JP2020026985A - Vehicle position estimation device and program - Google Patents

Abstract

To accurately estimate the position of a vehicle, even when the state of a compartment line is different from a map data.SOLUTION: A reliability determination unit 208 determines the reliability of a detected vehicle position, on the basis of a reliability vector having, as an element, a plurality of types of indices showing the coincidence between a compartment line of a road of the map data at the detected vehicle position and a detected compartment line, and on the basis of a previously-learned learned model for determining the reliability of the reliability vector. A learning vector update unit 209 updates the learned model, on the basis of the reliability vector obtained when the reliability is a previously determined first threshold or more.SELECTED DRAWING: Figure 1

Description

  The present invention relates to a vehicle position estimation device and a program.

  2. Description of the Related Art Conventionally, in an automatic driving system, when a difference has occurred between a road state and map data due to an update of a road state, updating of the map data based on a predetermined threshold or automatic There is a technique for stopping operation control.

  For example, Patent Literature 1 includes a map database and a process of calculating map update data from a surrounding environment, and compares the calculated data with the map database. When it is determined that an error between the two is large, the database is updated. Update. In this case, the determination is made based on the fact that the error is equal to or greater than the determination value, the number of times the deviation has occurred, and the reliability of the data. The reliability of the data is set based on the recognition state of the external environment by the stereo camera, the vehicle behavior (yaw rate and lateral acceleration change) detected by the on-board sensor, and the GPS reception status.

  Patent Document 2 further includes a map database, a lane marking detecting unit by image processing, and a vehicle position estimating unit by GPS. When it is determined that the detected road shape is different from the road shape obtained from the map, Stop the vehicle control. In this case, the determination is made on the condition that the center position of the road is different from the estimation result in the map.

Japanese Patent No. 5997797 JP 2006-2900072 public relations

  The technique of Patent Document 1 uses the data reliability during traveling and the number of deviations for the determination of map data update, and compares it with a predetermined threshold. However, the quality of road lane markings varies depending on the region and the management company, and a predetermined threshold value may not be applicable in a specific region. In this case, there is a possibility that erroneous warnings may be continued in road sections having different lane marking specifications.

  Further, in the technique of Patent Literature 2, the stop of the control is determined by detecting the deviation of the road shape. However, a small scale such as a change in the position of only one lane marking or a change in the lane width is performed. The difference to the road update cannot be determined. Therefore, there is a possibility that the control cannot be switched even if the vehicle position accuracy is reduced.

  The present invention has been made in order to solve the above problems, and has a vehicle position estimating apparatus and a program that can accurately estimate the position of a vehicle even when the state of lane markings is different from map data. The purpose is to provide.

  In order to achieve the above object, a vehicle position estimating device of the present invention includes: a lane marking detecting unit that detects a lane marking of a road on which a vehicle travels, based on an image input by an imaging device; The lane markings of the roads included in the map data including the information on the lane markings and the lane markings detected by the lane marking detector are compared with each other. A position corresponding to the map data, a vehicle position detecting unit that detects as a vehicle position, a lane marking of a road of the map data at the vehicle position detected by the vehicle position detecting unit, and a lane marking detected by the lane marking detecting unit. A reliability vector calculation unit that calculates a reliability vector having a plurality of types of indices indicating the degrees of coincidence with the lane markings, and the reliability vector, A reliability determination unit that determines the reliability of the vehicle position detected by the vehicle position detection unit based on the learned model for determining the reliability of the reliability vector, and the reliability is predetermined. A learning vector updating unit that updates the learned model based on the reliability vector obtained when the reliability vector is equal to or greater than a first threshold.

  In addition, the program according to the present invention includes the computer in a map data including information on a lane marking of a road based on an image input by the imaging device, the lane marking detecting unit detecting a lane marking of a road on which the vehicle travels. A lane marking verification unit that verifies the lane markings of the road to be checked and the lane markings detected by the lane marking detection unit, and a position corresponding to the map data that is most closely matched by the lane marking verification unit. A vehicle position detecting unit that detects a position, indicating a degree of coincidence between a lane marking of the road in the map data at the vehicle position detected by the vehicle position detecting unit and the lane marking detected by the lane marking detecting unit; A reliability vector calculation unit that calculates a reliability vector having a plurality of types of indices as elements, the reliability vector, and a reliability of the reliability vector that has been learned in advance. A reliability determination unit that determines the reliability of the vehicle position detected by the vehicle position detection unit based on a learned model for determining the reliability, and the reliability is equal to or greater than a predetermined first threshold. A program for functioning as a learning vector updating unit that updates the learned model based on the obtained reliability vector.

