OBJECT RECOGNITION DEVICE, OBJECT RECOGNITION METHOD, AND OBJECT RECOGNITION PROGRAM

Abstract

To add noise corresponding to distance to each point group around the point group of an object without losing a feature quantity added to the object.SOLUTION: An object recognition device 100 includes: a point group measurement unit 10 for measuring a position of point groups included in an object and distance to each point group; a noise addition unit 20 for adding noise according to distance to each point group around the measured point group; a noise number storage unit 30 for storing a total number and a position of the noise added by the noise addition unit; and a recognition unit 40 for recognizing the object from the point group and noise added to the point group.SELECTED DRAWING: Figure 1

Description

[Technical Field]

[0001] The present invention relates to an object recognition device, an object recognition method, and an object recognition program. [Background technology]

[0002] Patent Document 1 discloses an object recognition device that adds a predetermined noise to multiple ranging points measured by a ranging device, then forms a cluster representing an object based on the ranging points to which the noise has been added, and recognizes the object based on the feature values of the formed cluster. [Prior art documents] [Patent documents]

[0003] [Patent Document 1] Patent Publication No. 2019-153188 Summary of the Invention [Problem to be solved by the invention]

[0004] For example, in the technology disclosed in Patent Document 1, noise is added to multiple distance measurement points (hereinafter also referred to as "point cloud") measured by a distance measurement device such as LiDAR (Light Detection and Ranging). Adding noise adds features to the object, which causes a stochastic resonance phenomenon and improves the recognition rate of the object. The recognition rate can be further improved by varying the number and range of noise added depending on the distance to the object. However, simply adding multiple noises overwrites previously added noise data, which may result in the loss of features added to the object.

[0005] The present invention has been made in consideration of these circumstances, and its purpose is to add noise around the point cloud of an object according to the distance to each point cloud without losing the features added to the object. [Means for solving the problem]

[0006] In order to solve the above problems, an object recognition device according to one embodiment of the present invention is an object recognition device for recognizing an object, and includes a point cloud measurement unit that measures the positions of point clouds included in the object and the distance to each point cloud, a noise addition unit that adds noise around the measured point cloud according to the distance to each point cloud, a noise number memory unit that stores the number of noises added by the noise addition unit for each position of the added noise, and a recognition unit that recognizes the object from the point clouds and the noise added to the point clouds.

[0007] Another aspect of the present invention is an object recognition method for recognizing an object, the method comprising: a point cloud measurement step of measuring positions of point clouds included in the object and distances to each of the point clouds, a noise addition step of adding noise around the measured point clouds according to the distance to each of the point clouds, a noise count storage step of storing the number of added noises for each position of the added noise, and a recognition step of recognizing the object from the point clouds and the noise added to the point clouds.

[0008] Yet another aspect of the present invention is an object recognition program, which is an object recognition method for recognizing an object, and causes a computer to execute the following steps: a point cloud measurement step of measuring positions of point clouds included in the object and distances to each of the point clouds, a noise addition step of adding noise around the measured point clouds according to the distance to each of the point clouds, a noise number storage step of storing the number of added noises for each position of the added noise, and a recognition step of recognizing the object from the point clouds and the noise added to the point clouds.

[0009] Any combination of the above components, and any transformation of the present invention into a method, device, system, recording medium, computer program, etc., are also valid aspects of the present invention. [Effects of the Invention]

[0010] According to the present invention, noise according to the distance to each point cloud can be added to the periphery of the point cloud of the object without losing the feature amount added to the object. [Brief explanation of the drawings]

[0011] [Figure 1] 1 is a functional block diagram of an object recognition device according to a first embodiment. [Figure 2] 10A and 10B are diagrams showing examples of noise added by a noise adding unit to the periphery of point clouds measured by a point cloud measuring unit, where (a) is a point cloud of a relatively close object, and (b) is a point cloud of a relatively far object. [Figure 3] FIG. 3 is a diagram showing the distribution range of noise added in (a) and (b) of FIG. 2. [Figure 4] 2(a) is a diagram showing points P1 and P2 on the XY plane. [Figure 5] FIG. 10 is a diagram showing noise added around points P1 and P2. [Figure 6] FIG. 10 is a diagram showing the number of noises for each noise position stored in a noise number storage unit. [Figure 7] The numbers of noises stored in the noise number storage unit are shown. (a) is the number of noises added around point P1. (b) is the number of noises added around point P2. [Figure 8] 13 is a flowchart illustrating a processing procedure of an object recognition method according to a fifth embodiment. DETAILED DESCRIPTION OF THE INVENTION

[0012] The present invention will be described below based on preferred embodiments with reference to the drawings. In the embodiments and modifications, identical or equivalent components and parts are designated by the same reference numerals, and redundant description will be omitted where appropriate. The dimensions of the components in the drawings are enlarged or reduced as appropriate for ease of understanding. Some elements that are not important for explaining the embodiments are omitted from the drawings. Terms including ordinal numbers such as "first" and "second" are used to describe various components, but these terms are used only to distinguish one component from another and do not limit the components.

