Vehicle azimuth angle estimation method, program, vehicle azimuth angle estimation device, and autonomous mobile body

The method uses Hough transform and LiDAR/GNSS/INS integration for precise vehicle azimuth estimation, addressing accuracy issues in user-dependent methods by enhancing estimation robustness and accuracy.

JP2024160569APending Publication Date: 2024-11-14KANAZAWA UNIV
View PDF 1 Cites 0 Cited by

Patent Information

Application Number
JP2023075709
Authority / Receiving Office
JP · JP
Patent Type
Applications
Current Assignee / Owner
Filing Date
2023-05-01
Publication Date
2024-11-14

AI Technical Summary

Technical Problem

Existing vehicle azimuth estimation methods rely on user corrections, leading to insufficient accuracy in determining vehicle orientation.

Method used

A method involving Hough transform of road and map images to calculate correlation distributions and probability distributions for accurate azimuth estimation, using LiDAR for data acquisition and integrating GNSS/INS for position and attitude detection.

Benefits of technology

Enhances vehicle azimuth estimation accuracy, particularly in environments with complex road layouts, by reducing ambiguity and improving robustness through probability distribution updates and image division techniques.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure 2024160569000001_ABST
    Figure 2024160569000001_ABST
Patent Text Reader

Abstract

To provide a vehicle azimuth angle estimation method, or the like which can accurately estimate an azimuth angle of a vehicle.SOLUTION: A vehicle azimuth angle estimation method for estimating an azimuth angle in a front-back direction of an own vehicle includes: a correlation distribution calculation step S20 for calculating a distribution of correlation between a first Hough spectrum obtained by performing a Hough transform on an observation image as an image of a road on which the own vehicle travels and a second Hough spectrum obtained by performing the Hough transform on a map image corresponding to the observation image; and a probability distribution calculation step S30 for calculating an observation probability distribution as a probability distribution of an azimuth angle error relative to the map image of the observation image on the basis of the correlation distribution.SELECTED DRAWING: Figure 12
Need to check novelty before this filing date? Find Prior Art

Description

[Technical field]

[0001] The present invention relates to a vehicle azimuth angle estimation method, a program, a vehicle azimuth angle estimation device, and an autonomous moving body. [Background technology]

[0002] Currently, research into autonomous driving of moving bodies such as automobiles is being actively conducted (see Patent Document 1, etc.). In autonomous driving of a moving body, it is necessary to estimate the azimuth angle in the forward and backward directions of the moving body. For example, Patent Document 1 describes an invention that corrects the position information and azimuth angle information of a vehicle based on a surrounding image based on the surrounding environment including structures around the vehicle and a map around the vehicle. The invention described in Patent Document 1 attempts to display the surrounding image and the map in a superimposed manner, and correct the position information and azimuth angle information of the vehicle based on a correction instruction from the user of the vehicle in response to the display. [Prior art documents] [Patent documents]

[0003] [Patent Document 1] Patent No. 6669059 Summary of the Invention [Problem to be solved by the invention]

[0004] However, in the invention described in Patent Document 1, the accuracy of correction depends on the contents of the user's instruction, and therefore sufficient accuracy of correction cannot necessarily be obtained.

[0005] Therefore, an object of the present invention is to provide a vehicle azimuth angle estimation method capable of estimating the vehicle azimuth angle with high accuracy. [Means for solving the problem]

[0006] In order to achieve the above object, a vehicle azimuth angle estimation method according to one aspect of the present invention is a method for estimating a longitudinal azimuth angle of a host vehicle, and includes a correlation distribution calculation step of calculating a correlation distribution between a first Hough spectrum obtained by subjecting an observation image, which is an image of a road along which the host vehicle travels, to a Hough transform, and a second Hough spectrum obtained by subjecting a map image corresponding to the observation image to a Hough transform, and a probability distribution calculation step of calculating, based on the correlation distribution, an observation probability distribution, which is a probability distribution of an azimuth angle error of the observation image with respect to the map image.

[0007] In order to achieve the above object, a program according to one aspect of the present invention is a program for causing a computer to execute the vehicle azimuth angle estimation method.

[0008] In order to achieve the above object, a vehicle azimuth angle estimation device according to one embodiment of the present invention is a device for estimating a longitudinal azimuth angle of a host vehicle, and includes: a correlation distribution calculation unit that calculates a correlation distribution between a first Hough spectrum obtained by subjecting an observation image acquired in the host vehicle to a Hough transform and a second Hough spectrum obtained by subjecting a map image to a Hough transform; and a probability distribution calculation unit that calculates, based on the correlation distribution, an observation probability distribution which is a probability distribution of an azimuth angle error of the observation image with respect to the map image.

[0009] In order to achieve the above object, an autonomous moving body according to one aspect of the present invention includes the vehicle azimuth angle estimation device and an acquisition unit that acquires the observed image.

[0010] In addition, these comprehensive or specific aspects may be realized by a system, a method, an integrated circuit, a computer program, or a non-transitory computer-readable recording medium such as a CD-ROM, or may be realized by any combination of the system, the method, the integrated circuit, the computer program, and the recording medium. Effect of the Invention

[0011] According to the present invention, it is possible to provide a vehicle azimuth angle estimation method capable of estimating a vehicle azimuth angle with high accuracy. [Brief description of the drawings]