  According to the state prediction device and the program of the present invention, the lane marking detection unit detects the lane marking of the road on which the vehicle travels based on the image input by the imaging device, and the lane marking verification unit determines the lane marking of the road. The lane markings of the roads included in the map data including the information about the lines are collated with the lane markings detected by the lane marking detection unit, and the vehicle position detection unit compares the map data with the best matching result by the lane marking verification unit. Is detected as a vehicle position.

  Then, the reliability vector calculation unit calculates a plurality of types of indices indicating the degree of coincidence between the lane markings of the road in the map data at the vehicle position detected by the vehicle position detecting unit and the lane markings detected by the lane marking detecting unit. The reliability determination unit calculates a reliability vector as an element, and the reliability determination unit uses the vehicle position detection unit based on the reliability vector and a learned model for determining the reliability of the reliability vector that has been learned in advance. The reliability of the detected vehicle position is determined, and the learning vector updating unit updates the learned model based on the reliability vector obtained when the reliability is equal to or greater than a predetermined first threshold.

  As described above, a reliability vector having a plurality of types of indices indicating the degrees of coincidence between the detected lane markings and the lane markings of the map data at the detected vehicle position as elements, The reliability of the detected vehicle position is determined based on the learned model for determining the reliability of the vector, and the reliability vector obtained when the reliability is equal to or greater than a predetermined first threshold is determined. By updating the learned model based on the map data, the position of the vehicle can be accurately estimated even when the state of the lane markings differs from the map data.

  Further, the learned model of the vehicle position estimating device of the present invention has been learned in advance for each of a plurality of areas, and the learning vector updating unit may determine that the reliability is equal to or greater than the first threshold. The learned model for the area to which the vehicle position belongs can be updated based on the reliability vector.

  Further, the lane marking detection unit of the vehicle position estimating device of the present invention detects a point on the lane marking based on the image, and a plurality of types of indices serving as elements of the reliability vector are included in the map data. And the percentage of the detected inline points, which are included in a predetermined range from the position of the lane marking of the road, which is the point of the lane marking detected.

  Further, the ratio of the inlier point of the vehicle position estimating device of the present invention is a density of the inlier point with respect to the length of the lane marking included in the map data, a ratio of the inlier point with respect to all the detected points, and a traveling direction of the vehicle. The ratio may be at least one of the ratio of the inlier point to each of the left and right lane markings.

  In addition, the vehicle position estimating device of the present invention further includes an acquisition unit that acquires positioning information obtained by measuring the position of the vehicle from a GPS sensor, wherein the learning vector updating unit is configured such that the reliability is equal to or greater than a predetermined first threshold. And, when the difference between the vehicle position indicated by the positioning information and the vehicle position detected by the vehicle position detection unit is equal to or less than a predetermined second threshold value, based on the reliability vector, The trained model can be updated.

  Further, the vehicle position estimating device of the present invention further includes an acquisition unit that acquires an override detection result from an override detector that detects the presence or absence of an override, wherein the learning vector updating unit includes a first threshold value whose reliability is predetermined. As described above, when the override detection result indicates that there is no override, the learned model can be updated based on the reliability vector.

  Further, the vehicle position estimating device of the present invention, when the reliability is equal to or more than the first threshold value, controls the vehicle to automatically drive using the vehicle position and the map data. When the value is less than the first threshold value, the vehicle may further include an automatic driving control unit that controls the vehicle to perform automatic driving using the lane marking detected by the lane marking detecting unit.

  Further, the vehicle position estimating device of the present invention, when the reliability is equal to or more than the first threshold value, controls the vehicle to automatically drive using the vehicle position and the map data. Is less than the first threshold value, and when the number of inlier points is equal to or more than a predetermined third threshold value, the automatic driving of the vehicle using the lane marking detected by the lane marking detection unit is performed. An automatic operation control unit that controls to switch to manual operation when the reliability is less than the first threshold and the number of inlier points is less than a predetermined third threshold. May be further included.

