Control device and control method for moving body

Abstract

To safely and autonomously travel a moving body.SOLUTION: A control device 30A of a mobile body 100 includes an image analysis unit 31 that identifies a travelable area of the mobile body 100 by performing semantic segmentation on an omnidirectional image obtained by imaging the mobile body 100 in all directions and estimates a range from a current position of the mobile body 100 to an arbitrary position in the omnidirectional image, a pixel histogram calculation unit 32 that calculates a pixel histogram representing a distribution of the number of pixels of the travelable area in an image longitudinal direction, a path determination unit 33 that determines a local path from the current position of the mobile body 100 to a target point on a straight line passing through the current position of the mobile body 100 and a position corresponding to a peak point of the pixel histogram in the omnidirectional image, and a control unit 34 that causes the mobile body 100 to travel along the local path.SELECTED DRAWING: Figure 1

Description

[Technical Field]

[0001] The present invention relates to a control device and a control method for a moving body, and more particularly to a control device and a control method suitable for autonomous driving of personal mobility such as an electric wheelchair. [Background technology]

[0002] There are two main types of wheelchairs: manual wheelchairs and electric wheelchairs. Electric wheelchairs, which place less strain on the body, are often used by people with physical disabilities. The movement of an electric wheelchair is controlled by operating a joystick, but operating a joystick can be difficult for people with upper limb disabilities. To address this issue, a technology is known that estimates where the occupant of the electric wheelchair is looking from their line of sight and controls the direction of movement of the electric wheelchair (see, for example, Patent Document 1).

[0003] Furthermore, to autonomously navigate a mobile object such as an electric wheelchair, it is necessary to constantly observe and recognize the road, people, and obstacles, plan a local route, and move by following that route. Regarding such vision-based route following, a known method is to detect the drivable area by performing semantic segmentation on a bird's-eye view synthesized from cameras attached around the vehicle, and to determine the predicted point where the vehicle should head by imitation learning (see, for example, Non-Patent Document 1). [Prior art documents] [Non-patent literature]

[0004] [Non-Patent Document 1] Ahn, J., Kim, M. & Park, J., “Autonomous driving using imitation learning with look ahead point for semi structured environments,” Scientific Reports 12, 21285 (2022). [Patent documents]

[0005] [Patent Document 1] Japanese Patent Application Publication No. 2023-29216 Summary of the Invention [Problem to be solved by the invention]

[0006] When observing the driving environment ahead of a moving vehicle using a bird's-eye view, it becomes impossible to detect distant obstacles, and there is a risk that the proportion of the drivable area within a range of 1 m ahead of the moving vehicle will decrease when approaching an obstacle or a corner.In particular, compared to outdoor environments, indoor environments have narrower driving paths and are more likely to contain obstacles, so when an electric wheelchair is autonomously driving in an indoor environment, if the drivable area ahead is reduced, there is a risk of it colliding with a wall or obstacle.

[0007] Therefore, an object of the present invention is to provide a control device and a control method that can allow a moving body to travel safely and autonomously. [Means for solving the problem]

[0008] According to one aspect of the present invention, there is provided a control device for a mobile body that causes the mobile body to travel autonomously, the control device comprising: an image analysis unit that performs semantic segmentation on an omnidirectional image that captures the mobile body in all directions to identify a drivable area of ​​the mobile body and estimates the distance from the current position of the mobile body to an arbitrary position in the omnidirectional image; a pixel histogram calculation unit that calculates a pixel histogram that represents the distribution of the number of pixels in the vertical direction of the image of the drivable area; a route determination unit that determines a local route from the current position of the mobile body to a target point a predetermined distance away on a straight line that passes through the current position of the mobile body and a position in the omnidirectional image that corresponds to the peak point of the pixel histogram; and a control unit that causes the mobile body to travel along the local route, as well as a corresponding control method. [Effects of the Invention]

[0009] According to the present invention, a local route that a mobile object can travel is determined from an omnidirectional image of the mobile object, and the mobile object is made to travel along this route, thereby enabling the mobile object to travel safely and autonomously. [Brief explanation of the drawings]

