JP4899657B2 - Road shape acquisition device - Google Patents

Description

  The present invention relates to a road shape acquisition device that acquires the shape of a road on which a vehicle travels, and more particularly to a road shape acquisition device that acquires a road shape based on a detection result of a white line.

  In the navigation system, the position of the host vehicle is detected by positioning by radio wave positioning means, and the road map corresponding to the position is displayed on the display device, so that the position where the vehicle is traveling is visually provided to the driver. can do.

  However, the accuracy of the road map data for displaying the road map is not sufficient. For example, a road is represented by a link connecting intersections (nodes), but the link is configured by a straight line, and may not match the actual road shape. For this reason, even if the position of the own vehicle is detected with a certain degree of accuracy, the position of the own vehicle may be displayed differently with respect to the road map data.

  The road map data is stored in a storage device such as a DVD or a hard disk drive and mounted on the vehicle. Even if the road network changes due to construction work, there is a time lag until the road map is updated. When updating is forgotten, there is a problem that accurate road map data cannot be used forever.

  On the other hand, in order to obtain high-accuracy road map data, it is difficult to realize this on all roads because accurate surveying of roads and high definition of links are necessary. Even if high-accuracy road map data is obtained, it is still impossible to cope with changes in the road network due to construction.

  By the way, for a self-supporting mobile robot, a method of simultaneously performing self-position estimation and map creation in an unknown environment called SLAM (Simultaneous Localization and Mapping) has been proposed. In SLAM, a laser range finder is used to calculate the presence and distance of an obstacle to create a map having information on both an accessible area and an inaccessible area that are useful for generating robot behavior. Is expensive. Also, since there are no obstacles on the road except for the roadside belt and guardrail, it is difficult to obtain road map data with a laser.

On the other hand, a method for acquiring road map data by a probe car equipped with a camera has been proposed (for example, see Patent Document 1). In Patent Document 1, a white line is recognized from an image taken in front of the vehicle, and the road width, the number of lanes, the position of an intersection, and the like are detected based on the white line. Then, the detected road width or the like is compared with the road map data stored in advance, and if there is a mismatch, the road map data is rewritten with information such as the detected road width.
Japanese Patent Laid-Open No. 2000-230834

  However, since the method described in Patent Document 1 detects whether or not there is a change in the road map data, and updates it when there is a change, the change in the road width and the number of lanes can be updated. Yes, but it does not improve the accuracy of road map data. Therefore, the actual road shape cannot be associated with the position of the measured vehicle with high accuracy.

  An object of this invention is to provide the road shape acquisition apparatus which can acquire a highly accurate road shape at low cost in view of the said subject.

In order to solve the above-described problems, the present invention is a road shape acquisition apparatus that divides a link plane generated based on links constituting a road into grids and acquires a road shape from the grid having a predetermined white line existence probability or more. Te, a map data storage means for storing link information of the road, the white line existence probability memory means for storing white line existence probability of last updated white presence probability or the initial value for each of the grid, detecting the position of the vehicle Position detecting means (for example, GPS receiver 11, vehicle speed sensor 12, gyro sensor 13), white line detecting means (for example, camera 14, image processing device 15) for detecting the white line of the road, and the detected white line the white line existence probability from the grid in a predetermined range grid, and, white line existence probability stored in the white line existence probability memory means corresponding to (e.g., 4 P | a (Z m) is applied to Bayesian update equation and having a a white line existence probability setting means for updating the white line existence probability stored in the white line existence probability memory means.

  According to the present invention, the system can be configured at a low cost by using a camera, and the white line existence probability is introduced in consideration of the detection error and updated by the Bayesian update formula. A road shape acquisition device capable of acquiring a road shape can be provided.

A road shape acquisition device that can acquire a high-precision road shape at low cost can be provided.
An obtaining device can be provided.