  As described above, according to the vehicle position estimating apparatus and the program of the present invention, even when the state of the lane markings is different from the map data, the position of the vehicle can be accurately estimated. Can be

It is a block diagram showing a schematic structure in an embodiment of the invention. It is an image figure showing the example of the extracted inlier point. It is a figure which shows an example of the presence or absence of reliability. It is an image figure showing the example of the difference of the inlier distribution according to a road. 5 is a flowchart showing the contents of an automatic operation processing routine according to the embodiment of the present invention.

  Hereinafter, embodiments of the present invention will be described in detail with reference to the drawings.

<System configuration of automatic operation control system>
As shown in FIG. 1, an automatic driving control system 10 according to an embodiment of the present invention includes an imaging device 100, a GPS sensor 110, an override detector 120, and a vehicle position estimating device 200. .

  The imaging device 100 is mounted on a vehicle, and captures an image of a runway. Then, the imaging device 100 passes the captured image to the acquisition unit 201.

  The GPS sensor 110 receives positioning information obtained from each GPS satellite. The positioning information includes the pseudo distance and the Doppler frequency. Then, the GPS sensor 110 passes the received positioning information to the acquisition unit 201.

  The override detector 120 detects whether or not the vehicle has been overridden, and passes the detection result to the acquisition unit 201.

  The vehicle position estimating device 200 includes a CPU, a RAM, and a ROM storing a program for executing a state estimating process routine to be described later, and is functionally configured as follows.

  The vehicle position estimating device 200 includes an acquisition unit 201, a lane marking detection unit 202, a map data storage unit 203, a lane marking collation unit 204, a vehicle position detection unit 205, a reliability vector calculation unit 206, a learning vector The storage unit 207, the reliability determination unit 208, the learning vector updating unit 209, the automatic driving control unit 210, and the output unit 211 are provided.

  The acquisition unit 201 receives an input of an image from the imaging device 100 and passes the received image to the lane marking detection unit 202.

  Further, the acquisition unit 201 receives the input of the positioning information from the GPS sensor 110 and the input of the override detection result from the override detector 120, and passes the received positioning information and the override detection result to the learning vector updating unit 209. In addition, the acquisition unit 201 passes the received positioning information to the reliability determination unit 208.

  The lane marking detection unit 202 detects a lane marking on the road on which the vehicle travels based on the image input by the imaging device 100.

  Specifically, the lane marking detection unit 202 detects, from the image, a point group representing the center position of the lane marking. More specifically, the marking line detection unit 202 scans each row of the image, and detects both edges of the marking line using a difference filter or the like. When the interval between both end edges of the division line is equal to or smaller than a predetermined threshold (for example, 60 cm), the center position of the both end edges is set as a point representing the center position of the division line.

  The marking line detection unit 202 uses, for each of the point groups representing the center position of the detected marking line, the focal length of the camera and the center coordinate parameter, and converts the coordinates to coordinates around the vehicle by Inverse Perspective Mapping. Group. The division line point group represents the division line of the road on which the vehicle travels at coordinates around the vehicle.

  Then, the lane marking detection unit 202 passes the lane marking point group representing the obtained lane markings to the lane marking verification unit 204 and the automatic operation control unit 210.

  The map data storage unit 203 stores map data including information on lane markings of roads.

  The map data is a high-precision map storing the shape of the lane markings on the road, and is described by a set of nodes constituting the lane markings with respect to the shape of the lane markings on the road. Each node is described by latitude, longitude, and altitude.

In addition, for each node, the vehicle position and the vehicle azimuth before correction by the vehicle position detection unit 205 are used to convert the coordinates into the east / west / north / south coordinate system and the coordinates (lateral direction, traveling direction) centered on the vehicle (reference Reference 1).
[Reference Document 1] "Calculation of plane rectangular coordinates x, y and meridional aberration angle from latitude and longitude", GSI.

  The lane marking collation unit 204 collates the lane markings of the road included in the map data with the lane marking point groups representing the lane markings detected by the lane marking detection unit 202.

  Specifically, the lane marking collation unit 204 acquires the map data from the map data storage unit 203, and collates the lane markings of the map data with the detected lane marking point group. At this time, a detected demarcation line point that is near (for example, ± 25 cm) the demarcation line of the map data is regarded as an inlier.

Detected parcel line point

Is the parcel of map data

The following conditions 1 or 2 can be adopted as conditions for inliers. However, the acquisition range of the lane marking node in the map data is

And

<< Condition 1 >>
The detected lane mark point exists within the range of the vertical position of the lane mark of the map data, that is, the lane mark point that satisfies the following equation (1).