[0013] [First embodiment] 1 is a functional block diagram of an object recognition device 100 according to a first embodiment. The object recognition device 100 includes a point cloud measurement unit 10, a noise addition unit 20, a noise count storage unit 30, and a recognition unit 40.

[0014] The point cloud measurement unit 10 measures the position of a point cloud, which is one or more ranging points included on the surface of an object to be recognized (referred to as "object" in this specification), and the distance to each point cloud. The object recognition device 100 may be configured to include, for example, a LiDAR. The LiDAR measures the position of the point cloud by irradiating and scanning the object with laser light and receiving the light reflected from the object, and can measure the distance from the point cloud measurement unit 10 to the point cloud based on the time required from irradiation to reception of the light. When the scanning range of the LiDAR is an XY plane, the position and distance of each point included in the point cloud are acquired as a two-dimensional array in which the position of each point is represented by coordinates on the XY plane and the distance of each point is represented by coordinate values.

[0015] Point clouds measured by distance measurement devices such as LiDAR may not adequately reflect the features of an object due to reasons such as an insufficient number of points. In such cases, the measured point cloud may be considered noise rather than features in object recognition. By adding noise to the surroundings of such point clouds, stochastic resonance can be triggered, making the features of the object stand out.

[0016] The noise addition unit 20 adds noise around each point cloud measured by the point cloud measurement unit 10 according to the distance to the point cloud. Because the scanning range of the laser light expands radially in the irradiation direction, the spacing between the irradiated laser beams increases as the distance from the light source increases. Therefore, the farther the object is, the fewer points can receive the reflected light from the laser light irradiated by the scanning. As a result, the number of points included in the point cloud output from the LiDAR decreases. Therefore, when adding noise around each point cloud, it is considered better to add more noise the shorter the distance to each point cloud and less noise the longer the distance to each point cloud. Furthermore, it is considered better to add a standard deviation (spatial spread of the added noise), which indicates the variation in the distribution range of the added noise, the shorter the distance to each point cloud and the smaller the distance to each point cloud.

[0017] The noise added by the noise adding unit 20 is based on the distance from the point cloud measuring unit 10 to each point. For example, the noise is generated by first acquiring the position of each point as coordinates on the XY plane, and then acquiring the distance to each point as a coordinate value. In this case, the noise value added to each point indicates the distance to each point. A specific noise value may be, for example, a binary value indicating the presence or absence of each point in the point cloud. Alternatively, the noise value may be a value according to the distance to each point, such as the distance to each point, the distance to each point multiplied by a predetermined coefficient, or a value with a certain degree of variation based on the distance to each point. Furthermore, a random value may be assigned as the noise value.

[0018] FIG. 2 shows an example in which noise has been added by the noise addition unit 20 to the periphery of a point cloud measured by the point cloud measurement unit 10. (a) is a point cloud of a relatively close object, and (b) is a point cloud of a relatively distant object. Seven point clouds are observed in (a), and two point clouds in (b). It can be seen that more noise has been added over a wider area around the point cloud in (a) compared to (b). Two of the seven point clouds in (a) are named P1 and P2. In FIG. 2, the points indicated by a mesh are the point cloud of the object, and the points indicated by diagonal lines are noise. Points P1 and P2 will be described later.

[0019] Figure 3 shows the distribution range of the noise added in Figure 2 (a) and (b). The solid line is the noise distribution in (a), and the dashed line is the noise distribution in (b). The horizontal axis of the graph shows the noise distribution range, and the vertical axis shows the distribution density. Compared to (b), (a) has a wider distribution range of the added noise and a lower distribution density. Therefore, it can be seen that (a) has a larger standard deviation, which indicates the variability in the distribution range.

[0020] The noise number storage unit 30 stores the number of noises added by the noise adding unit 20 for each position of the added noise.

[0021] Figure 4 shows points P1 and P2 in Figure 2(a) plotted on the XY plane. The coordinates of points P1 and P2 are (3, 8) and (5, 5), respectively.