The best mode for carrying out the present invention will be described below with reference to the drawings.
FIG. 1 shows a configuration diagram of a road shape acquisition device 10. The road shape acquisition apparatus 10 of this embodiment acquires a detailed road shape using the detected white line for map data mounted on a vehicle or distributed from a server.

  The road shape acquisition device 10 has a map database (hereinafter simply referred to as a map DB) 17 that stores position information of latitude and longitude of each point where the vehicle travels. The map DB 17 is shared with a navigation device mounted on the vehicle.

  The road shape acquisition device 10 is controlled by map data ECU (Electronic Control Unit) 16 mounted on the vehicle. The map data ECU 16 includes a CPU that executes the program, a ROM that stores the program, a storage device that stores the program and files (for example, Hard Disk Drive), a RAM that temporarily stores data and programs, and an NV-RAM (Non Volatile). RAM) and an input / output device for inputting and outputting data are configured as a computer connected via a bus. The white line existence probability setting means is realized by the CPU executing the program.

  The map data ECU 16 includes a GPS receiver 11 that receives radio waves from a GPS (Global Positioning System) satellite, a vehicle speed sensor 12 that detects a vehicle speed, a signal (INS ( A gyro sensor 13 for outputting (Inertial Navigation Sensor) data) is connected.

  The map data ECU 16 measures the current position (specifically, latitude, longitude, altitude) of the host vehicle based on the radio wave received by the GPS receiver 11, and the travel distance based on the vehicle speed of the vehicle speed sensor 12 and the gyro sensor 13. Is accumulated in the traveling direction based on the INS data output to detect the current current position of the host vehicle.

  The map data ECU 16 extracts a road map around the detected current position and a road map of the area designated by the vehicle occupant from the map DB 17 on a display device provided in the vehicle interior, and displays the map in accordance with the designated scale. To display. The map data ECU 16 displays the current position of the vehicle superimposed on the road map as necessary.

  The map data ECU 16 is connected to an image processing device 15 to which image data taken by the camera 14 is input. The camera 14 is mounted on a front bumper or an indoor mirror, and images the front of the vehicle.

  The camera 14 has a CMOS (Complementary Metal Oxide Semiconductor) or a CCD (Charge Coupled Device) photoelectric conversion element, and outputs image data of a predetermined luminance gradation (for example, 256 gradations). The camera 14 takes an image of a region extending in a predetermined angular range horizontally downward from the mounting position toward the front of the vehicle.

  The image processing device 15 performs image processing on the image data around the front of the vehicle supplied from the camera 14, and detects white lines that are displayed on the image and that separate the traveling lanes drawn on the road. For example, the white line is detected based on the luminance of the image data by searching an area having a luminance equal to or higher than a predetermined threshold value upward from the bottom of the image data. Since the white line has edges that are high-frequency components at both ends, when the image data ahead of the vehicle is differentiated in the horizontal direction, peaks are obtained at both ends of the white line, and the peaks are within the white line, the white line from the outside of the white line, and the white line from the white line to the white line. Since it is obtained positively and negatively with respect to the outside, the white line portion can be estimated. By applying a white line enhancement filter that performs such processing to the image data, the white line portion can be emphasized, and from the image data in which the white line is emphasized, features such as white line characteristics, high brightness, and a linear shape A white line can be detected by applying a technique such as matching to a certain area.

  The map data ECU 16 is connected to a road shape information storage unit 18 for storing the acquired road shape information. The map data stored in the map DB 17 described above includes an actual road network connected to nodes (for example, points where roads and roads intersect, points separated from intersections at predetermined intervals, etc.) and links (nodes and nodes are connected). Corresponding to (road), it is configured as a table-like database.

  The map data ECU 16 stores the link information of the map data extracted from the map DB 17 in the road shape information storage unit 18. That is, link information is stored in the road shape information storage unit 18 in association with latitude and longitude. In addition, you may provide the road shape acquisition apparatus 18 integrally with map DB17.