[0012] [Figure 1] FIG. 1 is a block diagram showing an example of a configuration of a vehicle azimuth angle estimation device and an autonomous moving body according to an embodiment. [Diagram 2] FIG. 1 is a diagram showing an example of an observation image according to an embodiment; [Diagram 3] 1 is a graph showing an example of a first Hough spectrum according to an embodiment; [Figure 4] FIG. 1 is a diagram showing an example of a map image corresponding to an observation image in the embodiment; [Diagram 5] Graph showing an example of a second Hough spectrum according to an embodiment. [Figure 6] Graph showing an example of a correlation distribution between a first Hough spectrum and a second Hough spectrum according to an embodiment of the present invention. [Figure 7] 1 is a schematic graph showing an observation probability distribution according to an embodiment; [Figure 8] 1 is a schematic graph showing a prior probability distribution according to an embodiment; [Figure 9] 1 is a schematic graph showing a posterior probability distribution according to an embodiment; [Figure 10] 1 is a schematic graph illustrating a method for converting an observation probability distribution according to an embodiment, and a probability range of the observation probability distribution; [Figure 11] Graph showing an example of the relationship between the variance of the observed probability distribution and the probability width of the observed probability distribution in the embodiment. [Figure 12] A flowchart showing the flow of a vehicle azimuth angle estimation method according to an embodiment. [Figure 13] A photograph showing a road on which an experiment was conducted with a vehicle azimuth angle estimation device according to an embodiment of the present invention. [Figure 14] FIG. 1 is a diagram showing an example of an observation image acquired by a vehicle azimuth angle estimation device of a comparative example; [Figure 15] FIG. 13 is a diagram showing the estimation results of the vehicle position when the vehicle azimuth angle estimation device according to the embodiment and the vehicle azimuth angle estimation device of the comparative example are used; [Figure 16] FIG. 13 is a diagram showing an example of a division mode of an observation image according to a modified example. [Figure 17] FIG. 13 is a diagram showing the relationship between the size of the divided observed image and the radius of curvature of the road on which the host vehicle is traveling according to a modified example. DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS

[0013] Hereinafter, an embodiment of the present invention will be described in detail with reference to the drawings.

[0014] The embodiments described below are all comprehensive or specific examples. The numerical values, shapes, materials, components, component arrangement and connection forms, steps, and order of steps shown in the following embodiments are merely examples and are not intended to limit the present invention. Furthermore, among the components in the following embodiments, components that are not described in an independent claim showing a top concept are described as optional components.

[0015] In addition, each drawing is a schematic diagram and is not necessarily a precise illustration. In addition, in each drawing, the same components are denoted by the same reference numerals.

[0016] (Embodiment) A vehicle azimuth angle estimation method, a vehicle azimuth angle estimation device, and an autonomous moving body according to an embodiment will be described.

[0017] [1. Configuration of vehicle azimuth angle estimation device and autonomous moving body] First, the configuration of a vehicle azimuth angle estimation device and an autonomous moving body according to this embodiment will be described with reference to FIG.

[0018] FIG. 1 is a block diagram showing an example of the configuration of a vehicle azimuth angle estimation device 100 and an autonomous moving body V10 according to this embodiment.

[0019] The autonomous moving body V10 is a moving body that includes a vehicle azimuth angle estimation device 100 and an acquisition unit 10 and performs automatic driving. The autonomous moving body V10 is not particularly limited as long as it is a moving body. In this embodiment, the autonomous moving body V10 is a vehicle that performs automatic driving.

[0020] The acquisition unit 10 acquires an observation image, which is an image of a road on which the vehicle travels. The vehicle is a vehicle for which the vehicle azimuth estimation device 100 estimates the azimuth angle in the forward and backward directions. In this embodiment, the vehicle is an autonomous moving body V10. Hereinafter, the autonomous moving body V10 is also referred to as the vehicle. In this embodiment, position information of objects around the vehicle is acquired. The objects for which the acquisition unit 10 acquires position information include moving objects and stationary objects around the vehicle. For example, the moving objects include vehicles and pedestrians, and the stationary objects include stopped vehicles, buildings, signs, and the like. In this embodiment, the acquisition unit 10 acquires three-dimensional information of the direction in which the surrounding objects are located, the distance to the surface of the object, and the height, based on the installation position of the acquisition unit 10. For example, a three-dimensional information acquisition device such as LiDAR (Light Detection and Ranging) or a stereo camera can be used as the acquisition unit 10. In addition, a system combining a monocular camera and AI (Artificial Intelligence) can be used as the acquisition unit 10. In such a system, it is possible to extract three-dimensional information from an image captured by a monocular camera using AI. In this embodiment, the acquisition unit 10 includes a LiDAR.

[0021] The LiDAR measures the distance between the target object and the observation body equipped with the LiDAR by irradiating the target object with a laser light and acquiring the laser light reflected from the target object. The acquisition unit 10 acquires 3D point cloud data as height information indicating the relationship between the horizontal position of the road surface and the height from the measured distance, the position of the LiDAR, and the direction of the laser light irradiation.

[0022] The LiDAR irradiates, for example, 128 laser beams in the vertical direction (i.e., the vertical direction) at different irradiation angles. The vertical viewing angle of the LiDAR is, for example, -25 deg or more and 15 deg or less, with the horizontal direction (i.e., the direction perpendicular to the vertical direction) being 0 deg. The LiDAR can acquire omnidirectional three-dimensional point cloud data by rotating the irradiation direction of these laser beams by 360 deg in the horizontal direction. The scan rate of the laser beams is, for example, 5 Hz or more and 20 Hz or less. In this embodiment, the scan rate is set to 10 Hz. That is, in this embodiment, the LiDAR acquires 10 frames of three-dimensional point cloud data per second.

[0023] The acquisition unit 10 acquires an observation image from the three-dimensional point cloud data. An example of the observation image will be described below with reference to FIG. 2. FIG. 2 is a diagram showing an example of an observation image according to the present embodiment. FIG. 2 shows an angle θ1 representing an angle in the azimuth direction. As shown in FIG. 2, the observation image includes information such as road markings, the positions of the lateral edges of the road, and the positions of the shoulders. This makes it possible to obtain information about the direction of the road on which the vehicle is traveling (i.e., the direction in which the road extends) from the observation image. Since the relative relationship between the observation direction of the acquisition unit 10 and the longitudinal direction of the vehicle can be specified, the relative relationship between the observation image and the longitudinal direction of the vehicle can be specified.

[0024] The acquisition unit 10 may acquire the position, attitude, speed, acceleration, and the like of the vehicle. To detect this information, the acquisition unit 10 uses, for example, a GNSS / INS (Global Navigation Satellite System / Inertial Navigation System). In this embodiment, the acquisition unit 10 detects information such as the position, attitude, speed, and acceleration of the vehicle at a cycle of, for example, 100 Hz by integrating position information obtained by observing an artificial satellite using the GNSS / INS with measurement information of an IMU (Inertial Measurement Unit). The IMU is a composite system including a gyro sensor, an acceleration sensor, and the like.

[0025] The vehicle azimuth angle estimation device 100 is a device that estimates an azimuth angle in the forward / rearward direction of the vehicle. As shown in Fig. 1, the vehicle azimuth angle estimation device 100 includes a correlation distribution calculation unit 20 and a probability distribution calculation unit 30. In this embodiment, the vehicle azimuth angle estimation device 100 further includes an update unit 40, a driving control unit 70, and a storage unit 80.

[0026] The correlation distribution calculation unit 20 is a processing unit that compares an observation image acquired by the vehicle with a map image. In this embodiment, the correlation distribution calculation unit 20 performs a Hough transform on the observation image acquired by the acquisition unit 10. The Hough transform is a type of processing method for extracting geometric elements such as straight lines from an image. In this embodiment, the correlation distribution calculation unit 20 obtains a first Hough spectrum by performing a Hough transform on the observation image.

[0027] Here, the first Hough spectrum will be described with reference to FIG. 3. FIG. 3 is a graph showing an example of the first Hough spectrum according to the present embodiment. The first Hough spectrum is a graph showing the relationship between the angle θ1 and the intensity of the straight line component in the angle θ1 direction in the observed image. The horizontal axis of FIG. 3 shows the angle θ1 representing the angle in the azimuth direction, and the vertical axis shows the intensity of the straight line component in the angle θ1 direction in the observed image. Such a first Hough spectrum can detect the angle θ1 at which many straight line components are included in the road markings, the lateral edges of the road, the shoulder strips, etc. Such an angle θ1 at which many straight line components are included often corresponds to the direction of the road. Therefore, the direction of the road can be estimated from the first Hough spectrum.

[0028] In addition, in this embodiment, the correlation distribution calculation unit 20 performs a Hough transform on a map image corresponding to the observation image. An example of the map image will be described below with reference to FIG. 4. FIG. 4 is a diagram showing an example of a map image corresponding to the observation image according to this embodiment. Note that FIG. 4 shows an angle θ2 representing an angle in the azimuth direction. The map image is an image showing an actual map of the position of the vehicle and its surroundings. The map image is extracted, for example, from high-precision three-dimensional map data (HD map), a dynamic map, etc. acquired in advance. The high-precision three-dimensional map data, the dynamic map, etc. are acquired by measuring in advance at each position. The high-precision three-dimensional map data, the dynamic map, etc. may be stored in advance in the storage unit 80 of the vehicle, etc., or may be sent to the vehicle by communication. The map image corresponding to the observation image is a map image including a map of the same area as the area shown by the observation image. In this embodiment, the correlation distribution calculation unit 20 acquires the map image from the storage unit 80 based on the position information of the vehicle acquired by the acquisition unit 10, etc.

[0029] 4, the map image includes information such as road markings, the positions of the lateral edges of the road, and the positions of the shoulders, just like the observed image. The map image is stored, for example, in the storage unit 80. In this embodiment, the correlation distribution calculation unit 20 obtains a second Hough spectrum by performing a Hough transform on the map image.

[0030] The second Hough spectrum will now be described with reference to Fig. 5. Fig. 5 is a graph showing an example of the second Hough spectrum according to this embodiment. The second Hough spectrum is a graph showing the relationship between the angle θ2 and the intensity of the straight line component in the angle θ2 direction in the map image. The horizontal axis of Fig. 5 shows the angle θ2 representing the azimuth angle, and the vertical axis shows the intensity of the straight line component in the angle θ2 direction in the map image. With such a second Hough spectrum, the road direction can be estimated in the same way as with the first Hough spectrum.

[0031] The correlation distribution calculation unit 20 calculates the correlation distribution between the first Hough spectrum and the second Hough spectrum. Such a correlation distribution will be described with reference to FIG. 6. FIG. 6 is a graph showing an example of the correlation distribution between the first Hough spectrum and the second Hough spectrum according to the present embodiment. The correlation distribution is a distribution showing the relationship between the azimuth angle error Δθ, which is the amount of angular deviation between the direction of the observation image and the direction of the map image, and the correlation value between the first Hough spectrum and the second Hough spectrum. The correlation value between the first Hough spectrum and the second Hough spectrum is a value showing the degree of agreement between the first Hough spectrum and the second Hough spectrum. The larger the correlation value is, the higher the degree of agreement between the graphs showing each spectrum is. In other words, it is estimated that the direction of the image obtained by rotating the observation image in the azimuth direction by the azimuth angle error Δθ at which the correlation value is the largest is likely to match the direction of the map image. In other words, it is estimated that the azimuth angle error Δθ with the largest correlation value is most likely to be the true azimuth angle offset of the direction of the observed image relative to the direction of the map image (the exact offset angle in the azimuth direction of the direction of the observed image relative to the direction of the map image).

[0032] As described above, the relationship between the direction of the observation image and the longitudinal direction of the vehicle can be specified, and therefore, by estimating the relationship between the direction of the observation image and the direction of the map image, the relationship between the longitudinal direction of the vehicle and the direction of the map image can be estimated. In other words, the azimuth angle of the longitudinal direction of the vehicle can be estimated with high accuracy. Since vehicles such as automobiles move at high speed, if the estimation accuracy of the azimuth angle decreases, the estimation accuracy of the position of the vehicle also decreases significantly. For this reason, the effect of the vehicle azimuth angle estimation device 100 according to this embodiment is remarkable, particularly when applied to vehicles such as automobiles.

[0033] The probability distribution calculation unit 30 is a processing unit that calculates an observation probability distribution, which is a probability distribution of the azimuth angle error Δθ of the observed image with respect to the map image, based on the correlation distribution calculated by the correlation distribution calculation unit 20. The observation probability distribution is a distribution of probability indicating the certainty of the azimuth angle error Δθ, and corresponds to the correlation distribution. The larger the correlation value in the correlation distribution for the azimuth angle error Δθ, the higher the probability. The observation probability distribution will be explained using FIG. 7. FIG. 7 is a schematic graph showing the observation probability distribution according to this embodiment. The horizontal axis of FIG. 7 indicates the azimuth angle error Δθ, and the vertical axis indicates the probability p(Δθ|z t ) shown in FIG. t It is estimated that the azimuth error Δθ for which Δθ is maximum is highly likely to be the true azimuth offset of the direction of the observed image relative to the direction of the map image.

[0034] The update unit 40 is a processing unit that updates the prior probability distribution, which is the probability distribution of the azimuth angle error Δθ calculated based on the observation probability distribution calculated in the past, to a posterior probability distribution based on the latest observation probability distribution. The prior probability distribution and the posterior probability distribution will be described below with reference to Figs. 8 and 9. Figs. 8 and 9 are schematic graphs showing the prior probability distribution and the posterior probability distribution according to this embodiment, respectively. The horizontal axis of Figs. 8 and 9 indicates the azimuth angle error Δθ. The vertical axis of Fig. 8 indicates the probability p(Δθ|z 1:t-1 ) The vertical axis of Fig. 9 shows the probability p(Δθ|z 1:t ) The update unit 40 is not an essential component of the vehicle azimuth estimation device 100. The vehicle azimuth estimation device 100 may estimate the azimuth based on the azimuth offset calculated by the probability distribution calculation unit 30 without using the update unit 40. By providing the vehicle azimuth estimation device 100 with the update unit 40, the azimuth can be estimated with even greater accuracy.

[0035] In this embodiment, every time an observed image is acquired by the acquisition unit 10 (that is, every 0.1 seconds), the correlation distribution calculation unit 20 calculates a correlation distribution and calculates an observation probability distribution based on the observed image. The update unit 40 according to this embodiment updates the prior probability distribution calculated based on the observation probability distribution calculated up to time t-1 (0.1 seconds before time t in this embodiment) immediately before the latest observation probability distribution calculated at time t is calculated, to a posterior probability distribution based on the latest observation probability distribution.

[0036] In this embodiment, the update unit 40 uses a BBF (Binary Bayes Filter) to update the probability distribution. From the BBF, the following formula (1) can be derived.

[0037]

number

[0038] In equation (1), p(Δθ i |z t ) is the azimuth angle error Δθ in the observation probability distribution at time t i p(Δθ i |z 1:t-1 ) is the azimuth angle error Δθ in the prior probability distribution up to time t-1 i p(Δθ i |z 1:t ) is the azimuth angle error Δθ in the posterior probability distribution up to time t i p(Δθ i ) is the azimuth error Δθ i represents the probability that the azimuth offset is the true azimuth offset. Here, the probability p(Δθ i ) can usually be considered to be 0.5, so using equation (1), the prior probability distribution (probability p(Δθ i |z 1:t-1 )) to the latest observed probability distribution (with probability p(Δθ i |z t ) to obtain the posterior probability distribution (probability p(Δθi |z 1:t )) can be updated.

[0039] In this way, by updating the prior probability distribution to the posterior probability distribution by the update unit 40 using the observation probability distribution, even when the accuracy of the azimuth angle estimation based on the latest observation probability distribution is low, a posterior probability distribution with high accuracy can be obtained by using the information of the prior probability distribution. That is, the robustness of the accuracy of the azimuth angle estimation can be improved. For example, when the vehicle is located near an intersection, the observed image may contain few center lines of the road, or may contain many straight line components indicating the direction of other roads intersecting the road in addition to the straight line components indicating the direction of the road on which the vehicle is traveling. For this reason, the accuracy of the azimuth angle estimation based on the observation probability distribution may decrease. However, in this embodiment, the update unit 40 uses the information of the prior probability distribution to suppress the decrease in the accuracy of the azimuth angle estimation.

[0040] In this embodiment, the updating method may be changed according to the variance of the observation probability distribution. Such an updating method will be described with reference to Fig. 10 and Fig. 11. Fig. 10 is a schematic graph illustrating a method for converting an observation probability distribution according to this embodiment, and the probability width Wp of the observation probability distribution. Fig. 11 is a graph showing an example of the relationship between the variance V of the observation probability distribution and the probability width Wp of the observation probability distribution according to this embodiment.

[0041] For example, when the vehicle is located near an intersection as described above, the variance V of the observation probability distribution corresponding to the observation image becomes large. That is, the ambiguity of the azimuth angle error Δθ of the observation image becomes high. If the prior probability distribution is updated using an observation probability distribution with high ambiguity, that is, low reliability, the accuracy of the posterior probability distribution may decrease. Therefore, in this embodiment, the update unit 40 may convert the observation probability distribution so that the probability of the observation probability distribution approaches 0.5 as the variance V of the observation probability distribution increases. For example, the observation probability distribution as shown by the dashed line in FIG. 10 may be converted to the observation probability distribution as shown by the solid line in FIG. 10. As a result, the probability of the observation probability distribution approaches 0.5, and the width Wp of the probability (the difference between the maximum value and the minimum value of the probability of the observation probability distribution) is reduced. Here, when the probability of the observation probability distribution is higher than 0.5, the probability of the posterior probability distribution is higher than the probability of the prior probability distribution. When the probability of the observation probability distribution is lower than 0.5, the probability of the posterior probability distribution is lower than the probability of the prior probability distribution. When the probability of the observation probability distribution is 0.5, the probability of the posterior probability distribution does not change relative to the probability of the prior probability distribution. Therefore, by converting the observation probability distribution, the probability approaches 0.5, and the influence of the observation probability distribution on the posterior probability distribution can be suppressed. Therefore, it is possible to suppress a decrease in the accuracy of the posterior probability distribution due to an observation probability distribution with a large variance V and high ambiguity.

[0042] The conversion method of the observation probability distribution according to the present embodiment is not particularly limited as long as it is a conversion method in which the probability of the observation probability distribution approaches 0.5 (i.e., the probability width Wp is reduced) as the variance V of the observation probability distribution increases. For example, as shown in FIG. 11, the probability width Wp may be converted according to the variance V. In the example shown in FIG. 11, when the variance V is less than 4.5, the probability width Wp is 1, and when the variance V is 4.5 or more, the probability distribution is converted so that the probability width Wp becomes (9 / V)-1. Note that, in the conversion method in which the probability of the observation probability distribution approaches 0.5 as the variance of the observation probability distribution increases, there may be a range of the variance V in which the probability width Wp is constant (in the example shown in FIG. 11, a range in which the variance V is less than 4.5).

[0043] The driving control unit 70 is a processing unit that controls the driving of the host vehicle. In the present embodiment, the driving control unit 70 controls the driving based on the azimuth angle in the forward and backward directions of the host vehicle estimated based on the update unit 40. The driving control unit 70 controls the traveling direction of the host vehicle, etc., based on, for example, map information stored in the storage unit 80, the position of the host vehicle, and the azimuth angle in the forward and backward directions of the host vehicle.

[0044] The storage unit 80 stores information for azimuth angle estimation. For example, the storage unit 80 stores map information including information on the road on which the vehicle is driven. The storage unit 80 may also store information for automatic driving.

[0045] [2. Vehicle azimuth angle estimation method] The vehicle azimuth angle estimation method according to this embodiment will be described with reference to Fig. 12. Fig. 12 is a flowchart showing the flow of the vehicle azimuth angle estimation method according to this embodiment.

[0046] First, as shown in FIG. 12, the acquisition unit 10 acquires an observation image (acquisition step S10).

[0047] Next, the correlation distribution calculation unit 20 calculates the correlation distribution between a first Hough spectrum obtained by applying a Hough transform to an observed image, which is an image of a road on which the vehicle is traveling, and a second Hough spectrum obtained by applying a Hough transform to a map image corresponding to the observed image (correlation distribution calculation step S20).

[0048] Next, the probability distribution calculation unit 30 calculates an observation probability distribution, which is a probability distribution of the azimuth angle error Δθ of the observation image relative to the map image, based on the correlation distribution calculated in the correlation distribution calculation step S20 (probability distribution calculation step S30). Based on the azimuth angle error Δθ with the maximum probability in the thus obtained observation probability distribution, the relationship between the direction of the observation image and the direction of the map image can be estimated with high accuracy. As described above, since the relationship between the direction of the observation image and the forward / backward direction of the host vehicle can be specified, the relationship between the direction of the observation image and the direction of the map image can be estimated by estimating the relationship between the direction of the observation image and the direction of the map image. In other words, the azimuth angle in the forward / backward direction of the host vehicle can be estimated with high accuracy.

[0049] Next, the update unit 40 updates the prior probability distribution, which is the probability distribution of the azimuth angle error Δθ calculated based on the observation probability distribution calculated in the past, to a posterior probability distribution based on the latest observation probability distribution (updating step S40). In this way, by updating the prior probability distribution to a posterior probability distribution using the observation probability distribution by the update unit 40, even when the accuracy of the azimuth angle estimation based on the latest observation probability distribution is low, a posterior probability distribution with high accuracy can be obtained using the information of the prior probability distribution. In other words, the robustness of the accuracy of the azimuth angle estimation can be improved. In the updating step S40, the update unit 40 may use the BBF as described above.

[0050] Furthermore, in the updating step S40, the updating section 40 may convert the observation probability distribution so that the probability of the observation probability distribution approaches 0.5 as the variance V of the observation probability distribution increases, as described above.

[0051] Note that the updating step S40 is not an essential component of the vehicle azimuth angle estimation method according to the present embodiment. The vehicle azimuth angle estimation method may not include the updating step S40, and may estimate the azimuth angle based on the azimuth angle offset calculated by the probability distribution calculation step S30. By including the updating step S40 in the vehicle azimuth angle estimation method, the azimuth angle can be estimated with even greater accuracy.

[0052] Then, the process returns to the acquisition step S10 and the above steps are repeated.

[0053] By using the above-described vehicle azimuth angle estimation method, the azimuth angle in the forward / rearward direction of the vehicle can be estimated with high accuracy.

[0054] [3. Experimental Results] The experimental results of the vehicle azimuth estimation device 100 according to the present embodiment and the vehicle azimuth estimation method will be described with reference to Figs. 13 to 15 in comparison with a comparative example. Fig. 13 is a photograph showing a road on which an experiment was conducted with the vehicle azimuth estimation device 100 according to the present embodiment. Fig. 14 is a diagram showing an example of an observation image acquired by the vehicle azimuth estimation device of the comparative example. In this comparative example, azimuth estimation is not performed using an observation image or the like, and azimuth information and position information acquired by GNSS / INS are used. In addition, in this comparative example, azimuth information and position information of the vehicle are estimated by image matching between an observation image acquired by LiDAR and a map image. Fig. 15 is a diagram showing the estimation result of the vehicle position when the vehicle azimuth estimation device 100 according to the present embodiment and the vehicle azimuth estimation device of the comparative example are used. The dotted line L1 shown in FIG. 15 indicates the actual trajectory traveled by the vehicle and the trajectory of the vehicle estimated using the vehicle azimuth estimation device 100 according to the present embodiment, the dotted line L2 indicates the estimated trajectory of the vehicle when the vehicle azimuth estimation device of the comparative example is used, and the dotted line L3 indicates the trajectory of the vehicle based on the position information acquired by the GNSS / INS.

[0055] As shown in Fig. 13, the experiment was carried out on a road located below an overpass (see the dashed ellipse in Fig. 13). On such a road located below an overpass, the accuracy of the position information and azimuth angle information acquired by the acquisition unit 10 using the GNSS / INS decreases. Specifically, as shown by the dotted line L3 in Fig. 15, the position information and azimuth angle information of the vehicle itself are significantly deviated from the correct vehicle position and azimuth angle (see the dotted line L1 in Fig. 15). For this reason, in the vehicle azimuth angle estimation device of the comparative example that estimates the azimuth angle of the vehicle only using the GNSS / INS, problems such as blurring of white lines (road markings) occur in the observation image as shown in Fig. 14 (see the arrows in Fig. 14, etc.).

[0056] Moreover, according to the vehicle azimuth angle estimation device of the comparative example, the vehicle position can be estimated by image matching with higher accuracy than the vehicle position acquired by the GNSS / INS (see dotted line L3 in FIG. 15), but there is an error from the actual vehicle position (see dotted line L1 in FIG. 15). Also, as shown in FIG. 15, the error becomes larger as the distance traveled under the elevated road becomes longer.

[0057] On the other hand, when the vehicle azimuth angle estimation device 100 and the vehicle azimuth angle estimation method according to the present embodiment are used, the estimated position of the vehicle coincides with the actual position of the vehicle, as shown by the dotted line L1 in Fig. 15. In this way, the vehicle azimuth angle estimation device 100 and the vehicle azimuth angle estimation method according to the present embodiment can accurately estimate the azimuth angle in the forward and backward directions of the vehicle, and therefore can accurately estimate the position of the vehicle.

[0058] [4. Modifications] A modified example of the vehicle azimuth estimation device 100 and vehicle azimuth estimation method according to the present embodiment will be described. This modified example differs from the above embodiment in the configuration of the correlation distribution calculation unit 20 and the correlation distribution calculation step S20. In this modified example, the correlation distribution calculation unit 20 differs from the above embodiment in that it performs a Hough transform on the divided observation image. Hereinafter, the correlation distribution calculation method in the correlation distribution calculation unit 20 according to this modified example will be described with reference to Figs. 16 and 17. Fig. 16 is a diagram showing an example of a division mode of the observation image according to this modified example. Fig. 17 is a diagram showing the relationship between the dimension L of the divided observation image according to this modified example and the curvature radius R of the road on which the vehicle is traveling.

[0059] In the vehicle azimuth estimation device 100 and the vehicle azimuth estimation method according to the present embodiment, as described above, the direction of the observation image is estimated by extracting the linear components of the observation image. Therefore, for example, when the curvature radius of the road is small (i.e., the curvature is large), the azimuth estimation accuracy may decrease. Therefore, in this modification, when the curvature radius of the road on which the vehicle runs is small, the observation image is divided as shown by the broken lines in FIG. 16. In the example shown in FIG. 16, the observation image is divided into eight parts. Although not shown, the map image has a size similar to that of the divided observation image. In this modification, the map image may be divided in the same way as the observation image. In the correlation distribution calculation step S20 according to this modification, the correlation distribution calculation unit 20 calculates the correlation distribution between a first Hough spectrum obtained by performing a Hough transform on each of the divided observation images and a second Hough spectrum obtained by performing a Hough transform on a map image corresponding to the divided observation image. This makes it possible to reduce the length of the road included in the divided observation image, thereby suppressing the ambiguity (i.e., the spread) of the correlation between the first Hough spectrum and the second Hough spectrum caused by the curvature radius of the road, and therefore suppresses the deterioration of the azimuth angle estimation accuracy caused by the curvature radius of the road.

[0060] The curvature radius of the road may be detected by, for example, an acceleration sensor included in the acquisition unit 10, or may be included in the map information stored in the storage unit 80. In addition, in autonomous driving, not only images of roads but also information on the connections between roads is used, so information on the curvature of the road may be included as part of such information.

[0061] The observed image and the map image after division may be determined based on the radius of curvature of the road. This allows the observed image to be divided into appropriate dimensions according to the radius of curvature of the road. For example, as shown in FIG. 17, the observed image may be divided so that the width ΔR of the area including a curve in the divided observed image is equal to or smaller than the image resolution. Here, in the observed image, which is a square with a side length of L as shown in FIG. 17, the width ΔR of the area including a curve with a radius of curvature R is expressed by the following formula (2).

[0062]

number

[0063] For example, if the resolution of the observed image is 0.125 m, in order for the width ΔR of the region including the curve with radius of curvature R in the divided observed image to be equal to or less than the image resolution, the length L [m] of one side of the divided observed image needs to satisfy the following formula (3).

[0064]

number

[0065] As a result, the width ΔR of the region including the curve with the radius of curvature R is equal to or smaller than the resolution of the observed image, so that the curve with the radius of curvature R [m] in the divided observed image can be regarded as a substantially straight line. Therefore, the direction of the observed image can be estimated with high accuracy by the Hough transform, so that the estimation accuracy of the vehicle azimuth angle can be improved.

[0066] [4. Effects, etc.] As described above, the vehicle azimuth angle estimation method according to this embodiment is a method for estimating an azimuth angle in the forward / rearward direction of the vehicle, and includes a correlation distribution calculation step S20 for calculating a correlation distribution between a first Hough spectrum obtained by subjecting an observation image, which is an image of a road along which the vehicle is traveling, to a Hough transform, and a second Hough spectrum obtained by subjecting a map image corresponding to the observation image to a Hough transform, and a probability distribution calculation step S30 for calculating an observation probability distribution, which is a probability distribution of an azimuth angle error Δθ of the observation image relative to the map image, based on the correlation distribution.

[0067] Based on the azimuth error Δθ that maximizes the probability in the obtained observation probability distribution, the relationship between the direction of the observation image and the direction of the map image can be estimated with high accuracy. Since the relationship between the direction of the observation image and the longitudinal direction of the vehicle can be specified, the relationship between the longitudinal direction of the vehicle and the direction of the map image can be estimated by estimating the relationship between the direction of the observation image and the direction of the map image. In other words, the azimuth angle of the longitudinal direction of the vehicle can be estimated with high accuracy. Vehicles such as automobiles move at high speed, and therefore if the estimation accuracy of the azimuth angle decreases, the estimation accuracy of the position of the vehicle also decreases significantly. For this reason, the effect of the vehicle azimuth angle estimation method according to the present embodiment is particularly noticeable when applied to vehicles such as automobiles.

[0068] In addition, the vehicle azimuth angle estimation method according to this embodiment may further include an updating step S40 of updating the prior probability distribution, which is the probability distribution of the azimuth angle error Δθ calculated based on the observation probability distribution calculated in the past, to a posterior probability distribution based on the latest observation probability distribution.

[0069] In this way, by updating the prior probability distribution to the posterior probability distribution using the observation probability distribution, even if the accuracy of the azimuth angle estimation based on the latest observation probability distribution is low, a highly accurate posterior probability distribution can be obtained using the information on the prior probability distribution. In other words, the robustness of the accuracy of the azimuth angle estimation can be improved.

[0070] Furthermore, in the updating step S40 of the vehicle azimuth estimation method according to this embodiment, BBF may be used.

[0071] By using the BBF in this way, the above formula (1) can be derived, and therefore the prior probability distribution can be easily and accurately updated to the posterior probability distribution using the observation probability distribution based on the above formula (1).

[0072] In the updating step S40 of the vehicle azimuth estimation method according to this embodiment, the observation probability distribution may be converted so that the probability of the observation probability distribution approaches 0.5 as the variance V of the observation probability distribution increases.

[0073] This makes it possible to prevent the accuracy of the posterior probability distribution from being reduced due to an observation probability distribution with a large variance V and high ambiguity.

[0074] In the correlation distribution calculation step S20 of the vehicle azimuth angle estimation method according to this embodiment, the divided observed images may be subjected to a Hough transform.

[0075] This reduces the effect of the radius of curvature of the road on the accuracy of azimuth angle estimation.

[0076] In the correlation distribution calculation step S20 of the vehicle azimuth angle estimation method according to this embodiment, the observed image may be divided into dimensions determined based on the radius of curvature of the road on which the host vehicle is traveling.

[0077] This allows the observed image to be divided into appropriate dimensions according to the radius of curvature of the road. For example, the observed image may be divided so that the width ΔR of the region including the curve in the divided observed image is equal to or smaller than the resolution of the image. As a result, the width ΔR of the region including the curve with radius of curvature R is equal to or smaller than the resolution of the observed image, so that the curve with radius of curvature R [m] in the divided observed image can be regarded as substantially a straight line. Therefore, the direction of the observed image can be estimated with high accuracy by the Hough transform, thereby improving the estimation accuracy of the vehicle azimuth angle.

[0078] The program according to this embodiment is a program for causing a computer to execute a vehicle azimuth angle estimation method according to this embodiment.

[0079] This provides the same effects as those of the vehicle azimuth angle estimation method according to the present embodiment.

[0080] The vehicle azimuth angle estimation device 100 according to this embodiment is a device for estimating an azimuth angle in the forward / rearward direction of the vehicle, and includes a correlation distribution calculation unit 20 that calculates a correlation distribution between a first Hough spectrum obtained by applying a Hough transform to an observation image acquired in the vehicle and a second Hough spectrum obtained by applying a Hough transform to a map image, and a probability distribution calculation unit 30 that calculates an observation probability distribution, which is a probability distribution of an azimuth angle error Δθ of the observation image with respect to the map image, based on the correlation distribution.

[0081] This provides the same effects as those of the vehicle azimuth angle estimation method according to the present embodiment.

[0082] An autonomous moving body V10 according to this embodiment includes a vehicle azimuth angle estimation device 100 and an acquisition unit 10 that acquires an observed image.

[0083] This provides the same effects as those of the vehicle azimuth angle estimation method according to the present embodiment.

[0084] (Variations, etc.) Although the vehicle azimuth estimation method and the like according to one aspect of the present invention have been described based on the embodiment, the present invention is not limited to the embodiment. As long as the embodiment is modified in various ways that are conceivable by a person skilled in the art, the modifications may be included within the scope of the present invention.

[0085] For example, in the above embodiment, the vehicle azimuth angle estimation method includes the acquisition step S10 and the update step S40, but the vehicle azimuth angle estimation method does not have to include these steps.

[0086] In the above embodiment, the vehicle azimuth estimation device 100 includes the update unit 40, the driving control unit 70, and the storage unit 80, but the vehicle azimuth estimation device 100 does not have to include these components. The vehicle azimuth estimation device 100 may also include an acquisition unit 10.

[0087] In addition, in the above embodiment, an example is shown in which the autonomous moving body V10 is used as an example of a moving body, but the moving body is not limited to the autonomous moving body V10. It is sufficient if the moving body is an object that can move on a road and whose azimuth angle can be estimated by the vehicle azimuth angle estimation device 100.

[0088] In addition, in the vehicle azimuth estimation method according to the present embodiment, information acquired from a source other than the acquisition unit 10 may be used. For example, the vehicle azimuth estimation device 100 may acquire information by communicating with an external device.

[0089] The following embodiments may also be included within the scope of one or more aspects of the present invention.

[0090] (1) Some of the components included in the vehicle azimuth estimation device 100 may be a computer system including a microprocessor, a ROM, a RAM, a hard disk unit, a display unit, a keyboard, a mouse, etc. A computer program is stored in the RAM or the hard disk unit. The microprocessor operates according to the computer program to achieve its functions. Here, the computer program is configured by combining a plurality of instruction codes indicating commands for a computer to achieve a predetermined function.

[0091] (2) Some of the components included in the vehicle azimuth estimation device 100 may be configured as one system LSI (Large Scale Integration). The system LSI is an ultra-multifunctional LSI manufactured by integrating multiple components on a single chip, and specifically, is a computer system including a microprocessor, a ROM, a RAM, etc. A computer program is stored in the RAM. The system LSI achieves its functions by the microprocessor operating in accordance with the computer program.

[0092] (3) Some of the components included in the vehicle azimuth estimation device 100 may be configured as an IC card or a standalone module that is detachable from each device. The IC card or the module is a computer system configured with a microprocessor, ROM, RAM, etc. The IC card or the module may include the ultra-multifunction LSI. The microprocessor operates according to a computer program, causing the IC card or the module to achieve its functions. The IC card or the module may be tamper-resistant.

[0093] (4) Furthermore, some of the components included in the vehicle azimuth estimation device 100 may be the computer program or the digital signal recorded on a computer-readable recording medium, such as a flexible disk, a hard disk, a CD-ROM, an MO, a DVD, a DVD-ROM, a DVD-RAM, a BD (Blu-ray (registered trademark) Disc), a semiconductor memory, etc. Also, the components may be the digital signal recorded on such a recording medium.

[0094] In addition, some of the components included in the vehicle azimuth angle estimation device 100 may transmit the computer program or the digital signal via a telecommunication line, a wireless or wired communication line, a network such as the Internet, data broadcasting, or the like.

[0095] (5) The present invention may be the vehicle azimuth angle estimation method described above. It may also be a computer program for causing a computer to realize the vehicle azimuth angle estimation method, or a digital signal comprising the computer program. Furthermore, the present invention may be realized as a non-transitory computer-readable recording medium, such as a CD-ROM, on which the computer program is recorded.

[0096] (6) The present invention may also be a computer system having a microprocessor and a memory, the memory storing the above-mentioned computer program, and the microprocessor operating in accordance with the computer program.

[0097] (7) The program or the digital signal may also be implemented by another independent computer system by recording it on the recording medium and transferring it, or by transferring the program or the digital signal via the network, etc.

[0098] (8) The above-described embodiments and modifications may be combined with each other.

[0099] (Additional Note) Furthermore, based on the above description, the following techniques are disclosed.

[0100] (Technology 1) A vehicle azimuth angle estimation method for estimating a longitudinal azimuth angle of a host vehicle, the vehicle azimuth angle estimation method including: a correlation distribution calculation step of calculating a correlation distribution between a first Hough spectrum obtained by subjecting an observation image, which is an image of a road along which the host vehicle travels, to a Hough transform, and a second Hough spectrum obtained by subjecting a map image corresponding to the observation image to a Hough transform; and a probability distribution calculation step of calculating, based on the correlation distribution, an observation probability distribution, which is a probability distribution of an azimuth angle error of the observation image with respect to the map image.

[0101] (Technology 2) The vehicle azimuth angle estimation method according to Technology 1, further comprising an updating step of updating a priori probability distribution, which is a probability distribution of the azimuth angle error calculated based on the observation probability distribution calculated in the past, to a posterior probability distribution based on the latest observation probability distribution.

[0102] (Technology 3) The vehicle azimuth angle estimation method according to Technology 2, in which BBF is used in the updating step.

[0103] (Technique 4) The vehicle azimuth angle estimation method according to Technique 2 or 3, wherein in the updating step, the observation probability distribution is transformed so that the probability of the observation probability distribution approaches 0.5 as the variance of the observation probability distribution increases.

[0104] (Technology 5) The vehicle azimuth angle estimation method according to any one of Techniques 1 to 4, wherein in the correlation distribution calculation step, the divided observed image is subjected to a Hough transform.

[0105] (Technology 6) The vehicle azimuth angle estimation method according to Technology 5, wherein in the correlation distribution calculation step, the observed image is divided into areas having dimensions determined based on a radius of curvature of a road on which the host vehicle is traveling.

[0106] (Technology 7) A program for causing a computer to execute the vehicle azimuth angle estimation method according to any one of Technologies 1 to 6.

[0107] (Technology 8) A vehicle azimuth angle estimation device that estimates an azimuth angle in a forward / rearward direction of a vehicle, the vehicle azimuth angle estimation device including: a correlation distribution calculation unit that calculates a correlation distribution between a first Hough spectrum obtained by subjecting an observation image acquired in the vehicle to a Hough transform and a second Hough spectrum obtained by subjecting a map image to a Hough transform; and a probability distribution calculation unit that calculates an observation probability distribution, which is a probability distribution of an azimuth angle error of the observation image with respect to the map image, based on the correlation distribution.

[0108] (Technology 9) An autonomous moving body comprising the vehicle azimuth angle estimation device according to Technology 8 and an acquisition unit that acquires the observed image. [Industrial Applicability]

[0109] A vehicle azimuth angle estimation method according to an aspect of the present invention can be applied to a vehicle for autonomous driving, for example. [Explanation of symbols]

[0110] 10 Acquisition Department 20 Correlation distribution calculation section 30 Probability distribution calculation unit 40 Update section 70 Driving control unit 80 Storage section 100 Vehicle azimuth angle estimation device V10 autonomous vehicle

Claims

1. A vehicle azimuth angle estimation method for estimating a forward / rearward azimuth angle of a host vehicle, comprising: a correlation distribution calculation step of calculating a correlation distribution between a first Hough spectrum obtained by subjecting an observation image, which is an image of a road along which the host vehicle travels, to a Hough transform and a second Hough spectrum obtained by subjecting a map image corresponding to the observation image to a Hough transform; and calculating an observation probability distribution, which is a probability distribution of an azimuth angle error of the observation image with respect to the map image, based on the correlation distribution. Vehicle azimuth angle estimation method.

2. The method further includes an updating step of updating a prior probability distribution, which is a probability distribution of the azimuth angle error calculated based on the observation probability distribution calculated in the past, to a posterior probability distribution based on the latest observation probability distribution. The method for estimating a vehicle azimuth angle according to claim 1 .

3. In the update step, a Binary Bayes Filter (BBF) is used. The method for estimating a vehicle azimuth angle according to claim 2.

4. In the updating step, the observation probability distribution is transformed so that the probability of the observation probability distribution approaches 0.5 as the variance of the observation probability distribution increases. The method for estimating a vehicle azimuth angle according to claim 2 or 3.

5. In the correlation distribution calculation step, the divided observed image is subjected to a Hough transform. The vehicle azimuth angle estimation method according to any one of claims 1 to 3.

6. In the correlation distribution calculation step, the observed image is divided into dimensions determined based on a radius of curvature of a road on which the host vehicle is traveling. The method for estimating a vehicle azimuth angle according to claim 5.

7. A method for estimating a vehicle azimuth angle according to any one of claims 1 to 3, program.

8. A vehicle azimuth angle estimation device that estimates a forward / rearward azimuth angle of a host vehicle, a correlation distribution calculation unit that calculates a correlation distribution between a first Hough spectrum obtained by subjecting an observation image acquired by the host vehicle to a Hough transform and a second Hough spectrum obtained by subjecting a map image to a Hough transform; a probability distribution calculation unit that calculates an observation probability distribution, which is a probability distribution of an azimuth angle error of the observation image with respect to the map image, based on the correlation distribution. Vehicle heading angle estimation device.

9. The vehicle azimuth angle estimation device according to claim 8 ; An acquisition unit for acquiring the observation image. Autonomous mobile body.

Citation Information

Patent Citations

  • Position calculation device

    JP6669059B2