Position estimation device, and position estimation method

Abstract

To provide a position estimation device capable of self-location estimation accurately with fewer errors, and a position estimation method.SOLUTION: The position estimation device includes: an identification circuit that identifies a stationary model representing stationary objects around a vehicle out of multiple models that model objects based on the stationary model and a point group corresponding to the stationary objects detected at each time by the point group detection device; and a position estimation circuit that detects the relative position of the vehicle relative to stationary objects at each time and estimates the position of the vehicle at each time based on the relative positional relationship at each time.SELECTED DRAWING: Figure 3

Description

Technical Field

[0001] The present disclosure relates to a position estimation device and a position estimation method.

Background Art

[0002] In existing technologies, in order for a vehicle to park without colliding, it accurately determines where the available parking space is. For this purpose, a surrounding map in which the location of three-dimensional objects is known is used. To generate such a map, a technique called SLAM (Simultaneous Localization and Mapping) that overlaps Lidar (Light Detection and Ranging) point clouds for each frame is known.

[0003] This SLAM technique estimates the amount of vehicle movement by making the features of stationary objects accurately overlap using the Lidar point cloud for each frame during vehicle movement. Then, the vehicle generates an expected path between the target parking position and the vehicle position on the surrounding map and enters the warehouse along that path. For example, Patent Document 1 discloses a technique in which a vehicle determines the error in the position of an object detected by the SLAM technique and travels in consideration of the error.

[0004] Also, the currently mainstream mechanical Lidar scans a beam to acquire surrounding point cloud data, but what is expected as a future sensor device is Flash-Lidar. Flash-Lidar acquires point cloud data at once like a camera without scanning a beam, as compared with mechanical Lidar.

Prior Art Documents

Patent Documents

[0005]

Patent Document 1

Summary of the Invention

Problems to be Solved by the Invention

[0006] As described above, in SLAM, Lidar point clouds for each frame are superimposed. Specifically, if it is determined that the point cloud of the latest frame output from the Lidar is the same as the point cloud in the map generated so far, the former point cloud is further superimposed on the latter point cloud to update the map. At this time, the vehicle estimates its own position assuming that the amount of movement of the vehicle for exactly superimposing the two point clouds is the amount of movement of the vehicle in the time for one frame.

[0007] In this case, even if the error in the amount of movement of the vehicle in the time for one frame is small, it will become large if it accumulates. For example, when the vehicle travels on a loop route, it is known that if the error accumulates, a so-called loop closure problem occurs where the vehicle cannot return to the starting point and the loop cannot be closed.

[0008] The technology of SLAM has a weakness that once a small error occurs in self-position estimation, the vehicle cannot return to the correct position because it superimposes a new point cloud assuming that the map superimposed so far is correct.

[0009] In addition, mechanical Lidar is vulnerable to vibration due to the rotation mechanism for scanning the beam, and it is difficult to ensure reliability at the in-vehicle level. On the other hand, Flash-Lidar can measure the relative positional relationship to surrounding objects, but it is difficult to obtain its own position information by itself. And without accurate own position information, it becomes difficult to create a surrounding map by superimposing point clouds between frames.

[0010] Furthermore, in Flash-Lidar, although device development for the light-receiving part of the beam is actively underway, the ranging accuracy is still insufficient. This is because in mechanical Lidar, the laser power per pixel can be increased by narrowing the beam, while in Flash-Lidar, since the laser power is radiated to all pixels at once, the laser power per pixel decreases, and the SNR (Signal to Noise Ratio) between the received beam and sunlight deteriorates, resulting in an increase in noise in the ranging result.

[0011] Non-limiting embodiments of the present disclosure contribute to providing a position estimation device and a position estimation method capable of accurate self-position estimation with little error.

Means for Solving the Problems

[0012] Therefore, one aspect of the position estimation device according to the present disclosure includes a specific circuit that specifies a stationary object model representing a stationary object existing around the vehicle from among a plurality of models that model an object, the stationary object model, and a point cloud corresponding to the stationary object detected at each time by a point cloud detection device, detects the relative positional relationship of the vehicle with respect to the stationary object at each time based on these, and a position estimation circuit that estimates the position of the vehicle at each time based on the relative positional relationship at each time.

[0013] Also, one aspect of the position estimation method according to the present disclosure specifies a stationary object model representing a stationary object existing around the vehicle from among a plurality of models that model an object, detects the relative positional relationship of the vehicle with respect to the stationary object at each time based on the stationary object model and the point cloud corresponding to the stationary object detected at each time by a point cloud detection device, and estimates the position of the vehicle at each time based on the relative positional relationship at each time.