[0010] [Figure 1] 1 is a block diagram of a control device for a moving body according to a first embodiment of the present invention. [Figure 2] 10 is a control flowchart relating to autonomous traveling of a moving body. [Figure 3] 10A and 10B are diagrams illustrating an example of image analysis of an omnidirectional image related to autonomous driving of a moving body. [Figure 4] FIG. 10 is a diagram illustrating an example configuration of a neural network that estimates a segmentation map and a depth map from an omnidirectional image and a cube map image. [Figure 5] FIG. 10 is a block diagram of a control device for a moving body according to a second embodiment of the present invention. [Figure 6] FIG. 10 is a diagram illustrating an example of the configuration of a neural network that estimates the intended traveling direction of a passenger from the passenger's gaze point data and the speed data of a moving object. [Figure 7] FIG. 10 is a block diagram of a control device for a moving body according to a third embodiment of the present invention. [Figure 8] 1A and 1B are diagrams illustrating a method for generating a bird's-eye view from an omnidirectional image. [Figure 9] FIG. 1 illustrates an example of an omnidirectional image and a bird's-eye view generated from it. [Figure 10] FIG. 10 is a diagram showing an example of a bird's-eye view map of the predicted course of a moving object. DETAILED DESCRIPTION OF THE INVENTION

[0011] Hereinafter, embodiments of the present invention will be described in detail with reference to the accompanying drawings as appropriate. However, more detailed description than necessary may be omitted. For example, detailed description of well-known matters or redundant description of substantially identical configurations may be omitted. This is to avoid unnecessary redundancy in the following description and to facilitate understanding by those skilled in the art. Note that the inventors provide the accompanying drawings and the following description to enable those skilled in the art to fully understand the present invention, and do not intend for them to limit the subject matter described in the claims.

[0012] (First embodiment) 1 is a block diagram of a control device for a mobile body according to a first embodiment of the present invention. Mobile body 100 is a personal mobility device, such as an electric wheelchair, in which a single passenger sits and which moves at a speed of approximately 10 km / h or less. The passenger can operate a joystick to move mobile body 100 at their own will, and mobile body 100 can also autonomously travel by assessing the surrounding conditions and determining a possible route.

[0013] The mobile object 100 includes a driving unit 10, an omnidirectional camera 20, and a control device 30A. The driving unit 10 includes a wheel and a motor for driving the wheel to rotate. The rotation of the wheel allows the mobile object 100 to move forward and backward, and the direction of travel of the mobile object 100 can be changed by rotating the mobile object 100 around a vertical axis.

[0014] The omnidirectional camera 20 is a device that captures omnidirectional images of the mobile object 100. For example, the omnidirectional camera 20 may be composed of two fisheye cameras, and the omnidirectional image is obtained by combining the images captured by each camera. The omnidirectional camera 20 can transfer the captured omnidirectional images to the control device 30A via wireless or wired communication. Note that, since the omnidirectional image of the mobile object 100 is an important source of information for the mobile object 100 to determine the surrounding situation during autonomous driving, it is desirable to install the omnidirectional camera 20 at a certain height so that it can capture a 180-degree field of view in front of the mobile object 100.

[0015] The control device 30A is a device that controls the drive device 10. Specifically, the control device 30A is composed of electronic components such as a processor and a memory. In particular, as functions related to the autonomous driving of the moving body 100, the control device 30A is equipped with an image analysis unit 31, a pixel histogram calculation unit 32, a route determination unit 33, and a control unit 34. These components can be realized as either hardware modules or software modules. Alternatively, they can be realized by combining hardware modules and software modules. In particular, the trained model described below operates when parameters stored in memory are read out by the processor.

[0016] Fig. 2 is a control flowchart for the autonomous traveling of a moving body. Fig. 3 is a diagram showing an example of image analysis of an omnidirectional image for the autonomous traveling of a moving body. Hereinafter, a control method for the autonomous traveling of the moving body 100 and the operation of each component of the control device 30A will be described with reference to Figs. 2 and 3.