<< Condition 2 >>
The detected demarcation line points should be located near the left and right of the demarcation line in the map data. That is, it is a section line point satisfying the following equation (2).

Here, in the formula (2), _{a i} is the slope represented by the above formula (3), _{b i} is the x-intercept of the above formula (4). Also,

Is near the division line of the map data, and can be set to, for example, 0.25 [m].

The detected parcel line points that correspond to the parcel lines in the map data are inline points (

) (FIG. 2).

  Then, the lane marking verification unit 204 passes the obtained inlier point group to the vehicle position detection unit 205.

  The vehicle position detection unit 205 detects, as a vehicle position, a position corresponding to the map data whose matching result obtained by the lane marking matching unit 204 matches the most.

  Specifically, the vehicle position detection unit 205 obtains the vehicle position at which the inlier point and the division line of the map data are most associated by the least square method or the like.

  For example, in the case of obtaining by the least square method, a variable u [m] representing the amount of horizontal displacement (positive in the right direction), a variable v [m] representing the amount of vertical displacement (positive in the traveling direction), and The vehicle position is determined such that the least square sum F is minimized by using a variable ω [rad] representing the amount of azimuth deviation (clockwise as positive) as a variable.

  In this case, the minimized sum of squares F is represented by the following equation (5). It should be noted that some terms are approximated to reduce to a linear least squares problem.

Here, in the above equation (5), a _{i [k]} is the slope of the division line i of the map data associated with the inlier point k, and bi _[k] is the map data of the map data associated with the inlier point k. It is an x-intercept of the division line i.

Also,

Is solved, and the shift amount of the horizontal position, the vertical position, and the azimuth angle is obtained by the following equation (6).

As a vehicle position before the correction in the present process, a rough vehicle position can be estimated by road surface image collation using the method of Reference 2.
[Reference Document 2] Japanese Patent No. 6225889 (vehicle position estimation device and program).

  The vehicle position detection unit 205 performs a correction to subtract the obtained lateral position, vertical position, and deviation amount of the azimuth angle from the vehicle position and the vehicle azimuth before the correction, respectively, so that the vehicle position and the vehicle after the correction are corrected. Find the azimuth.

Further, a result obtained by applying the Kalman filter described in Reference 3 to the corrected vehicle position may be used as the vehicle position.
[Reference Document 3] JP-A-2018-21777

  Then, the vehicle position detection unit 205 passes the corrected vehicle position to the reliability determination unit 208 and the automatic driving control unit 210.

  The reliability vector calculation unit 206 includes a plurality of types of indices indicating the degree of coincidence between the lane markings of the road in the map data at the vehicle position detected by the vehicle position detector 205 and the lane markings detected by the lane marking detector 202. Is calculated as a reliability vector.

Specifically, the reliability vector calculation unit 206 calculates the inlier density, the inlier ratio, and the inlier left / right Calculate the ratio. The calculation formula is as follows.
Inlier density = MIN ((L + R) / S, 2)
However, when S is 0, the inlier ratio to substitute 0 = (L + R) / J
However, when J is 0, 0 is substituted.
Inlier left / right ratio = (LR) / MAX (L, R).
However, when MAX (L, R) is 0, -1 is substituted.

Where L is

Where R is the number of inlier points

Is the number of inlier points. S is the total length of the lane marking

(Only the i that satisfies the above equation (1) is subject to the summation), and J is the total number of points included in the detected lane marking point group.

  The reliability vector calculation unit 206 uses the calculated inlier density, inlier ratio, and inlier left-right ratio as reliability vectors.

  In addition, the reliability vector may include a least square sum F in addition to the inlier density, the inlier ratio, and the inlier left-right ratio.

  When the Kalman filter (Reference Document 3) is used in the processing of the vehicle position detection unit 205, the difference between the vehicle lateral position before applying the Kalman filter and the vehicle lateral position after applying the Kalman filter may be included in the reliability vector element. Good.

  When the technique of Reference 2 is used in the processing of the lane line collation unit 204, the correlation value of the image collation may be included in the element of the reliability vector. The correlation value of the image comparison is represented, for example, by the following equation (7).

  Here, h (i, j) is the correlation image in Reference 2.

  Then, the reliability vector calculation unit 206 passes the reliability vector to the reliability determination unit 208 and the learning vector update unit 209.

  The learning vector storage unit 207 stores a learned model that has been learned in advance and that determines the reliability of the reliability vector. In the present embodiment, a learning vector will be described as an example of a learned model.