[0022] Figure 5 shows noise added around points P1 and P2. The horizontal stripes indicate the location of the noise added around point P1, and the vertical stripes indicate the location of the noise added around point P2. Specifically, the noise added around point P1 is located at (1, 8), (2, 7), (2, 8), (2, 9), (3, 6), (3, 7), (3, 8), (3, 9), (3, 10), (4, 8), and (4, 9). The noise added around point P2 is located at (4, 4), (4, 5), (4, 6), (4, 7), (5, 3), (5, 4), (5, 5), (5, 6), (5, 7), (6, 4), (6, 5), (6, 6), (7, 5), and (7, 6).

[0023] Note that in Figure 5, noise added later overwrites previously added noise data. Specifically, in Figure 5, noise data added around point P1 is first recorded, followed by noise data added around point P2. At (4, 7), noise was first added around point P1. Then, noise was added around point P2 at (4, 7). Therefore, the noise data around point P1 recorded at (4, 7) was overwritten with the noise data around point P2 and lost. As a result, only the noise added around point P2 is recorded at (4, 7). This means that some of the features related to point P1 have been lost. Furthermore, when adding noise around a single point, noise may be added multiple times to the same coordinates. In other words, coordinates to which noise has already been added may be overwritten with the same noise data.

[0024] FIG. 6 shows the number of noises stored in the noise number storage unit 30 for each noise position. Specifically, two noises are added to (1, 8), one to (2, 7), two to (2, 8), one to (2, 9), one to (3, 6), three to (3, 7), six to (3, 8), three to (3, 9), two to (3, 10), one to (4, 4), four to (4, 5), two to (4, 6), six to (4, 7), two to (4, 8), one to (4, 9), one to (5, 3), three to (5, 4), seven to (5, 5), three to (5, 6), one to (5, 7), three to (6, 4), three to (6, 5), two to (6, 6), one to (7, 5), and one to (7, 6).

[0025] By storing the number of noises added to each position in this way, all information about the added noise can be retained without overwriting the noise data, meaning that the feature values added to the object will not be lost.

[0026] A large amount of noise added at a certain position indicates that the position is important as a feature of the object. In particular, in this example, the number of noises at (4, 7) is 6, which is higher than the surrounding positions. This is because noise was added to both points P1 and P2, but at the same time, this also shows that (4, 7) is an important position connecting points P1 and P2.

[0027] It should be noted that the amount of noise added at each position does not matter whether it is due to noise added around point P1 or point P2. The amount of added noise at each position (the distribution of values shown in Figure 6) continues to follow a similar pattern when measurements are repeated. This makes it possible to determine whether the noise reflects the feature of the object or is simply noise.

[0028] The noise value added by the noise adding unit 20 may be any value, such as a binary value, indicating the presence or absence of each point in the point cloud. It may also be a value based on the coordinate value of each point. In this embodiment, the coordinate value of each point is the distance from the point cloud measuring unit 10 to each point, and the noise value added to each point is a value according to the distance to each point, being larger the longer the distance and smaller the shorter the distance. For example, the noise value may be the distance to each point, the distance to each point multiplied by a predetermined coefficient, or a value with a certain degree of variation based on the distance to each point. Furthermore, a random value may be added as the noise value.

[0029] The recognition unit 40 recognizes the object from the point cloud and the noise added to the point cloud.

[0030] In the above example, noise is added to points P1 and P2 included in the same point cloud with the same distribution. The distribution of noise added to each point in the same point cloud is a distribution according to the distance from each point cloud. Here, the distance from the point cloud is a representative value of the distances to each point included in the point cloud. The representative value may be calculated based on the distance to each point. For example, the representative value may be the average, median, maximum, or minimum value of the distances to each point. Alternatively, a representative coordinate may be calculated from the coordinates of each point included in the point cloud, and the distance from that coordinate may be used as the distance from the point cloud. The representative coordinate may be, for example, the center of gravity of each point. However, this is not limited to this, and noise may be added with a distribution according to the distance from each point. In other words, noise may be added to points P1 and P2 with different distributions according to the distance from each point.

[0031] As described above, according to this embodiment, it is possible to add noise around the point cloud of an object according to the distance to each point cloud without losing the feature amount added to the object, thereby enabling highly accurate object recognition.

[0032] [Second embodiment] In one embodiment, the noise number storage unit 30 stores, for each point, the number of noises added around each point in the measured point cloud.