  When the map data ECU 16 stores the link information in the road shape information storage unit 18, the map data ECU 16 generates a link plane of a rectangular area based on each link, and divides the rectangular area into an array. FIG. 2 shows an example of a link plane. The link plane corresponds to the link length connecting the start end (node) and the end point (node) of the link and the width of the road. If the width is registered in the map data, a link plane corresponding to the width is generated. If the width is unknown, for example, a link plane is formed assuming a width of about 4 lanes on one side.

  As will be described later, the white line existence probability in the grid is stored in each grid constituting the array. Therefore, the length of the grid in the width direction is equal to or less than the width of the white line (about 1/1 to 1/10). Since white lines exist continuously in the link length direction, white lines once detected are expected to be detected continuously. For this reason, the length in the link length direction of the grid that stores the white line existence probability may be longer than the width direction.

  Each link plane of the road shape information storage unit 18 is divided into a number of grids corresponding to the width based on the size of the grid set based on such a reference. The size of the grid may be uniform or variable across all link planes. In the road shape information storage unit 18, the position of each grid is designated by a grid number (n, m) designated in the width direction and the link length direction, and is handled as a table in which the white line existence probability is associated with the grid number. Can do.

  As will be described in detail later, the map data ECU 16 sets the white line existence probability for updating in the grid in consideration of the error of the detection position, and updates the white line existence probability of each grid by the Bayesian update formula every time it travels repeatedly. By doing so, the position of the white line is detected with higher accuracy as the number of travels increases, and the road shape is acquired.

  In FIG. 2, as an example of road shape information obtained by a plurality of times of travel, a left white line on the left side and a right white line on the right side are shown with respect to the traveling direction of the vehicle. Since the white line existence probability is set for each grid, a curved road shape can be acquired even if the link plane is substantially rectangular, and a highly accurate road shape can be obtained.

  The white line existence probability will be described. First, the concept of the existence probability of a three-dimensional object when a three-dimensional object is detected by a radar will be described with reference to FIG. In the present embodiment, the existence probability of the detection target is updated using a Bayesian update formula described below.

  In FIG. 3A, the origin is the position of the radar, and a three-dimensional object is detected at a position away from it by a predetermined distance. In the case of a radar, when a three-dimensional object is detected, it can be determined that there is no object up to the three-dimensional object, but it is unknown whether there is a three-dimensional object behind the three-dimensional object. Therefore, in the three-dimensional object detection by the radar, for example, when a three-dimensional object is observed at a predetermined distance in a certain direction, the area is divided into three parts, that is, the observed position, the front side, and the rear side, and the corresponding grid is observed. The results are interpreted as event Z occurred, event Z ′ occurred, and no information was obtained. In FIG. 3A, the observation result is shown by a dotted line. That is, the probability of the vicinity of the grid where the three-dimensional object is detected is close to 1, the probability of the front grid is close to 0, and the back is between 0 and 1 (0.5).

  Since detection by radar also includes errors, the accuracy of determining the presence or absence of solid objects is improved by reflecting past observation results in the latest observation results rather than determining the presence or absence of solid objects in a single observation. Can do.

The event that a solid object exists is represented by m and the existence probability of a solid object in a grid is expressed by P (m). The existence probability of a solid object in each grid can be updated by calculating the following conditional probability. .
• For the grid where event Z occurred: P (m | Z)
• For the grid where event Z 'occurred: P (m | Z')
P (m | Z) = P (Z | m) · P (m) / {P (Z | m) · P (m) + P (Z | m ′) · P (m ′)} (1)
P (m | Z ′) = P (Z ′ | m) · P (m) / {P (Z ′ | m) · P (m) + P (Z ′ | m ′) · P (m ′)} ... (2)
It becomes.

  Equation (1) is a Bayesian update equation used in this embodiment. P (m) is the prior probability, and the initial value is 0.5. Other terms are P (Z ′ | m) = 1−P (Z | m), P (Z ′ | m ′) = 1−P (Z | m ′), P (m ′) = 1−P (m).