Specifically, the learning vector includes a total number of support vectors n and an i-th support vector x _i.

, The i-th alpha parameter α _i

, A raw parameter ρ, and a gamma parameter γ.

  The reliability determination unit 208 determines the reliability of the vehicle position detected by the vehicle position detection unit 205 based on the reliability vector and a learning vector that is learned in advance and determines the reliability of the reliability vector. judge.

  Specifically, the reliability determination unit 208 performs reliability determination based on the reliability vector y calculated by the reliability vector calculation unit 206 and the learning vector stored in the learning vector storage unit 207.

In the present embodiment, the value of the following equation (8) is used as a value indicating reliability using One class SVM [Ref. 4] and RBF kernel [Ref. 5].
[Reference 4] Scholkopf, B., Plattz, JC, “Estimating the Support of A High Dimensional Distribution”, Neurral Computation, vol. 13, pp. 1443-1472, 2001.
[Reference 5] C. Burges. “A tutorial on support vector machines for pattern recognition”, Knowledge Discovery and Data Mining, vol. 2, no. 2, 1998.

  The reliability determination unit 208 determines that there is reliability when the value obtained by the above equation (8) is equal to or greater than a predetermined first threshold, and that there is no reliability otherwise. For example, when the first threshold value is 0, it is determined that there is reliability when the value obtained by the above equation (8) is positive (≧ 0), and when there is a negative value (<0), there is no reliability. (FIG. 3).

  FIG. 3 shows the result of using a trained classifier for the inlier density (horizontal axis) and the inlier rate (vertical axis). Elements other than the inlier density (horizontal axis) and the inlier rate (vertical axis) are averaged. Indicates the discrimination boundary when considered. The white area is determined to be reliable, and the gray area is determined to be unreliable. Also, the reliability vector used for learning is plotted as a black point.

  Note that a learned model in an arbitrary neural network may be used instead of the learning vector.

  Then, the reliability determination unit 208 passes the determination result to the learning vector update unit 209 and the automatic driving control unit 210.

  The learning vector updating unit 209 updates the learning vector based on the reliability vector obtained when the reliability is equal to or greater than a predetermined first threshold.

  Specifically, the learning vector updating unit 209 updates the learning vector stored in the learning vector storage unit 207 using the reliability vector when the determination result of the reliability determining unit 208 is reliable.

  In the present embodiment, One class SVM (Ref. 4) is used, and RBF kernel (Ref. 5) is used for the support vector update processing.

  Since One class SVM is used, a positive sample, that is, a reliability vector when the vehicle position is reliable is used as a learning sample.

  As a method of selecting a learning sample, at least one of the following conditions 1 to 3 is used.

  (Condition 1) The reliability determination unit 208 registers a reliability vector determined to be reliable as a learning sample. In order to prevent over-learning, a configuration may be adopted in which a reliability vector that is not determined to be reliable at a rate of 1/100 is also registered as a learning sample.

  (Condition 2) If the difference between the vehicle position obtained by the vehicle position detection unit 205 and the vehicle position obtained from the positioning information of the GPS sensor 110 is equal to or smaller than a predetermined second threshold value, and the GPS positioning situation is good. The reliability vector at the time of the determination is registered as a learning sample. When the difference is a horizontal position difference and the second threshold value is 0.25, the difference can be determined by the following equation (9).

Here, (e _gps , n _gps ) is the position [m] obtained by ENU-converting the vehicle position based on the GPS positioning information according to Reference 1, and (e, n) is the estimated vehicle position according to Reference 1. This is the position [m] obtained by the ENU conversion, and θ is the vehicle azimuth [rad].

  (Condition 3) A reliability vector when the automatic control by the automatic driving control unit 210 is determined to be continued for the past 5 seconds is registered as a learning sample. Whether or not the condition 3 is satisfied may be determined by being notified by the automatic driving control unit 210, or may be determined by not being detected for 5 seconds by the override detector 120 that the override is present.

  The learning vector is learned in advance for each of the plurality of areas, and the reliability determination unit 208 determines the reliability using the learning vector for the area to which the vehicle position belongs, and updates the learning vector. The unit 209 can be configured to update the learning vector for the area to which the vehicle position belongs based on the reliability vector obtained when the reliability is equal to or greater than the first threshold.