[0017] First, the omnidirectional camera 20 installed on the moving object 100 captures an omnidirectional image of the moving object 100 (S1). Fig. 3(a) is an example of the omnidirectional image of the moving object 100. The omnidirectional image in this example is an image taken when the moving object 100 is traveling down a linear corridor. The captured omnidirectional image is temporarily stored in a memory (not shown) in the control device 30A.

[0018] Next, the image analysis unit 31 reads the omnidirectional image from memory and performs semantic segmentation (S2). Semantic segmentation can be performed using existing deep learning technologies such as DeepLabv3+. Specifically, a segmentation map is output by inputting the omnidirectional image into a trained model. Considering that semantic segmentation is used to control the autonomous driving of the mobile object 100, the segmentation classes are divided into three: (1) "drivable areas" where the mobile object 100 can travel, such as roads and corridors; (2) "passengers and the mobile object itself" of the mobile object 100; and (3) "other areas." Reducing the number of classes to be identified in this way enables faster semantic segmentation. Figure 3(b) shows an image in which the segmentation map, divided into three classes, is superimposed on (a). Semantic segmentation identifies the corridor as a drivable area.

[0019] Once semantic segmentation of the omnidirectional image is complete, the pixel histogram calculation unit 32 calculates a pixel histogram that represents the number of pixels in the vertical direction of the image in the drivable area in the segmentation map (S3). Figure 3(c) is an image in which the pixel histogram of the drivable area is superimposed on (b). In the case of a linear corridor, the pixel histogram of the drivable area will have a single-peaked mountain shape with a peak at the front end of the corridor.

[0020] Once the pixel histogram of the drivable area is calculated, the route determination unit 33 determines a local route from the current position of the mobile object 100 to a target point a predetermined distance away on a straight line passing through the current position of the mobile object 100 and a position corresponding to the peak point of the pixel histogram in the omnidirectional image (S4). More specifically, the route determination unit 33 draws a straight line (hereinafter sometimes referred to as the target route) passing through two points in the omnidirectional image, the current position of the mobile object 100 and the position corresponding to the peak point of the pixel histogram, to infinity ahead, and determines the local route from the current position of the mobile object 100 to the target point a predetermined distance away on the target route. Note that the current position of the mobile object 100 on the omnidirectional image shifts left or right depending on the installation position of the omnidirectional camera 20, so it is desirable to correct the omnidirectional image in advance so that the current position of the mobile object 100 is located at the center of the bottom edge of the omnidirectional image.

[0021] Figure 3(d) is an image with a local route superimposed on (c). The line extending from the center of the bottom edge of the image to the center of the image corresponds to the local route. The six horizontal lines in the image represent, from bottom to top, points 1 m, 2 m, 3 m, 4 m, 5 m, and 10 m from the current position of the mobile object 100. In this example, the target point of the local route is set to a point 10 m ahead of the current position of the mobile object 100. The circles at the intersections of the target route and each horizontal line represent passing points on the way to the target point.

[0022] To determine the local route, it is necessary to know the distance from the current position of the mobile object 100 to an arbitrary position in the omnidirectional image. This distance can be estimated by calculation by the image analysis unit 31 based on various conditions such as the optical characteristics of the omnidirectional camera 20, the height of the lens from the ground or floor, the angle and direction relative to the ground or floor, etc. From these conditions, the zenith angle θ to an arbitrary point in the omnidirectional image can be determined, and if the height from the ground or floor to the lens of the omnidirectional camera 20 is H, the distance from the current position of the mobile object 100 to that point can be calculated as Htan(θ-π).

[0023] The mobile object 100 travels autonomously along a local route. For this reason, the local route must be one that can safely guide the mobile object 100. In an omnidirectional image, objects that are further away appear smaller due to perspective, so the estimated distance to a distant position has a large error. For this reason, if the position corresponding to the peak point of the pixel histogram is set as the target point of the local route, the error in the distance to the target point will be large if that position is far away, and there is a risk that the mobile object 100 will not be able to be safely guided. Therefore, by dividing the local route into small sections at a predetermined distance as described above and repeatedly determining short-distance local routes, the mobile object 100 can be safely guided.