[0033] FIG. 7 shows the number of noises stored in the noise number storage unit 30. (a) is the number of noises added around point P1. (b) is the number of noises added around point P2. Specifically, the number of noises added around point P1 is There are two in (1, 8), one in (2, 7), two in (2, 8), one in (2, 9), one in (3, 6), three in (3, 7), six in (3, 8), three in (3, 9), two in (3, 10), two in (4, 7), two in (4, 8), and one in (4, 9). The number of noises added around point P2 is 1 at (4, 4), 4 at (4, 5), 2 at (4, 6), 4 at (4, 7), 1 at (5, 3), 3 at (5, 4), 7 at (5, 5), 3 at (5, 6), 1 at (5, 7), 3 at (6, 4), 3 at (6, 5), 2 at (6, 6), 1 at (7, 5), and 1 at (7, 6).

[0034] According to this embodiment, the number of noises added around each point in the measured point cloud is stored for each point, so it is possible to know the point around which the added noise was added, thereby enabling more accurate object recognition.

[0035] [Third embodiment] In one embodiment, the point cloud measurement unit 10 not only measures the positions of the point clouds included in the object and the distances to each point cloud, but also measures the luminance of the point clouds included in the object. The luminance may be acquired by providing a luminance sensor and acquiring the luminance within the range of the distance acquired by LiDAR. Alternatively, the luminance may be calculated from information on the reflection intensity of the object acquired by LiDAR. The luminance is acquired as a two-dimensional array in which the position of each point is represented by a coordinate on the XY plane and the luminance of each point is represented by a coordinate value.

[0036] As in the first embodiment, the noise addition unit 20 adds noise based on the luminance of each point. Specifically, the noise addition unit 20 adds noise around the point group whose luminance has been acquired, with a value based on the luminance of each point, with a distribution according to the distance to each point group. The noise value is given so that it is larger the higher the luminance and smaller the lower the luminance. For example, it may be the luminance up to each point, the luminance of each point multiplied by a predetermined coefficient, or a value with a certain degree of variation based on the luminance of each point. The noise number storage unit 30 stores the number of added noises for each position of the added noise.

[0037] According to this embodiment, noise can be added to the brightness of an object in addition to the distance of the object without losing the feature amount added to the object, thereby enabling more accurate object recognition.

[0038] [Fourth embodiment] In one embodiment, the point cloud measurement unit 10 not only measures the positions of the point clouds included in the object and the distances to each point cloud, but also measures the colors of the point clouds included in the object. The color information may be acquired by providing an image sensor and acquiring color information within a range where distances are acquired using LiDAR. Alternatively, the color information may be acquired using LiDAR that emits laser light of multiple different wavelengths. Alternatively, the color information may be calculated from information on the reflection intensity of the object acquired by LiDAR. The color information is acquired as a two-dimensional array of each component (e.g., R component, G component, B component) where the position of each point is the coordinate on the XY plane and the pixel value of each color component (e.g., RGB) of each point is the coordinate value. For example, if the color components are acquired as RGB values, the information is acquired as three two-dimensional arrays of the R component, the G component, and the B component.

[0039] As in the first embodiment, the noise addition unit 20 adds noise based on the value of each color component (e.g., RGB) of each point. Specifically, in the two-dimensional array of each color component, the noise addition unit 20 adds noise based on the value of the color information of each point, around the point group from which color information has been acquired, with a distribution according to the distance to each point group. The noise value is given so that it increases as the pixel value of each RGB component increases and decreases as the pixel value decreases. For example, the noise value may be the pixel value up to each point, the pixel value of each point multiplied by a predetermined coefficient, or a value with a certain degree of variation based on the pixel value of each point. The noise number storage unit 30 stores the number of added noises for each position of the added noise.

[0040] According to this embodiment, noise can be added to the color of an object in addition to the distance of the object without losing the feature amount added to the object, thereby enabling more accurate object recognition.

[0041] In the above description, the point cloud measurement unit 10 has been described as including a LiDAR. However, the present invention is not limited to this, and the point cloud measurement unit 10 may be any unit that can measure the positions of point clouds included in an object and the distances to each point cloud.

[0042] The various processes of the object recognition device described above can be realized not only as a device using hardware such as a CPU and memory, but also as firmware stored in a ROM (read-only memory) or flash memory, or software for a computer, etc. The firmware program or software program can be provided by being recorded on a computer-readable recording medium, or can be transmitted and received with a server via a wired or wireless network, or can be transmitted and received as data broadcasting via terrestrial or satellite digital broadcasting.

[0043] [Fifth embodiment] FIG. 8 is a flowchart illustrating a processing procedure of the object recognition method according to the fifth embodiment.

[0044] First, in this method, the position of a point cloud included in the object and the distance to each point cloud are measured using a point cloud measurement unit (S1).

[0045] Next, in this method, a noise adding unit is used to add noise around the measured point cloud according to the distance to each point cloud (S2).

[0046] Next, the method uses a noise number storage unit to store the number of added noises for each position of the added noise (S3).

[0047] Next, the method uses a recognition unit to recognize an object from the point cloud and the noise added to the point cloud (S4).

[0048] According to this embodiment, noise according to the distance to each point cloud can be added to the periphery of the point cloud of the object without losing the feature amount added to the object.

[0049] [Sixth embodiment] The sixth embodiment is a computer program that causes a computer to execute a point cloud measurement step of measuring the positions of point clouds included in an object and the distances to each point cloud, a noise addition step of adding noise around the measured point cloud according to the distance to each point cloud, a noise count storage step of storing the number of added noises for each position of the added noise, and a recognition step of recognizing the object from the point clouds and the noise added to the point clouds.

[0050] According to this embodiment, a program can be implemented as software that adds noise around the point cloud of an object according to the distance to each point cloud without losing the feature amount added to the object.

[0051] Figure 4-7 shows the distribution of points in a 10 x 10 coordinate system. However, this is not limited to this, and the acquired points may be distributed in any range.

[0052] In the above embodiment, the case where one object is recognized has been described, but the present invention is not limited to this, and an embodiment may acquire a range image within a predetermined range and recognize multiple objects that exist at different distances.

[0053] In the above embodiment, the case where noise is added around each point included in the point cloud has been described, but the present invention is not limited to this, and noise may be added to the periphery of the point cloud.

[0054] The present invention has been described above based on the embodiments. The embodiments are merely examples, and it will be understood by those skilled in the art that various modifications are possible in the combination of the respective components and treatment processes, and that such modifications are also within the scope of the present invention. [Explanation of symbols]

[0055] 10...Point cloud measurement unit, 20...Noise addition unit, 30...Noise number storage unit, 40...Recognition unit, 50...Each point noise storage unit, S1...Step of measuring the position and distance of the point cloud, S2...Step of adding noise, S3...Step of storing the number of added noises for each position of the added noise, S4...Step of recognizing an object, 100...Object recognition device.

Claims

1. An object recognition device for recognizing an object, a point cloud measurement unit that measures the positions of point clouds included in the object and the distances to each point cloud; a noise adding unit that adds noise around the measured point cloud according to the distance to each point cloud; a noise number storage unit that stores the number of noises added by the noise adding unit for each position of the added noise; and a recognition unit that recognizes the object from the point cloud and noise added to the point cloud.

2. 2. The object recognition device according to claim 1, wherein the number of noises added around the measured point clouds increases as the distance to each point cloud decreases.

3. 3. The object recognition device according to claim 1, wherein the variation in the range in which the noise added around the measured point cloud is distributed increases as the distance to each point cloud decreases.

4. the noise number storage unit stores the number of noises added by the noise addition unit for each point of the point cloud measured by the point cloud measurement unit; 4. The object recognition device according to claim 1, wherein the number of added noises at each position is stored for each point of the point cloud.

5. the point cloud measurement unit further measures the brightness of the point cloud included in the object; 5. The object recognition device according to claim 1, wherein the noise adding unit adds noise around the measured point clouds according to the distance to each point cloud and the brightness of each point cloud.

6. the point cloud measurement unit further measures the color of the point cloud included in the object; 6. The object recognition device according to claim 1, wherein the noise adding unit adds noise around the measured point clouds according to the distance to each point cloud and the color of each point cloud.

7. An object recognition method for recognizing an object, comprising: a point cloud measurement step in which the object recognition device measures the positions of point clouds included in the target object and the distances to each point cloud; a noise addition step of adding noise around the measured point cloud according to the distance to each point cloud; a noise number storage step for storing the number of added noises and the positions of the added noises; a recognition step of recognizing the object from the point cloud and noise added to the point cloud.

8. An object recognition program for recognizing an object, a point cloud measurement step of measuring positions of point clouds included in the object and distances to each point cloud; a noise addition step of adding noise around the measured point cloud according to the distance to each point cloud; a noise number storage step for storing the number of added noises and the positions of the added noises; a recognition step of recognizing the object from the point cloud and noise added to the point cloud.

Images