Advantages of the Invention

[0014] According to the present disclosure, accurate self-position estimation with little error can be performed.

Brief Description of the Drawings

[0015]

Figure 1

Figure 2

Figure 3

Figure 4

Figure 5

Figure 6

Figure 7

Figure 8

Figure 9

Embodiments for Carrying Out the Invention

[0016] Hereinafter, embodiments of the present disclosure will be described with reference to the drawings. Note that each of the embodiments described below shows a specific example of the present disclosure. Therefore, each component shown in the following embodiments, the arrangement position and connection form of each component, and the order of each step and each step are examples and are not intended to limit the present disclosure. In addition, among the components in the following embodiments, the components not described in the independent claims are described as optional components.

[0017] In addition, each figure is a schematic diagram and is not necessarily drawn precisely. In each figure, the same reference numerals are given to substantially the same configurations, and duplicate explanations are omitted or simplified.

[0018] The position estimation device according to the present disclosure shown in FIG. 1 estimates the position of a vehicle based on the position of a stationary anchor vehicle.

[0019] As shown in FIG. 1, the position estimation device identifies the vehicle type of the anchor vehicle from Image 1 obtained by photographing the anchor vehicle with a camera (image capturing device), and identifies a three-dimensional model 2 representing the shape of the anchor vehicle.

[0020] On the other hand, a Flash-Lidar (point cloud detection device) 3 measures the distance to each point on the surface of the anchor vehicle using a reflected beam obtained by reflecting the emitted beam by the anchor vehicle, and generates point cloud data 4 obtained as a result. Then, the position estimation device estimates the position of the vehicle by matching the positions of the three-dimensional model 2 and the point cloud included in the point cloud data 4.

[0021] For example, the position estimation device detects the position of the surface of the three-dimensional model 2 where the three-dimensional distance from the three-dimensional point cloud 4 is the smallest by using the three-dimensional least squares method, thereby detecting the point cloud on the surface of the three-dimensional model 2 corresponding to the three-dimensional point cloud 4.

[0022] Then, the position estimation device calculates the relative position of the vehicle with respect to the anchor vehicle at each time based on the point cloud on the surface of the three-dimensional model 2 and the point cloud corresponding to the anchor vehicle detected by the Flash-Lidar 3 at each time, and estimates the absolute position of the vehicle at each time based on the relative position at each time.

[0023] In this way, by calculating the relative position of the vehicle with respect to the anchor vehicle at each time and estimating the absolute position of the vehicle at each time based on the relative position at each time, it is possible to prevent the measurement error from accumulating like in the SLAM technology, and to accurately create the movement information of the vehicle and the surrounding map.

[0024] Also, when measuring the relative distance between the anchor vehicle and the vehicle, by using the point cloud on the surface of the 3D model 2 instead of the point cloud itself detected by Flash-Lidar3, the influence of the measurement noise of Flash-Lidar3 can be removed, and the shape information of the anchor vehicle without noise can be obtained.

[0025] In FIG. 2, the change in the positional relationship of the stationary anchor vehicle 11 (the position and angle of the anchor vehicle 11 as seen from the host vehicle 12) as viewed in the Flash-Lidar coordinate system can be converted into the movement information of the host vehicle 12.

[0026] For example, in the Flash-Lidar coordinate system of FIG. 2(a), the position of the anchor vehicle 11 when the time t changes from 1 to 4 can be converted into the movement information of the host vehicle 12 as shown in FIG. 2(b).

[0027] In FIG. 3, the position estimation device 20 includes an image capturing device 21, a 3D model database 22, a point cloud detection device 23, a model identification unit 24, and a position estimation unit 27.

[0028] The image capturing device 21 is a camera installed on the vehicle and captures an image of an object existing around the vehicle in real time.

[0029] The 3D model database 22 is a database that stores data of a 3D model obtained by modeling an anchor vehicle.

[0030] The point cloud detection device 23 irradiates an object existing around the vehicle with a beam, measures the distance to each point on the surface of the object using the reflected beam reflected from the object, and generates point cloud data 4 obtained as a result in real time. For example, the point cloud detection device 23 is a Flash-Lidar installed on the vehicle. Note that the point cloud detection device 23 may be another device capable of detecting a point cloud, such as a mechanical Lidar.