[0024] Once the local route is determined, the control unit 34 causes the moving object 100 to travel along the local route. Prior to this, the control unit 34 determines whether the moving object 100 can travel (S5). Specifically, the control unit 34 determines whether the moving object 100 can travel along the local route by determining whether the height from the ground or floor to the lens of the omnidirectional camera 20 is H and the safe distance is D. T , where θ is the zenith angle to the end point of the travelable area of ​​the vehicle width of the moving body 100, and Htan(θ-π)-D T If Htan(θ-π) is too short, it is dangerous to allow the moving object 100 to travel. Therefore, a safe distance D T is set.

[0025] If the control unit 34 determines that the moving body 100 is capable of traveling (yes in S5), it determines the traveling conditions of the moving body 100 (S6). The traveling of the moving body 100 is determined by two elements: rotation about a vertical axis and translational movement, which can be proportionally controlled. Specifically, the control unit 34 issues a command to the driving device 10 of the moving body 100 to cause the moving body 100 to travel using the rotation amount ω and translational movement amount s expressed by the following equations (S7). ω=κ ω φ ω s=κ s (Htan(π-θ f )-D T ) However, κ ω is the proportional constant of the rotation control, φω is the directional angle of the peak point of the pixel histogram when the front direction of the moving object 100 is 0° and the counterclockwise direction is positive, and κ s is a proportional constant for translational movement control, H is the height from the ground or floor to the lens of the omnidirectional camera 20, and θ f is the zenith angle to the tip of the drivable area, D T is the safety distance.

[0026] When the moving body 100 travels and reaches the target point of the local route (yes in S8), the process returns to step S1, where an omnidirectional image is acquired again, a local route is determined, and the moving body 100 travels. On the other hand, if the control unit 34 determines in step S5 that the moving body 100 cannot travel (no in S5), the moving body 100 cannot move forward any further, so the direction of the moving body 100 is changed (S9), and the process returns to step S1 to search for a route that can be traveled in a different direction.

[0027] (Simultaneous Semantic Segmentation and Distance Estimation) The image analysis unit 31 can simultaneously perform semantic segmentation and distance estimation for omnidirectional images using a trained model. Figure 4 shows an example of the configuration of a neural network that estimates a segmentation map and a depth map from an omnidirectional image and a cube map image. A cube map is obtained by converting an omnidirectional image. This neural network is a reconfiguration of UniFuse, which outputs a depth map from an omnidirectional image and a cube map image, so that it can also output a segmentation map simultaneously. For convenience, this neural network is called multitasking UniFuse.

[0028] The multitasking UniFuse 310 includes a two-branch encoder 311 and a two-branch decoder 312. The encoder 311, like UniFuse, functions as a feature extractor for omnidirectional and cube-map images. "Fuse" in the figure represents a fusion module. The decoder 312 has the function of restoring the features extracted by the segmentation and distance estimation encoders. A feature sharing process by the concatenation unit 313 is added to achieve synergy between tasks before inputting the data to the output layer. The concatenation unit 313 concatenates the final outputs of the branched hidden layers in the neural network in the column direction, performs depth-only convolution on the concatenated feature map, and then sends it to the output layer.

[0029] 4, learning data is provided to the multitasking UniFuse 310 so that it can learn in advance, and the image analysis unit 31 inputs an omnidirectional image of a moving object and its cube map image into the learned model, thereby enabling semantic segmentation and distance estimation for the omnidirectional image to be performed simultaneously. In this case, it is not necessary to calculate the distance from the zenith angle to an arbitrary position in the omnidirectional image to that position.

[0030] As described above, according to the control device 30A of this embodiment, a local route on which the mobile body 100 can travel is determined from an omnidirectional image of the mobile body 100, and the mobile body can travel along that route, thereby allowing the mobile body 100 to travel safely and autonomously.

