL is connected.

(II) The method of using the up / down attribute in the lane line, that is, the processing unit 10, as a second method in the case illustrated in FIG. 23, when there is a road link RL for the road connected to the intersection CR, The ascending attribute and the descending attribute of the Nijo road (see FIG. 6B) in the road link RL are transferred to the lane line LL, and it is determined that the ascending lanes and the descending lanes can be connected. As a second method, when the uplink attribute and the downlink attribute of the lane line LL are set by a method other than using the road link RL, the uplink lanes and the downlink lanes can be connected based on the attributes. You may decide that there is.

(III) The method using the road marking, that is, the processing unit 10 refers to the road link RL when there is a road link RL for the road connected to the intersection CR as a third method in the case illustrated in FIG. Can be connected within the intersection CR based on the connection state between the groups G determined in this way, the up or down attribute in each lane line LL, and the direction indicated by the direction sign (arrow mark) as a road sign Lane line LL is estimated. More specifically, for example, when the left lane of the group G2 illustrated in FIG. 23 has an arrow sign indicating straight ahead or left turn, the leftmost lane line LL of the group G2 is connected to the group G1 and the group G3. Then decide. In this case, as a result, the connection from one group G to a plurality of groups G can be determined.

(IV) Utilization of probe information, that is, the processing unit 10 determines the trajectory of a vehicle or the like passing through the intersection CR by collecting so-called probe information as a fourth method in the case illustrated in FIG. Is detected (or a trajectory representing each trajectory is obtained) and used to determine the connection relationship of each group G. In this case, it is possible to determine a more realistic lane (curved running line) and its shape in the intersection CR.

  As described above, according to the lane network generation process according to the embodiment, the line L reflected in the actual image is extracted from the ortho image data corresponding to the actual image obtained by capturing the road R including the lane, and the road A distance between the lines L arranged in a direction intersecting with R in the crossing direction is compared with a legal width as a lane, and a lane network is generated based on the comparison result. Therefore, since the lane network is generated based on the comparison between the distance between the lines L reflected in the actual image in the direction intersecting the road and the legal width, the lane network indicating the lane included in the road R is used as the actual image information. Based on this, it can be generated automatically and accurately.

  In addition, when there is a line L whose interval is longer than the legal width, the interval is divided based on the legal width, and a temporary line L ′ passing through the division point and extending in the road R is generated. Based on the line L ′, the lane network is generated by comparing the case where the temporary line L ′ is generated and the case where the temporary line L ′ is not generated. Therefore, since the lane network is generated based on the comparison results between the case where the division based on the legal width is performed and the case where the division is not performed, for example, even when the lane boundary line on the road is rubbed, the lane information is more accurately detected. Can be generated.

  Furthermore, since the length of each line L is detected, a vertical network N connecting the lines L is generated, and a lane network is generated based on the detected length, the vertical network N, and each comparison result. Considering the length of each line L and the vertical network N, a lane network can be generated more accurately.

  Furthermore, in the case shown in FIG. 22B, a lane network is generated based on a comparison between the link direction of each lane line LL and the road link RL, so that each of the lane directions indicated by each lane line LL Considering the direction in which the road R including the lane extends, the lane network can be generated more accurately.

  Further, as illustrated in FIG. 18B, when generating a lane network based on the ortho image data indicating the road marking drawn on the center line of the lane, the lane network indicating the lane included in the road is displayed. It can be generated more accurately.

  Furthermore, when generating a lane network based on the selection result by a user, the lane network according to a user's intention can be generated.

In addition to the above-described embodiments, examples of technologies that can be used for the lane network generation processing according to the embodiments include the following technologies.
A lane network in an intersection CR or in a shielding part (such as under an overpass or a tunnel) can be generated by using a virtual trajectory generated based on curve interpolation or a running image.
-By using the so-called real-time evaluation value variation method, the lane network that matches the actual lane condition is left out of the multiple lane networks, and the lane network on the road adjacent to the road is compared with the remaining lane network. The evaluation value is changed in consideration of the connection relation. Thereby, the accuracy as a lane network candidate can be improved.
・ Introducing so-called machine learning methods. That is, a plurality of lane network candidates are input, and a correct lane network visually determined by the user is determined and output, or neural network learning that determines and outputs the correct edit is utilized. In this case, if the learning is sufficiently performed, the output result of the neural network may be the highest result instead of the priority order of candidates based on the evaluation value.