[0031] The model identification unit 24 identifies a 3D model representing an anchor vehicle existing around the vehicle from among a plurality of 3D models that model the vehicle. The model identification unit 24 includes a 3D model extraction unit 25 and a point cloud matching unit 26.

[0032] The 3D model extraction unit 25 extracts a 3D model corresponding to the image captured by the image capturing device 21 from the 3D model database 22 and estimates the type of the anchor vehicle.

[0033] The point cloud matching unit 26 performs a matching process to determine whether the 3D model extracted by the 3D model extraction unit 25 matches the point cloud detected by the point cloud detection device 23, and based on the result of the matching process, performs a type determination process to determine whether the type of the 3D model extracted by the 3D model extraction unit 25 is correct as the type of the anchor vehicle.

[0034] And when the point cloud matching unit 26 identifies that the type of the 3D model is correct as the type of the anchor vehicle, it outputs the position information of the 3D model in the Flash-Lidar coordinate system in which the position of the anchor vehicle changes with the host vehicle as the origin. This position information of the 3D model corresponds to information indicating the relative positional relationship of the vehicle with respect to the anchor vehicle at each time.

[0035] This position information of the 3D model is different from the point cloud detected by the point cloud detection device 23, and since there is no measurement noise in the point cloud, the shape of the anchor vehicle can be accurately identified, and the subsequent vehicle position estimation process can be performed with high accuracy.

[0036] The position estimation unit 27 detects the relative positional relationship of the vehicle with respect to the anchor vehicle at each time based on the 3D model of the anchor vehicle and the point cloud corresponding to the anchor vehicle detected by the point cloud detection device 23 at each time, and estimates the position of the vehicle at each time based on the relative positional relationship at each time. The position estimation unit 27 includes an object position identification unit 28 and a vehicle position identification unit 29.

[0037] The object position specifying unit 28 specifies the position of the anchor vehicle based on the surrounding map. The vehicle position estimating unit 29 estimates the position of the vehicle at each time based on the relative positional relationship of the vehicle with respect to the anchor vehicle detected by the point cloud matching unit 26, with the position of the specified anchor vehicle as the reference position. Then, the vehicle position estimating unit 29 outputs the estimated position of the vehicle as the self-position estimation result.

[0038] In FIG. 4, the image capturing device 21 acquires an image of the anchor vehicle 1 with a camera, and the three-dimensional model extraction unit 25 estimates the type of the vehicle from the image (step S1). Then, the three-dimensional model extraction unit 25 searches for the three-dimensional model 2 corresponding to the vehicle type from the three-dimensional model database 22 (step S2).

[0039] Subsequently, the point cloud detection device 23 extracts the three-dimensional point cloud 4 of the anchor vehicle 11 with the Flash-Lidar 3 (step S3). Then, the point cloud matching unit 26 matches the point cloud 4 of the anchor vehicle with the three-dimensional model 2 of the anchor vehicle (step S4).

[0040] Thereafter, the point cloud matching unit 26 determines whether the detection condition of the vehicle type is a difficult condition in which it is difficult to specify the vehicle type (step S5). The difficult condition will be described in detail later.

[0041] When the detection condition of the vehicle type is a difficult condition (step S5, YES), the point cloud matching unit 26 processes the three-dimensional point cloud 4 (step S6), and matches the three-dimensional point cloud 4 of the anchor vehicle 11 with the three-dimensional model 2 of the anchor vehicle 11 again (step S4).

[0042] When it is determined that the detection condition of the vehicle type is not a difficult condition (step S5, NO), the point cloud matching unit 26 determines whether the type of the three-dimensional model 2 is correct from the degree of matching between the point cloud 4 and the three-dimensional model 2 (step S7).

[0043] If the type of the three-dimensional model 2 is incorrect (step S7, NO), the point cloud matching unit 26 selects a three-dimensional model 2 of another vehicle type (step S8), and rematches the point cloud 4 and the three-dimensional model 2 (step S4).

[0044] If the type of the three-dimensional model 2 is correct (step S7, YES), the point cloud matching unit 26 outputs the position information of the three-dimensional model 2 in the Flash-Lidar coordinate system in which the position of the anchor vehicle 11 changes with the host vehicle as the origin (step S9).

[0045] Thereafter, the object position specifying unit 28 estimates the movement amount of the host vehicle from the position information of the three-dimensional model 2 (step S10). Then, the vehicle position specifying unit 29 estimates its own position from the movement amount of the host vehicle (step S11), and outputs the estimation result (step S12).

[0046] Hereinafter, the determination process of whether the type of the three-dimensional model performed in step S7 of FIG. 4 is correct will be described in detail. In the existing technology, there is a practical example of identifying the vehicle type (manufacturer, vehicle name, model) of a vehicle using a camera image. For vehicle type identification using a camera image, for example, machine learning for learning the feature amounts of the anchor vehicle is used.

[0047] However, although there are many feature amounts in the front part and the rear part of the anchor vehicle, there are few feature amounts in the side part of the anchor vehicle in the first place, and in a camera image in which the side is captured, the vehicle type may not be correctly determined.

[0048] The point cloud matching unit 26 performs matching processing of the point cloud that is the output of the three-dimensional model 2 and the Flash-Lidar 3. Based on the index of the distance from each point cloud to the surface of the three-dimensional model 2 in addition to the existing matching degree indexes thereof, it is determined whether the vehicle type is correct. For example, when the vehicle type is correct, the average value of the distances from each point cloud to the surface of the three-dimensional model 2 corresponds to the ranging error performance of the Flash-Lidar alone.

[0049] As described above, the distance from each point cloud shown in FIG. 5 to the surface of the three-dimensional model corresponds to the ranging error (noise) performance of the Flash-Lidar alone, and its probability density distribution becomes a normal distribution as shown in FIG. 6. In this case, if the average value of the ranging error is μ and the standard deviation is σ, 99.7% of the data will be included within the range of μ ± 3σ.

[0050] Therefore, when the data included outside the range of μ ± 3σ is greater than a predetermined threshold value of more than 0.3%, the point cloud matching unit 26 determines that the vehicle type is incorrect, with the average value of the distance from each point cloud to the surface of the three-dimensional model being μ and the standard deviation being σ. Here, this threshold value is set to a value considering the noise performance for each manufacturer of Flash-Lidar. Thereby, the point cloud matching unit 26 can correctly determine whether the vehicle type is correct.

[0051] In this way, when selecting the three-dimensional model of the anchor vehicle using the camera image under conditions where machine learning is difficult, even if a three-dimensional model of a different vehicle type is selected, it is possible to determine that it is a difference in vehicle type, and the matching process between the point cloud and the three-dimensional model can be repeated using a three-dimensional model of another vehicle type.

[0052] Next, the determination process as to whether the vehicle type detection condition performed in step S5 of FIG. 4 corresponds to a difficult condition will be described. Difficult conditions where it is difficult to detect the vehicle type include the following. For example, when there are obstacles in the parking area or pedestrians enter, a part of the anchor vehicle is hidden behind them, making it difficult to detect the anchor vehicle. Also, when the anchor vehicle is equipped with custom parts, or when it is nighttime or raining, making it difficult to obtain a clear camera image, or when the beam incident angle of Flash-Lidar3 is shallow.

[0053] As shown in FIG. 7, pedestrians 31 and obstacles enter the parking area, and a part of the anchor vehicle 11 is hidden behind them. The pedestrians 31 and the stationary anchor vehicle 11 can be seen overlapping in the camera image, but a gap can be confirmed when viewing the three-dimensional point cloud of Flash-Lidar3 in a bird's-eye view.

[0054] Therefore, the point cloud matching unit 26 clusters the point cloud with spatial information, separates the pedestrian 31 and the anchor vehicle, deletes the point cloud of the pedestrian 31 in the foreground, and then executes the matching process between the point cloud and the 3D model. As a result, it becomes possible to detect the vehicle type with higher accuracy.

[0055] In addition, the anchor vehicle may be equipped with custom parts. In this case, the point cloud matching unit 26 performs a matching process between the point cloud and the 3D model, calculates the distance between the point cloud and the 3D model, and detects the custom parts based on the distance. The stationary object anchor vehicle is divided into a plurality of cubes for evaluation.

[0056] For example, the point cloud matching unit 26 classifies the point cloud with a distance from the 3D model greater than or equal to a predetermined threshold and the point cloud with a distance from the 3D model less than the predetermined threshold. Then, the point cloud matching unit 26 clusters the 3D point cloud with spatial information and determines whether the 3D point cloud with a distance from the 3D model greater than or equal to the predetermined threshold exists at a spatially concentrated position.

[0057] When the point cloud exists at a spatially concentrated position, the point cloud matching unit 26 excludes the point cloud existing in this space and executes the matching process between the point cloud and the 3D model again. As a result, it becomes possible to detect the vehicle type with higher accuracy.

[0058] Note that the point cloud matching unit 26 may divide the 3D model into a plurality of cubes, extract the cubes containing a point cloud with a distance from the 3D model greater than or equal to the predetermined threshold at a predetermined ratio or more, and exclude the point cloud included in the cube and then execute the matching process between the point cloud and the 3D model again.

[0059] Note that the camera is weak in night-time shooting, but Flash-Lidar3 is not affected by object detection even at night. Also, for raindrops, for cameras inside and outside the vehicle, it becomes a factor that reduces the object detection rate, but Flash-Lidar3 has almost no effect on object detection if it is light rain. Therefore, even if the type of the anchor vehicle estimated from the camera image is incorrect, the final determination by Flash-Lidar3 reduces the likelihood of misidentifying the type.

[0060] Next, the case where the beam incident angle of Flash-Lidar3 is shallow will be described. As shown in Fig. 8(a), in Flash-Lidar3, when the incident angle of the beam emitted from Flash-Lidar3 to the anchor vehicle 11 is shallow, the ranging error of Flash-Lidar3 increases, and there is a possibility of misidentifying the vehicle type. For example, on the side surface or the bonnet part of the anchor vehicle 11, the beam incident angle becomes shallow. As shown in Fig. 8(b), when the beam incident angle is near 90 degrees, the ranging error becomes small.

[0061] On the other hand, the point cloud matching unit 26 performs three-dimensional polygonization processing for each frame on the three-dimensional point cloud obtained from Flash-Lidar3, and obtains the normal vector of the polygon model obtained as a result.

[0062] The point cloud matching unit 26 determines which point clouds have a beam incident angle shallower than a predetermined angle from the normal vector shown in Fig. 9 and the direction in which the anchor vehicle 11 is located, and after removing those point clouds, performs matching processing between the three-dimensional point cloud and the three-dimensional model. This reduces unnecessary errors in the matching process.

[0063] Note that the point cloud matching unit 26 may divide the three-dimensional polygon model into small areas and determine for each area whether the beam incident angle for the point clouds included in the area is shallower than a predetermined angle.

[0064] Furthermore, factors that make it difficult to identify the vehicle type include differences in the wheels due to differences in the grades of the anchor vehicles. In addition, since the shape of the anchor vehicle changes depending on whether the door mirror is open or closed, it is necessary to prepare a 3D model of the anchor vehicle with the door mirror open and a 3D model of the anchor vehicle with the door mirror closed, resulting in an increase in the number of 3D models.

[0065] Therefore, it is desirable to exclude the wheel and door mirror parts from the 3D model 2. In addition, since the glass part of the anchor vehicle causes the beam of the Flash-Lidar 3 to pass through and be distorted, it is also desirable to exclude it from the 3D model 2 in the same way.

[0066] As a result, these parts will be excluded from the target of the matching process, and it is possible to prevent errors from occurring in the matching process.

[0067] The embodiments have been described above, but the present disclosure is not limited to the above embodiments.

[0068] For example, in the above embodiment, the anchor vehicle is described as an example of a stationary object serving as a reference for identifying the position of the host vehicle. However, any stationary object from which a 3D model can be obtained may be used instead of the anchor vehicle.

[0069] In addition, part or all of the 3D model database 22, the model identification unit 24, and the position estimation unit 27 may be mounted on the vehicle, or may be mounted on an external device such as a cloud server device, and the vehicle may communicate to exhibit the functions described above.

[0070] In addition, in the above-described embodiment, each component may be realized by executing a software program suitable for each component. Each component may be realized by a program execution unit such as a CPU or a processor reading and executing a software program recorded on a recording medium such as a hard disk or a semiconductor memory.

[0071] Furthermore, the general or specific aspects of the present disclosure may be realized by an apparatus, a method, an integrated circuit, a computer program, or a recording medium such as a computer-readable CD-ROM. Further, it may be realized by any combination of an apparatus, a method, an integrated circuit, a computer program, and a recording medium.

[0072] Note that in the impact detection apparatus of the above-described embodiment, the notation "··· section" used for each component may be replaced with other notations such as "··· circuitry", "··· assembly", "··· device", "··· unit", or "··· module" as described above.

[0073] In addition, forms obtained by applying various modifications conceivable by those skilled in the art to each embodiment, or forms realized by arbitrarily combining the components and functions in each embodiment without departing from the spirit of the present disclosure are also included in the present disclosure.

Industrial Applicability

[0074] The present disclosure can be used for a position estimation apparatus and a position estimation method capable of estimating its own position.

Explanation of Signs

[0075] 1 Anchor vehicle 2 3D model 3 Flash-Lidar 4 3D point cloud 5 Position matching 11 Anchor vehicle 12 Own vehicle 20 Position estimation apparatus 21 Image capturing device 22 3D model database 23 Point cloud detection device 24 Model identification unit 25 3D model extraction unit 26 Point cloud matching unit 27 Position estimation unit 28 Object position identification unit 29 Vehicle position identification unit 31 Pedestrian

Claims

1. A specific circuit that identifies a stationary object model representing a stationary object existing around a vehicle from among a plurality of models that model an object, a position estimation circuit that detects a relative positional relationship of the vehicle with respect to the stationary object at each time based on the stationary object model and a point cloud corresponding to the stationary object detected at each time by a point cloud detection device, and estimates the position of the vehicle at each time based on the relative positional relationship at each time; A position estimation device comprising the above.

2. The position estimation device according to claim 1, wherein the specific circuit identifies the stationary object model based on an image of the stationary object captured by an image capturing device.

3. The position estimation device according to claim 1, wherein the specific circuit identifies the stationary object model based on a determination result as to whether or not the point cloud matches the shape of the stationary object model.

4. The position estimation device according to claim 3, wherein the specific circuit performs clustering to classify the point cloud into a first point cloud corresponding to the stationary object and a second point cloud corresponding to an object other than the stationary object, and determines whether or not the first point cloud matches the shape of the stationary object model.

5. The position estimation device according to claim 3, wherein the specific circuit determines whether or not the point cloud matches the shape of the stationary object model based on a distance between the point cloud and the surface of the stationary object model.

6. The position estimation device according to claim 5, wherein the specific circuit determines whether or not point clouds having a distance greater than a threshold value among the point clouds are spatially concentrated, excludes the spatially concentrated point clouds having a distance greater than the threshold value, and determines whether or not the point cloud matches the shape of the stationary object model.

7. The position estimation device according to claim 3, wherein the specific circuit generates a polygon model of the stationary object based on the point cloud, determines an incident angle of a beam emitted by the point cloud detection device to the stationary object based on the polygon model, excludes point clouds corresponding to beams having an incident angle shallower than a predetermined angle, and determines whether or not the point cloud matches the shape of the stationary object model.

8. The position estimation device according to claim 1, wherein the stationary object is a stationary vehicle, and the stationary object model does not include data of wheels, door mirrors, or glass portions of the stationary vehicle.

9. Identify a stationary object model representing a stationary object existing around a vehicle from among a plurality of models that model an object, A position estimation method for detecting a relative positional relationship of the vehicle with respect to the stationary object at each time based on the stationary object model and the point cloud corresponding to the stationary object detected by the point cloud detection device at each time, and estimating the position of the vehicle at each time based on the relative positional relationship at each time.

10. The stationary object model is specified based on an image of the stationary object photographed by an image photographing device. The position estimation method according to claim 9.

11. The stationary object model is specified based on a determination result as to whether or not the point cloud matches the shape of the stationary object model. The position estimation method according to claim 9.

12. Furthermore, clustering is performed to classify the point cloud into a first point cloud corresponding to the stationary object and a second point cloud corresponding to an object other than the stationary object. A determination is made as to whether or not the first point cloud matches the shape of the stationary object model. The position estimation method according to claim 11.

13. Based on a distance between the point cloud and a surface of the stationary object model, it is determined whether or not the point cloud matches the shape of the stationary object model. The position estimation method according to claim 11.

14. Furthermore, it is determined whether or not point clouds having a distance greater than a threshold value among the point clouds are spatially concentrated. The point cloud is determined as to whether or not it matches the shape of the stationary object model by excluding point clouds having a distance greater than the threshold value that are spatially concentrated. The position estimation method according to claim 13.

15. Furthermore, a polygon model of the stationary object is generated based on the point cloud. Based on the polygon model, an incident angle of a beam emitted by the point cloud detection device to the stationary object is determined. The point cloud is determined as to whether or not it matches the shape of the stationary object model by excluding point clouds corresponding to beams having an incident angle shallower than a predetermined angle. The position estimation method according to claim 11.

16. The stationary object is a stationary vehicle, and the stationary object model does not include data of wheels, door mirrors, or glass portions of the stationary vehicle. The position estimation method according to claim 9.

Images