  Specifically, the reliability determination unit 208 determines the area to which the vehicle position belongs from the positioning information obtained by the GPS sensor 110 or the road ID information accompanying the detected map data, and uses a learning vector for the area. And the reliability is determined. Further, the learning vector updating unit 209 updates different learning vectors for each area. This makes it possible to make a determination reflecting the characteristics of each area.

  If One class SVM is used, a unique distribution for each area can be learned by unsupervised learning, but using the above (Condition 1) to (Condition 3) is more reliable.

  The characteristics of each area include the following.

<< Differential marking line >>
The length and cycle of the broken line differs for each road. For example, the standard of the lane marking differs between the expressway and the general road, and the standard also differs between the expressways. As a result, there is a difference in the tendency of the “inlier density” and the “inlier left-right ratio” of the reliability vector. On a road with sparse lane markings, it is desirable that the learning vectors be distributed so as not to be sensitive to the “inlier density” and the “inlier left-right ratio”.

<< Differences in road maintenance status >>
On an unmanaged road, the “inlier density” of the reliability vector tends to decrease as a whole due to aging of the lane markings. It is desirable to lower the threshold within appropriate limits (FIG. 4, right).

  On the other hand, on a well-managed road, it is necessary to increase the sensitivity to malfunctions caused by updating of the road. Therefore, it is desirable to set a higher threshold.

<< Differences by cartography companies >>
Different companies create maps for each area, resulting in differences in the accuracy of the location of the lane markings. Since an incorrect warning is more likely to be issued to a map with lower accuracy, it is necessary to set an identification boundary according to the tendency of the map error.

<< Differences due to climatic conditions >>
If there is snow, the distribution of "inlier rate" will be different by recognizing snow and snow remover as lane markings.

  In view of the characteristics of each area, parameter tuning is required for each area. However, unlike parameters for vehicle position detection, parameters for vehicle position reliability detection have high ambiguity and are difficult to adjust. Therefore, in the present embodiment, clustering is performed for each area, and classified into areas where it is better to use a learning vector common to all areas and areas where it is better to use a learning vector specific to the area.

  For example, the following method is used for clustering for each area.

  First, a case where a learning vector common to all areas is used and a case where a learning vector specific to an area are used are tested in parallel. Thereafter, when it is determined that the determination based on the learning vector peculiar to the area is less likely to be erroneous, utilizing the feedback by manual intervention or the like, the class of the area is made independent.

  According to such a method, it is possible to classify into areas where it is better to use a learning vector common to all areas and areas where it is better to use an area-specific learning vector, and perform parameter tuning for each area.

  When the reliability is equal to or higher than the first threshold, the automatic driving control unit 210 controls the vehicle to perform automatic driving using the vehicle position and the map data, the reliability is lower than the first threshold, and When the number of inlier points is equal to or more than a predetermined third threshold value, control is performed such that the vehicle automatically drives using the lane marking detected by the lane marking detection unit 202, and the reliability is less than the first threshold. If there is and the number of inlier points is less than a predetermined third threshold, control is performed to switch to manual operation.

  Specifically, when the reliability determination unit 208 determines that there is reliability, the automatic driving control unit 210 performs vehicle control based on the vehicle position obtained by the vehicle position detection unit 205. In this case, referring to the map data, vehicle control based on the road structure such as branching and merging can be performed.

  On the other hand, when the reliability determination unit 208 determines that there is no reliability, the automatic driving control unit 210 determines the reliability of lane line detection. For example, when the number of detected inlier points is equal to or greater than a predetermined third threshold (for example, 10), it is determined that the reliability of the lane marking detection is high.

When the reliability of lane marking detection is high, the automatic driving control unit 210 performs vehicle control based on lane marking detection. For example, the method of Reference 6 can be used.
[Reference Document 6] JP-A-2018-005567

  When the reliability determination unit 208 determines that the reliability is not high and determines that the reliability of the lane marking detection is low (when the number of detected inlier points is less than a predetermined third threshold). ), Control to stop the vehicle control and switch to manual driving.

  Then, the automatic driving control unit 210 passes the control command to the output unit 211.

  The output unit 211 outputs a control command from the automatic operation control unit 210.

<Operation of Vehicle Position Estimating Device 200>
Next, an automatic operation control processing routine of the automatic operation control system 10 of the present embodiment will be described with reference to FIG.

  First, in step S100, the lane marking collation unit 204 acquires map data from the map data storage unit 203.

  In step S110, the acquisition unit 201 receives an input of an image from the imaging device 100.