  Furthermore, a program corresponding to the flowchart shown in FIG. 3 is recorded on a recording medium such as an optical disk or a hard disk, or is acquired via a network such as the Internet, and is read out to a general-purpose microcomputer or the like. It is possible to cause the microcomputer or the like to function as the processing unit 10 according to the embodiment.

1 Extraction means (extraction unit)
2 comparison means (detector)
3 Generation means (generation unit)
10 Processing Unit S Lane Information Generation Device SV Lane Network Generation Device

Claims (8)

Extraction means for extracting line segment information indicating a line segment reflected in a real image obtained by imaging a road from real image information corresponding to the real image;
Comparing means for comparing, based on the extracted line segment information, an interval in the intersecting direction between the line segments arranged in a direction intersecting the road with a width as a lane included in the road; ,
Generating means for generating lane information indicating the lane based on a comparison result by the comparing means;
A lane information generation device comprising: In the lane information generation device according to claim 1,
The generating means includes
Based on the comparison result in the comparison means, the dividing means for dividing the interval of the line segment whose interval is longer than the width based on the width;
Virtual line segment information generating means for generating virtual line segment information indicating a virtual line segment in a direction in which the road extends through the dividing point in the divided interval;
With
When there is the line segment whose interval is longer than the width, the generation means is the first comparison result when the division based on the width is not performed based on the generated virtual line segment information. Generating the lane information based on a comparison result and a second comparison result that is a comparison result between the line segment corresponding to the divided virtual line segment, an interval between the virtual line segment, and the width. A lane information generation device characterized by the above. In the lane information generation device according to claim 2,
A length detection means for detecting the length of the line segment constituting the interval based on the extracted line segment information;
Connection information generating means for generating connection information corresponding to a connection line connecting the line segments constituting the interval based on the extracted line segment information;
Further comprising
The generation means generates the lane information based on length information indicating the detected length, the generated connection information, the first comparison result, and the second comparison result. A lane information generation device characterized by the above. In the lane information generation device according to claim 2,
The generation means uses i) the direction of the lane indicated by the first candidate lane information generated using the first comparison result as the lane information candidate, and ii) uses the second comparison result as the candidate. A lane information generation device, wherein the lane information is generated based on a comparison between each of the lane directions indicated by the second candidate lane information generated in this way and a direction in which the road extends. In the lane information generation device according to any one of claims 1 to 4,
Road marking extracting means for extracting road marking information indicating the road marking drawn on the center line of the lane that is a road marking other than the line segment reflected in the actual image from the actual image information;
The generation means generates the lane information based on the extracted road marking information. In the lane information generation device according to claim 2,
The generating means includes
Output that outputs first candidate lane information generated using the first comparison result as the lane information candidate and second candidate lane information generated using the second comparison result as the candidate. Means,
Receiving means for accepting selection of either one of the output first candidate lane information or second candidate lane information;
Including
The generation means generates the lane information based on any one of the selected ones. A lane information generation method executed in a lane information generation device that generates lane information indicating a lane included in a road,
An extraction step of extracting line segment information indicating a line segment reflected in the actual image obtained by imaging the road from the actual image information corresponding to the actual image;
Based on the extracted line segment information, a comparison step for comparing the interval between the line segments arranged in the direction intersecting the road and the width as the lane, and
A generation step of generating the lane information based on a comparison result in the comparison step;
A lane information generation method comprising: A computer included in a lane information generation device that generates lane information indicating a lane included in a road,
Extraction means for extracting line segment information indicating a line segment reflected in the actual image obtained by imaging the road from actual image information corresponding to the actual image;
Based on the extracted line segment information, a comparison means for comparing an interval between the line segments arranged in a direction intersecting the road and the width as the lane, and
Generating means for generating the lane information based on a comparison result by the computer functioning as the comparing means;
A program for generating lane information, characterized in that it functions as: