Three-dimensional map update system and three-dimensional map update method

Abstract

To provide a three-dimensional map update system that can reduce cost required for updating three-dimensional map data corresponding to a city level space.SOLUTION: A three-dimensional map update system comprises: point group map creation means 10 that creates a creation point group map that is a point group map on the basis of a three-dimensional model representing a target space; a log point group map creation means 12 that acquires log data when an automatic driving vehicle 22 travels in a travel permitted area in the target space on the basis of the creation point group map, and creates a log point group map that is the point group map on the basis of the log data; difference detection means 14 that compares the creation point group map and the log point group map and detects a difference between the maps; and update means 16 that updates the three-dimensional model on the basis of a result of the detection made by the difference detection means 14. The point group map creation means 10 creates the creation point group map on the basis of the three-dimensional model updated by the update means 16.SELECTED DRAWING: Figure 1

Description

[Technical field]

[0001] The present invention relates to a 3D map updating system and a 3D map updating method that utilize spatial point cloud data. [Background technology]

[0002] In recent years, various self-location estimation technologies have been developed as technologies related to autonomous vehicles. In particular, in order to stabilize autonomous driving on public roads, it is considered better to estimate self-location by matching the illumination point cloud of a LiDAR (Light Detection and Ranging) sensor with a map that converts the space into point cloud data (hereinafter referred to as a point cloud map), since it is difficult to ensure a certain level of accuracy in urban areas and other areas using self-location estimation only with GPS and cameras due to the characteristics of the sensors.

[0003] A conventional method for creating point cloud maps is to use a dedicated measurement vehicle. However, this method has high operating costs, making it difficult to update data frequently. There is also a method for converting 3D (three-dimensional) data that represents the shape of a building into a point cloud and map, but this requires that the original 3D data be synchronized with the real space, making it difficult to apply in areas where incomplete buildings exist, such as areas under construction. Furthermore, in areas under construction, the spatial configuration is constantly changing as the construction progresses, so the above point cloud map creation method cannot keep up with the frequency of data updates, and self-location estimation may no longer function.

[0004] Meanwhile, a technology is known that detects planes from the density of spatial point cloud data and automatically creates 3D drawings of the space, including the interior of a building (for example, see Patent Document 1). This technology has the following characteristics: (1) it is specialized or limited to creating drawings of office spaces, (2) it does not capture changes in conditions because new data is created each time, and (3) the entire space is calculated each time a drawing is created. [Prior art documents] [Patent documents]

[0005] [Patent Document 1] Patent Publication No. 2022-101927 Summary of the Invention [Problem to be solved by the invention]

[0006] However, the technology disclosed in the above Patent Document 1 is difficult to use for periodic update maintenance of 3D map data corresponding to city-level space due to the target space and calculation cost. For this reason, there has been a demand for technology that can reduce the cost of updating 3D map data corresponding to city-level space.

[0007] The present invention has been made in consideration of the above, and aims to provide a 3D map updating system and a 3D map updating method that can reduce the cost of updating 3D map data corresponding to city-level space. [Means for solving the problem]

[0008] In order to solve the above-mentioned problems and achieve the object, the 3D map updating system of the present invention is a 3D map updating system that updates a 3D map, which is a point cloud map that represents the 3D terrain of a target space as a 3D point cloud, and is equipped with a point cloud map creation means that creates a created point cloud map, which is the point cloud map, based on a 3D model that represents the target space, a log point cloud map generation means that acquires log data when an autonomous vehicle drives within a drivable area in the target space based on the created point cloud map and generates a log point cloud map, which is the point cloud map, based on the log data, a difference detection means that compares the created point cloud map with the log point cloud map and detects the difference therebetween, and an update means that updates the 3D model based on the detection result by the difference detection means, and is characterized in that the point cloud map creation means creates the created point cloud map based on the 3D model updated by the update means.

[0009] Furthermore, a 3D map updating method according to the present invention is a 3D map updating method for updating a 3D map, which is a point cloud map that represents the 3D terrain of a target space as a 3D point cloud, and includes a point cloud map creation step of creating a created point cloud map, which is the point cloud map, based on a 3D model that represents the target space; a log point cloud map generation step of acquiring log data when an autonomous vehicle drives within a drivable area in the target space based on the created point cloud map, and generating a log point cloud map, which is the point cloud map, based on the log data; a difference detection step of comparing the created point cloud map with the log point cloud map and detecting a difference therebetween; and an update step of updating the 3D model based on the detection result by the difference detection step, and is characterized in that the point cloud map creation step creates the created point cloud map based on the 3D model updated by the update step. Effect of the Invention

[0010] According to the 3D map updating system of the present invention, there is provided a 3D map updating system for updating a 3D map, which is a point cloud map that represents the 3D terrain of a target space as a 3D point cloud, and the system includes a point cloud map creation means for creating a created point cloud map, which is the point cloud map, based on a 3D model that represents the target space, a log point cloud map generation means for acquiring log data when an autonomous vehicle drives within a drivable area in the target space based on the created point cloud map, and generating a log point cloud map, which is the point cloud map, based on the log data, a difference detection means for comparing the created point cloud map with the log point cloud map and detecting the difference therebetween, and an update means for updating the 3D model based on the detection result by the difference detection means, and the point cloud map creation means creates the created point cloud map based on the 3D model updated by the update means, thereby achieving the effect of reducing the cost of updating 3D map data corresponding to city-level space.

[0011] Furthermore, according to the 3D map updating method of the present invention, there is provided a 3D map updating method for updating a 3D map, which is a point cloud map that represents the 3D topography of a target space as a 3D point cloud, and the method includes a point cloud map creation step of creating a created point cloud map, which is the point cloud map, based on a 3D model that represents the target space; a log point cloud map generation step of acquiring log data when an autonomous vehicle travels within a drivable area in the target space based on the created point cloud map, and generating a log point cloud map, which is the point cloud map, based on the log data; a difference detection step of comparing the created point cloud map with the log point cloud map and detecting a difference therebetween; and an update step of updating the 3D model based on the detection result by the difference detection step. Since the point cloud map creation step creates the created point cloud map based on the 3D model updated by the update step, it has the effect of reducing the cost of updating 3D map data corresponding to city-level space. [Brief description of the drawings]

[0012] [Figure 1] FIG. 1 is a schematic diagram showing the configuration of an embodiment of a 3D map updating system according to the present invention. [Diagram 2] FIG. 2 is a schematic procedure diagram showing an embodiment of a 3D map updating method according to the present invention. [Diagram 3] FIG. 3 is an explanatory diagram of difference detection of point clouds. [Figure 4] FIG. 4 is a diagram showing an example of a pattern that is erroneously recognized as a new pattern. [Diagram 5] FIG. 5 is an explanatory diagram of the first step of 3D modeling. [Figure 6] FIG. 6 is an explanatory diagram of the second step of 3D modeling. [Figure 7] FIG. 7 is an explanatory diagram of geometric model matching. DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS

[0013] DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS Hereinafter, preferred embodiments of a 3D map updating system and a 3D map updating method according to the present invention will be described in detail with reference to the accompanying drawings. However, the present invention is not limited to these preferred embodiments.

[0014] As shown in Fig. 1, a 3D map updating system 100 according to an embodiment of the present invention is a system for updating a 3D map showing the topography of a city-level space S (target space) of a certain region, and includes a computer equipped with a point cloud map creating means 10, a log point cloud map generating means 12, a difference detecting means 14, an updating means 16, a storage means 18, and a control means 20 for controlling these. A 3D map is a map (point cloud map) in which a 3D topography is shown by 3D point cloud data. Fig. 2 shows an outline of the processing procedure of a map updating method using the 3D map updating system 100.

[0015] The point cloud map creation means 10 creates a created point cloud map based on a 3D model that represents a three-dimensional shape of the space S. The point cloud map creation means 10 also creates a created point cloud map based on a 3D model updated by an update means 16, which will be described later. The 3D model is a model in which a three-dimensional shape is polygonized using CAD data or the like, and is created in advance and registered in the storage means 18. For generating the created point cloud map, for example, a method of converting data constituting the 3D model into point cloud data can be used. The created created point cloud map is registered in the storage means 18 and is mounted on an autonomous vehicle 22, which will be described later.

[0016] The log point cloud map generating means 12 acquires log data when the autonomous vehicle 22 travels in a drivable area in the space S based on the created point cloud map, and generates a log point cloud map (point cloud map) based on the log data. The log data includes point cloud data acquired when traveling in the real space S. The generated log point cloud map is registered in the storage means 18.

[0017] The autonomous vehicle 22 is a vehicle capable of autonomous driving, and includes a laser sensor 24, a satellite positioning sensor 26, and a driving control device 28. The autonomous vehicle 22 may include an imaging device such as a camera that captures images of the surroundings.

[0018] The laser sensor 24 emits a laser into the surroundings and detects the laser reflected from surrounding objects, thereby acquiring point cloud data (three-dimensional coordinates) indicating the position of the object from which the laser is reflected. The acquired point cloud data is sent to the cruise control device 28 as log data. The laser sensor 24 uses a LiDAR that irradiates a laser within a predetermined angular range in the horizontal and vertical directions to acquire point cloud data with high accuracy. By acquiring point cloud data with the LiDAR, the three-dimensional shape of the terrain around the autonomous vehicle 22 can be detected.

[0019] The satellite positioning sensor 26 measures the self-position of the autonomous vehicle 22 using satellite radio waves such as GPS. The measured self-position data is sent to the cruise control device 28 as log data. In order to measure the self-position with high accuracy, the measured self-position may be corrected based on the result of scan matching between the point cloud data newly obtained by the laser sensor 24 while traveling in the space S and the created point cloud map.

[0020] The driving control device 28 is configured by a computer and includes a driving route setting unit 30, a storage unit 32, a control unit 34, and the like. The driving route setting unit 30 is a unit that sets a route along which the autonomous vehicle 22 autonomously drives. The route is set based on a drivable area in a created point cloud map stored in the storage unit 32. The drivable area is an area such as a road in which the vehicle can drive. The storage unit 32 stores map data such as the created point cloud map, data necessary for autonomous driving, and data (log data) acquired by the laser sensor 24, the satellite positioning sensor 26, and the like. The control unit 34 controls the traveling direction and speed of the autonomous vehicle 22 based on the route set by the driving route setting unit 30. This allows the autonomous vehicle 22 to drive within the space S along the set route. During driving, log data is acquired and stored in the storage unit 32. The log data and the like stored in the storage unit 32 are stored in the storage unit 18 via the control unit 20. In addition, instead of being installed in the autonomous vehicle 22, the driving control device 28 may be installed in a server or the like in a remote control room, and the operation of the autonomous vehicle 22 may be controlled from the server or the like via a network line.

[0021] The difference detection means 14 compares the created point cloud map stored in the storage means 18 with the log point cloud map to detect the difference. Specifically, the difference is detected by overlaying the point cloud maps in order to extract things that exist in the created point cloud map but not in the actual space S (hereinafter referred to as removal patterns) and things that do not exist in the created point cloud map but exist in the actual space S (hereinafter referred to as new construction patterns). Examples of removal patterns include removal of buildings and temporary structures through demolition, pruning of trees, etc. Examples of new construction patterns include newly constructed buildings, construction zones, etc.

[0022] In the case of the removal pattern, as shown in Figure 3 (1), the log point cloud map is overlaid with the created point cloud map as the base. At that time, as shown in Figures 3 (2) and (3), the shortest distance from each point on the created point cloud map to each point on the log point cloud map is calculated and a value is assigned to each point on the created point cloud map. In addition, in the case of the new establishment pattern, the log point cloud map is used as the base and the created point cloud map is overlaid to detect differences, in contrast to the removal pattern.

[0023] The update means 16 updates the 3D model based on the detection result by the difference detection means 14. This update means has a map update necessity determination unit 36 ​​and a point cloud data 3D modeling unit 38. These will be described below.

[0024] The map update necessity determination unit 36 ​​determines whether or not a map needs to be updated based on the value of the shortest distance when the difference detection means 14 detects the difference. As a basic determination criterion, the map is updated when the calculated shortest distance is large. For example, a threshold value of 0.4 m or more is set to perform the determination. In this case, it is desirable to perform the determination of whether or not the map needs to be updated in units of polygon data, such as one building or one plant, for the convenience of finally feeding back to the 3D model as shown in FIG. 2. If the determination criterion is only the threshold value of each point, rather than the unit of polygon data, all areas that cannot be captured by the log point cloud map, such as the high-rise part of a building as shown in FIG. 3, will be deleted from the map, and the original 3D model may have a shape different from reality. In addition, a pattern that is erroneously recognized as a new pattern is a pattern in which a temporary obstacle, such as a vehicle on a lane or a vehicle parked on the roadside, exists in the log point cloud map. In the example of FIG. 4, there is a risk of erroneously recognizing a vehicle as a newly built building. Therefore, in order to correctly determine whether or not to update the map, it is necessary to add attributes to the point cloud map and cluster the point cloud in units of polygon data.

[0025] One way to add attributes to the created point cloud map is to inherit the attributes of the original 3D data. This can be achieved by creating a text file that stores information that corresponds one-to-one to each point when converting a 3D building model into a point cloud, or by assigning attributes to the color information of the point cloud itself.

[0026] As a method for assigning attributes to a log point cloud map, a method can be used in which geometric clustering is performed using the distance between adjacent point clouds as a threshold, and attributes are estimated from the point cloud shape of the clustering result. For example, point cloud patterns of multiple vehicle shapes (sedans, SUVs, trucks, etc.) are created in advance as verification data. After that, the clustering result of the log point cloud map is compared with the verification data, and the optimal attribute can be assigned based on the matching rate.

[0027] In addition, attribute information obtained by object detection technology (such as YOLO) using image data may be added as necessary. In that case, images of the features that are candidates for the update target are required, so information from cameras (for sensing, drive recorders, etc.) mounted on the autonomous vehicle 22 and surveillance cameras on streets and premises can be used.

[0028] The point cloud data 3D modeling unit 38 performs processing to convert the detected point cloud data difference into a 3D model when it is determined that there is a feature to be added to the base 3D model as a result of the determination by the map update necessity determination unit 36. The task of converting the point cloud data difference into a 3D model can be performed in the following two steps (first and second steps).

[0029] First, in the first step, as shown in FIG. 5, the point cloud data that has been differentially detected from the feature that needs to be added is meshed and converted into a 3D model (strictly speaking, a mesh (polygon)). For the conversion, known techniques such as the Ball pivoting Algorithm and Poisson surface reconstruction can be used. If a precise 3D model is generated, it is adopted as the 3D model for the additional feature. However, simply applying the above-mentioned conversion technique may result in missing parts in the mesh, resulting in an incomplete 3D model. In such cases, the process proceeds to the next step, the second step.

[0030] In the second step, the incomplete 3D model generated in the first step is compared with the geometric model list and complemented with a geometrically corrected 3D model, as shown in Fig. 6. The geometric model list is a data set of geometric 3D models such as spheres, cones, and cylinders stored in the system. The corrected 3D model is a 3D model that is created by dividing the incomplete 3D model into multiple blocks as cross sections at points where the change in the maximum radius when viewed vertically is greater than a threshold, and then selecting and synthesizing geometric models with similar shapes for each block, as shown in Fig. 7.

[0031] Regarding the method of determining the threshold for the change in maximum radius, the threshold can be set to a small value in order to generate a more precise 3D model. The upper limit of the threshold varies depending on the accuracy required for the point cloud map, but basically it should be about the same as the accuracy required when using it as point cloud data. For example, the point cloud data used to estimate the self-position in an autonomous driving system uses downsampled point cloud scan data obtained from a dedicated vehicle. In this case, the point cloud is thinned out so that the intervals are approximately 30 cm in the horizontal direction, so the threshold can also be set to 30 cm.

[0032] The difference between the point cloud data thus 3D modeled is added to the original 3D model. This updates the 3D model. The updated 3D model is registered in the storage means 18.

[0033] The storage means 18 stores various data such as 3D models before and after the update, created point cloud maps, logged point cloud maps, and data acquired by the autonomously driven vehicle 22.

[0034] The operation and function of the above configuration will now be described. As shown in Fig. 2, an autonomous vehicle 22 performs autonomous driving using a created point cloud map created from a 3D model of space S, and a log point cloud map is generated from the log data at that time. The difference between the log point cloud map and the created point cloud map is detected, and feedback is provided based on this detection result to update the 3D model.

[0035] In this way, by extracting the time-dependent changes in the city-level space S during daily driving and updating the 3D model, regular management and maintenance of the 3D model can be efficiently performed. By creating a point cloud map from this 3D model, a map environment that constantly reproduces the real space S can be developed.

[0036] According to this embodiment, since log data of vehicles that are actually running, rather than dedicated vehicles, is used, it is possible to reduce the measurement work cost required for updating 3D map data corresponding to urban-level space. In addition, since it is possible to respond to frequent environmental changes, maintainability is improved. It is possible to update to a map that reproduces the current situation without waiting for administrative authorities to update the map.

[0037] As described above, the 3D map updating system of the present invention is a 3D map updating system that updates a 3D map, which is a point cloud map that represents the 3D topography of a target space as a 3D point cloud, and includes a point cloud map creation means that creates a created point cloud map, which is the point cloud map, based on a 3D model that represents the target space, a log point cloud map generation means that acquires log data when an autonomous vehicle drives within a drivable area in the target space based on the created point cloud map, and generates a log point cloud map, which is the point cloud map, based on the log data, a difference detection means that compares the created point cloud map with the log point cloud map and detects the difference therebetween, and an update means that updates the 3D model based on the detection result by the difference detection means, and the point cloud map creation means creates the created point cloud map based on the 3D model updated by the update means, thereby reducing the cost of updating 3D map data corresponding to city-level space.

[0038] Furthermore, according to a 3D map updating method of the present invention, there is provided a 3D map updating method for updating a 3D map, which is a point cloud map that represents the 3D topography of a target space as a 3D point cloud, comprising: a point cloud map creation step of creating a created point cloud map, which is the point cloud map, based on a 3D model that represents the target space; a log point cloud map generation step of acquiring log data when an autonomous vehicle drives within a drivable area in the target space based on the created point cloud map, and generating a log point cloud map, which is the point cloud map, based on the log data; a difference detection step of comparing the created point cloud map with the log point cloud map and detecting a difference therebetween; and an update step of updating the 3D model based on the detection result by the difference detection step, wherein the point cloud map creation step creates the created point cloud map based on the 3D model updated by the update step, thereby reducing the cost of updating 3D map data corresponding to city-level space.

[0039] The "Sustainable Development Goals (SDGs)" are among the 17 international goals adopted at the United Nations Summit in September 2015. The 3D map updating system and 3D map updating method according to the present embodiment can contribute to the achievement of, for example, "11. Sustainable cities and towns" among the 17 SDGs. [Industrial Applicability]

[0040] As described above, the 3D map updating system and 3D map updating method of the present invention are useful for updating 3D maps installed in self-driving vehicles, etc., and are particularly suitable for reducing the costs associated with updating 3D map data corresponding to city-level spaces. [Explanation of symbols]

[0041] 10 Point Cloud Mapping Methods 12 Log point cloud map generation method 14 Difference detection means 16 Update method 18 Memory means 20 Control Means 22 Self-driving vehicles 24 Laser sensor 26 Satellite positioning sensor 28 Driving control device 30 Travel route setting unit 32 Storage section 34 Control section 36 Map update necessity determination unit 38 3D modeling of point cloud data 100 3D map update system S space (target space)

Claims

1. A three-dimensional map updating system for updating a three-dimensional map that is a point cloud map that represents a three-dimensional topography of a target space as a three-dimensional point cloud, comprising: a point cloud map creation means for creating a created point cloud map, which is the point cloud map, based on a three-dimensional model representing the target space; a log point cloud map generating means for acquiring log data when an autonomous vehicle travels in a drivable area in the target space based on the created point cloud map, and generating a log point cloud map, which is the point cloud map, based on the log data; a difference detection means for comparing the created point cloud map with the log point cloud map and detecting a difference therebetween; an update means for updating the three-dimensional model based on a detection result by the difference detection means, The three-dimensional map updating system according to claim 1, wherein the point cloud map creating means creates the created point cloud map based on the three-dimensional model updated by the updating means.

2. A three-dimensional map updating method for updating a three-dimensional map that is a point cloud map that represents a three-dimensional topography of a target space as a three-dimensional point cloud, comprising the steps of: a point cloud map creation step of creating a created point cloud map, which is the point cloud map, based on a three-dimensional model representing the target space; a log point cloud map generation step of acquiring log data when an autonomous vehicle travels in a drivable area in the target space based on the created point cloud map, and generating a log point cloud map, which is the point cloud map, based on the log data; a difference detection step of comparing the created point cloud map with the log point cloud map to detect a difference therebetween; an updating step of updating the three-dimensional model based on a detection result by the difference detection step, The 3D map updating method, wherein the point cloud map creating step creates the created point cloud map based on the 3D model updated in the updating step.

Images