  In step S120, the lane marking detection unit 202 detects a lane marking on the road on which the vehicle travels based on the image received in step S110.

  In step S130, the lane marking matching unit 204 compares the lane markings of the road included in the map data with the lane markings detected in step S120.

  In step S140, the vehicle position detection unit 205 detects, as the vehicle position, a position corresponding to the map data whose matching result in step S130 is the best.

  In step S150, the reliability vector calculation unit 206 calculates a plurality of indices indicating the degree of coincidence between the lane markings of the road in the map data at the vehicle position detected in step S140 and the lane markings detected in step S120. Is calculated as a reliability vector.

  In step S160, the reliability determination unit 208 determines an area to which the vehicle position detected in step S140 belongs, and acquires a learning vector corresponding to the area stored in the learning vector storage unit 207.

  In step S170, the reliability determining unit 208 obtains the reliability of the vehicle position detected in step S140 based on the reliability vector and the learning vector acquired in step S160.

  In step S180, the reliability determination unit 208 determines whether the reliability obtained in step S170 is equal to or greater than a predetermined first threshold.

  If the reliability is equal to or greater than the predetermined first threshold (YES in step S180), in step S190, the learning vector updating unit 209 acquires the positioning information of the GPS sensor 110.

  In step S200, the learning vector updating unit 209 obtains a difference between the vehicle position obtained by the vehicle position detecting unit 205 and the vehicle position obtained from the positioning information of the GPS sensor 110.

  In step S210, the learning vector updating unit 209 determines whether the difference obtained in step S200 is equal to or smaller than a predetermined second threshold.

  If the difference is equal to or smaller than the second predetermined threshold (YES in step S210), in step S220, the learning vector updating unit 209 acquires an override detection result by the override detector 120.

  On the other hand, if the difference is equal to or smaller than the second threshold value (NO in step S210), the process proceeds to step S260.

  In step S230, the learning vector updating unit 209 determines whether or not there is an override.

  If there is an override (YES in step S230), the process proceeds to step S260.

  On the other hand, when there is no override (NO in step S230), in step S240, the learning vector updating unit 209 determines the vehicle based on the reliability vector obtained when the reliability is equal to or larger than the predetermined first threshold. Update the learning vector for the area to which the position belongs.

    In step S250, when the reliability is equal to or greater than the first threshold, the automatic driving control unit 210 controls the vehicle to perform automatic driving using the vehicle position and the map data.

  If the reliability is not equal to or higher than the predetermined first threshold (NO in step S180), in step S260, the automatic driving control unit 210 determines whether the number of inlier points is equal to or higher than the predetermined third threshold. Is determined.

  When the number of inlier points is equal to or greater than the third threshold value determined in advance (YES in step S260), in step S270, the automatic driving control unit 210 performs automatic driving of the vehicle using the lane marking detected in step S120. Is controlled to be performed.

  On the other hand, if the number of inlier points is not equal to or greater than the third threshold value (NO in step S260), in step S280, the automatic driving control unit 210 controls to switch to manual driving.

  In step S290, the output unit 211 outputs a control command from the automatic operation control unit 210.

  As described above, according to the vehicle position estimating device according to the embodiment of the present invention, a plurality of types of indicating the degree of coincidence between the lane markings of the road in the map data at the detected vehicle positions and the detected lane markings The reliability of the detected vehicle position is determined based on the reliability vector having the index of the element and the learned model for determining the reliability of the reliability vector that has been learned in advance, and the reliability is determined. By updating the learned model based on the reliability vector obtained when the value is equal to or greater than the predetermined first threshold value, the position of the vehicle can be accurately determined even if the state of the lane marking differs from the map data. Can be estimated.

  The present invention is not limited to the above-described embodiment, and various modifications and applications can be made without departing from the gist of the present invention.

  In addition, in the specification of the present application, the embodiment is described in which the program is installed in advance. However, the program may be stored in a computer-readable recording medium and provided.

Reference Signs List 10 automatic driving control system 100 imaging device 110 sensor 120 override detector 200 vehicle position estimating device 201 acquisition unit 202 lane line detection unit 203 map data storage unit 204 lane line collation unit 205 vehicle position detection unit 206 reliability vector calculation unit 207 learning Vector storage unit 208 Reliability determination unit 209 Learning vector update unit 210 Automatic operation control unit 211 Output unit

Claims (9)

A lane marking detection unit that detects a lane marking on a road on which the vehicle travels, based on an image input by the imaging device;
A lane marking of a road included in the map data including information on lane markings of the road, and a lane marking verification unit that verifies the lane marking detected by the lane marking detection unit;
A vehicle position detection unit that detects a position corresponding to the map data where the collation result obtained by the lane marking collation unit best matches as a vehicle position;
A reliability including a plurality of types of indices indicating the degree of coincidence between a lane marking of the road in the map data at the vehicle position detected by the vehicle position detecting unit and the lane marking detected by the lane marking detecting unit. A reliability vector calculation unit for calculating a degree vector,
A reliability for determining the reliability of the vehicle position detected by the vehicle position detection unit based on the reliability vector and a learned model for determining the reliability of the reliability vector that has been learned in advance. A sex determination unit;
A learning vector updating unit that updates the learned model based on the reliability vector obtained when the reliability is equal to or greater than a predetermined first threshold;
A vehicle position estimating device including: The learned model is learned in advance for each of a plurality of areas,
The vehicle according to claim 1, wherein the learning vector updating unit updates the learned model for an area to which the vehicle position belongs based on the reliability vector when the reliability is equal to or greater than the first threshold. Position estimation device. The lane marking detection unit detects a point on the lane marking based on the image,
The plurality of types of indices serving as elements of the reliability vector include a percentage related to an inlier point, which is a point of the detected lane marking, included in a predetermined range from a position of a lane marking included in the map data. Item 3. The vehicle position estimation device according to item 1 or 2.   The ratio of the inlier point is the density of the inlier point with respect to the length of the lane marking included in the map data, the ratio of the inlier point to all the detected points, and the left and right lane markings with respect to the traveling direction of the vehicle. The vehicle position estimating device according to claim 3, wherein the vehicle position estimating device is at least one of the inlier point ratios. An acquisition unit for acquiring positioning information obtained by measuring the position of the vehicle from a GPS sensor,
The learning vector updating unit is configured such that the reliability is equal to or greater than a predetermined first threshold, and a difference between a vehicle position indicated by the positioning information and a vehicle position detected by the vehicle position detecting unit is predetermined. The vehicle position estimating device according to claim 1, wherein the learned model is updated based on the reliability vector when the value is equal to or less than a second threshold. An acquisition unit that acquires an override detection result from an override detector that detects the presence or absence of an override;
The learning vector updating unit, based on the reliability vector, when the reliability is equal to or greater than a predetermined first threshold and the override detection result indicates that there is no override, based on the reliability vector. The vehicle position estimating device according to any one of claims 1 to 4. When the reliability is equal to or greater than the first threshold, control is performed to perform automatic driving of the vehicle using the vehicle position and the map data, and when the reliability is less than the first threshold, The vehicle position estimating device according to claim 1, further comprising: an automatic driving control unit configured to perform automatic driving of the vehicle using the lane marking detected by the lane marking detecting unit. When the reliability is equal to or more than the first threshold, control is performed to perform automatic driving of the vehicle using the vehicle position and the map data, and the reliability is less than the first threshold, and When the number of the inlier points is equal to or more than a predetermined third threshold value, control is performed such that the vehicle automatically drives using the lane markings detected by the lane marking detecting unit, and the reliability is the reliability. The vehicle according to claim 3, further comprising an automatic driving control unit configured to control to switch to manual driving when the number of inlier points is less than a first threshold value and the number of inlier points is less than a predetermined third threshold value. Position estimation device. Computer
A lane marking detection unit that detects a lane marking of a road on which the vehicle travels, based on an image input by the imaging device;
A lane marking matching unit that compares a lane marking of a road included in the map data including information on lane markings of the road with the lane marking detected by the lane marking detection unit;
A vehicle position detection unit that detects a position corresponding to the map data at which the collation result by the lane marking collation unit best matches as a vehicle position,
A reliability including a plurality of types of indices indicating the degree of coincidence between a lane marking of the road in the map data at the vehicle position detected by the vehicle position detecting unit and the lane marking detected by the lane marking detecting unit. A reliability vector calculation unit that calculates a degree vector,
A reliability for determining the reliability of the vehicle position detected by the vehicle position detection unit based on the reliability vector and a learned model for determining the reliability of the reliability vector that has been learned in advance. A program for functioning as a sex determination unit, and a learning vector updating unit that updates the learned model based on the reliability vector obtained when the reliability is equal to or greater than a predetermined first threshold.