  Also, it is necessary to determine P (Z | m) in order to calculate the Bayesian update formula, and these are models of the uncertainty of the radar sensor. That is, the white line existence probability P (Z | m) having a dispersion as shown by the solid line in FIG. Then, the white line existence probability P (Z | m) is applied to each corresponding grid.

  The calculated P (m | Z) and P (m | Z ′) are P (m) when a solid object is detected next time on the grid. In this way, every time a three-dimensional object is detected by scanning the radar with a certain grid, the existence probability of the detection target object of the grid where the event Z has occurred is increased according to the equations (1) and (2). It can be updated to reduce the existence probability of the grid where the 'occurred (the previous grid).

  In this embodiment, P (Z | m) is a model of the uncertainty of the position of white line detection. If there is no error in the position of the vehicle and the detection position of the white line, the road shape can be acquired by detecting the white line once, but in practice the error is inevitable.

Therefore, the road shape acquisition device 10 of the present embodiment is
a) White line existence probability P (Z | m) in the white line detection result at time t (latest)
b) White line existence probability P (m) in detection results from time 0 to t-1 (previous)
To the Bayesian renewal formula
c) White line existence probability P (m | Z) for detection results at times 0 to t
Is used to update P (m).

  FIG. 3B is a diagram for explaining the concept of white line existence probability applied to the link plane. Since the white line continuously exists in the longitudinal direction of the link plane, each grid corresponding to the white line existence probability P (Z | m) having a predetermined dispersion perpendicularly to the longitudinal direction of the link around the position where the white line is detected. Applies to For the front side in the longitudinal direction of the link, a white line existence probability P (Z | m) having a predetermined dispersion is set in the same manner as the solid line in FIG.

  However, for example, when the distribution of the white line existence probability is a simple normal distribution, it is not preferable that the edge in the width direction of the white line becomes unclear or a considerable number of times of traveling is required until the edge is detected.

  Therefore, in the present embodiment, the white line existence probability P (Z | m) as shown in FIG. 4 is applied. FIG. 4 shows the white line existence probability P (Z | m) applied to the Bayesian update formula by the road shape acquisition apparatus of the present embodiment. In FIG. 4, the X direction indicates the direction perpendicular to the longitudinal direction of the link, and the Y direction indicates the longitudinal direction of the link. The position where the white line was detected was taken as the origin.

  The white line existence probability P (Z | m) in FIG. 4 has a normal distribution centered on the detection position of the white line and has a probability distribution with concave portions on both sides. The portion of the normal distribution has a width of standard deviation ± dn, and the width of the recess is a predetermined length ds (for example, dn × constant). The recess has a certain probability of existence, and its value is set to less than 0.5. Further, the white line existence probability P (Z | m) of the portion of the normal distribution is added with 0.5 to the white line existence probability P (Z | m) of the concave portion. The white line existence probability P (Z | m) on the side farther from the recess is 0.5 (initial value).

  The white line existence probability P (Z | m) may not be a normal distribution centered on the detection position of the white line, and dn and ds are designable values. In particular, by making the distribution of the white line existence probability P (Z | m) variable according to the number of times of travel (the number of updates), the detection accuracy of the white line can be improved.

  Further, since the detected white line may be predicted to be continuous in the longitudinal direction of the link, for example, a white line existence probability P (Z | m) constant at 10 m is applied in the Y direction. This is because, as shown in FIG. 2, it is possible to predict that a white line is accurately detected in a region (Far region) that is a predetermined distance from the vehicle, and that the white line is continuous from there. Although the Far area is 10 m, this is variable depending on the performance of the camera 14 and the image processing device 15.

  When the white line is detected, the map data ECU 16 applies the white line existence probability P (Z | m) as shown in FIG. 4 to the equation (1). That is, the white line existence probability P (Z | m) in FIG. 4 is assigned to P (Z | m) in equation (1) in association with each grid.

  In Formula (1), if a value greater than 0.5 is substituted for P (Z | m), P (m | Z) is greater than 0.5, and P (Z | m) is less than 0.5. If P is substituted, P (m | Z) becomes a value smaller than 0.5. If 0.5 is substituted for P (Z | m), P (m | Z) becomes 0.5 (does not change).

  As shown in FIG. 4, for example, by providing concave portions on both sides of the normal distribution and setting the white line existence probability P (Z | m) of the concave portions to less than 0.5, information that can be detected by the radar but cannot be detected by the camera (three-dimensional in front) Can be compensated).

  Accordingly, the white line existence probability P (m) is increased in the range of standard deviation ± dn from the position where the white line is detected, and the white line existence probability P is in the range (recessed portion) of the standard deviation ± dn to ds from the position where the white line is detected. (M) can be reduced. In other regions, the white line existence probability P (m) is not changed and can be left at the initial value.

  FIG. 5 is a flowchart showing a processing procedure in which the map data ECU 16 acquires a road shape. The flowchart of FIG. 5 starts when the ignition of the vehicle is turned on, for example. The process in FIG. 5 is a batch process in which white line recognition results are once recorded, but the road shape may be acquired in real time.

  While traveling, the map data ECU 16 records the detection result of the white line in correspondence with these while acquiring the position of the vehicle by GPS or INS and the traveling direction by the gyro sensor 13 (S1). The detection result of the white line detects the relative distance from the position of the vehicle. Since white lines exist on both sides of the vehicle, the detection results are recorded for the left and right white lines.

  Next, the map data ECU 16 extracts a link corresponding to the vehicle position from the road shape information storage unit 18 (S2). A link plane is generated according to the link length and the like, a grid is formed, and an initial value of the white line existence probability P (m) is set in each grid (S3).

  Subsequently, the map data ECU 16 sets the white line existence probability P (Z | m) of FIG. 4 to each corresponding grid based on the detection result of the white line (S4).

  Then, using the Bayes formula, the white line existence probability P (m) of each grid is updated (S5).

  FIG. 6A is a diagram showing an image of the white line existence probability P (m) obtained for the link traveled for the first time. By applying the white line existence probability P (Z | m) of FIG. 4, the white line existence probability P (m) of the grid to which the white line existence probability P (Z | m) larger than 0.5 is applied is increased and is smaller than 0.5. The white line existence probability P (m) of the grid to which the white line existence probability P (Z | m) is applied can be reduced. In FIG. 6A, the elongated ellipse portion indicates a grid having a white line existence probability larger than 0.5. It is difficult to detect all white lines in a single run, and it is uncertain what kind of running trajectory is drawn and whether white lines are normally recognized, so the white line existence probability P (m) is more than 0.5. Large grids are randomly distributed in stripes.

  Subsequently, in the state as shown in FIG. 6A, the map data ECU 16 repeats steps S4 and S5 when a white line is detected next time by traveling on the same link.

  When traveling on the same link next time, a white line at the same position as the previous time may be detected, or a white line not detected the previous time may be detected. The white line existence probability P (m) of the grid corresponding to the white line at the same position as the previous time is determined by the white line existence probability P (Z | m) of FIG. Further, the white line existence probability of the grid to which the white line existence probability smaller than 0.5 is applied is further reduced. Therefore, as the number of runnings increases, the white line existence probability P (m) can be emphasized only in the grid around which the white line actually exists, so that the edge of the white line becomes clearer. Become.

  FIG. 6B is a diagram showing an image of the white line existence probability P (m) obtained for a link that has traveled a plurality of times. When a grid having a white line existence probability P (m) larger than 0.5 is continuously obtained by traveling a plurality of times, a road shape can be acquired based on the continuous grid.

  The map data ECU 16 obtains the road shape with the white line existence probability P (m) being a predetermined line or more as a white line position (S6).

  FIG. 7 is a diagram showing a road shape actually acquired by the road shape acquisition device of the present embodiment. In FIG. 7, a vehicle equipped with a road shape acquisition device travels the same general road four times in total and acquires the road shape based on the detection result of the white line. FIGS. 7A to 7D show road shapes acquired when the number of travels is 1 to 4, respectively. 7A and 7B, a grid having a white line existence probability of 0.7 or more is detected as a white line. In addition, the center line of the road was set at a location where the detected white line was offset to the right by a predetermined distance. Note that the near side of FIG. 7 includes a range in which a white line is not detected because the range not captured by the camera 14 is included.

  As shown in FIG. 7A, when the number of times of travel is 1, the grid detected as having a white line is sparse (s1 to s6), but a continuous grid is obtained as the number of times of travel increases. Has been.

  In addition, the detection result of FIG. 7 gradually decreases the dispersion in the width direction (using a narrower distribution instead of the normal distribution) as the number of times of travel increases. Thus, as the white line existence probability increases, the position of the white line can be detected with high accuracy by reducing the variance.

  As described above, according to the road shape acquisition device 10 of the present embodiment, the road shape can be acquired from the white line detection result, so that the driver is alerted before the signal or route guidance at an appropriate timing. Can do. Further, even when the road network changes due to construction or the like, the map DB 17 can be updated to a new road shape by traveling several times. Even when a new road is constructed, the road shape can be similarly obtained by adding a link.

  The road shape acquisition device 10 according to the present embodiment introduces a white line existence probability in consideration of an error in the detected position, and updates the white line existence probability by a Bayesian update formula every time it travels repeatedly, so that the number of times of traveling increases. White line position can be detected with high accuracy.

  In addition, the white line detection result compensates for the fact that the amount of information is less than the detection of three-dimensional objects by radar. Therefore, the white line existence probability P (Z | m) applied to the Bayesian update formula is devised, so that the white line portion is accurate. It became possible to detect well.

It is a block diagram of a road shape acquisition apparatus. It is a figure which shows an example of a link plane. It is a figure for demonstrating the concept of the white line presence probability applied to the detected white line. This is the white line existence probability applied to the Bayesian update formula by the road shape acquisition device. It is a flowchart figure which shows the process sequence in which map data ECU acquires a road shape. It is a figure which shows the image of the white line presence probability obtained about the drive | worked link. It is a figure which shows the road shape which the road shape acquisition apparatus actually acquired.

Explanation of symbols

DESCRIPTION OF SYMBOLS 10 Road shape acquisition apparatus 11 GPS receiver 12 Vehicle speed sensor 13 Gyro sensor 14 Camera 15 Image processing apparatus 16 Map data ECU
17 Map DB
18 Road shape information storage unit

Claims (2)

A road shape acquisition device that divides a link plane generated based on a link constituting a road into grids and acquires a road shape from the grid having a white line existence probability equal to or higher than a predetermined value,
A map data storage means for storing link information of the road,
For each grid, white line existence probability storage means for storing the updated white line existence probability or the initial white line existence probability ;
Position detecting means for detecting the position of the vehicle;
White line detecting means for detecting the white line of the road;
White lines existence probability of the grid in a predetermined range from said grid corresponding to said detected white lines, and the white line existence probability by applying the white line existence probability stored in the white line existence probability memory means Bayesian updating expression White line existence probability setting means for updating the white line existence probability stored in the storage means ;
A road shape acquisition apparatus comprising: When the width direction of the white line is X,
The white line existence probability of a grid in a predetermined range from the grid corresponding to the white line detected by the white line detection means is:
a) the white line within dn in the X direction from the grid position X1 detected is greater than the initial value of the white line existence probability,
b) The grid position X1 is smaller than the initial value in the grid away from dn in the X direction and within ds,
c) Same as the initial value in the grid distant from ds in the X direction from the grid position X1,
The road shape acquisition apparatus according to claim 1, wherein the road shape acquisition apparatus is a road shape acquisition apparatus.

Images