[0031] (Second embodiment) FIG. 5 is a block diagram of a control device for a mobile body according to a second embodiment of the present invention. When a mobile body 100 traveling autonomously approaches a fork in the road, such as a T-junction, it cannot determine whether to turn right or left and is unable to proceed any further. The control device 30B according to this embodiment adds a traveling direction estimation unit 35 to the control device 30A according to the first embodiment, so that when there are multiple possible traveling directions, the traveling direction intended by the passenger can be estimated and the mobile body 100 can automatically turn. Below, we will omit repetition of matters already explained and will only explain the features of this embodiment.

[0032] The traveling direction estimation unit 35 estimates the intended traveling direction of the passenger of the moving body 100 from gaze point data of the passenger and the odometry of the moving body 100. The gaze point data of the passenger is time-series data representing the gaze point of the passenger. The odometry of the moving body 100 is time-series data representing the translational velocity and angular velocity of the moving body 100.

[0033] The traveling direction estimation unit 35 can acquire gaze point data from the gaze measurement device 200 via wireless or wired communication and can acquire odometry from the drive unit 10. The gaze measurement device 200 is a wearable device equipped with an eye camera that detects the passenger's pupil and a world camera that captures the passenger's field of view, and generates gaze point data based on information obtained from these cameras. Specifically, the gaze point data is expressed as normalized coordinate values ​​on a two-dimensional image.

[0034] The traveling direction estimation unit 35 can estimate the intended traveling direction of the passenger using the trained model. Fig. 6 is a diagram showing an example of the configuration of a neural network that estimates the intended traveling direction of the passenger from the passenger's gaze point data and the speed data of the moving object. For example, a Unet model or an LSTM model can be used as the neural network 350.

[0035] For simplicity, the neural network 350 outputs three directions of travel intended by the passenger: left turn, forward, and right turn. Since it is believed that the left-right movement of the passenger's gaze point has a stronger correlation with such outputs than the up-down movement, time series data of the left-right movement of the passenger's pupil is input to the neural network 350 as gaze point data. In addition, speed time series data of one axis of the forward direction of the three-axis speed of the mobile body 100 is input as speed data of the mobile body 100. This is because, when approaching a T-junction, the mobile body 100 slows down sufficiently to make a left or right turn, and therefore it is believed that the forward speed has a stronger correlation with the passenger's intended direction of travel.

[0036] Learning data is provided to a neural network 350 such as that shown in Figure 6 so that it can learn in advance, and the traveling direction estimation unit 35 can input gaze point data of the passenger of the moving body 100 and the odometry of the moving body 100 into the trained model, thereby estimating the traveling direction intended by the passenger.

[0037] When the moving object 100 approaches a branch point such as a T-junction while traveling, the route determination unit 33 may determine multiple local routes. In this case, the control unit 34 selects a route in the direction estimated by the traveling direction estimation unit 35 from the multiple local routes, and causes the moving object 100 to travel along the selected local route.

[0038] As described above, the control device 30B according to this embodiment allows the moving body 100 to travel in the direction intended by the passenger at a branch point such as a T-junction.

[0039] (Third embodiment) 7 is a block diagram of a control device for a moving body according to a third embodiment of the present invention. If there is an obstacle on the path of the moving body 100, there is a risk that the moving body 100 will collide with the obstacle. The control device 30C according to this embodiment is configured by adding a path prediction unit 36 ​​and a path evaluation unit 37 to the control device 30A according to the first embodiment, so that it is possible to take action to avoid danger if there is an obstacle on the predicted path of the moving body 100. Below, a repetitive explanation of matters already explained will be omitted, and only the characteristic features of this embodiment will be explained.

[0040] The path prediction unit 36 ​​predicts the path of the moving body 100 from the odometry of the moving body 100. Specifically, the path prediction unit 36 ​​can acquire a translational velocity v and an angular velocity ω as the odometry of the moving body 100 from the driving device 10. The position of the moving body 100 can be expressed by xy coordinate values ​​(x, y) and a rotation angle θ around the vertical axis, and the current position of the moving body 100 (t=i) can be expressed as Oi=(x i ,y i ,θ i ) T Then, the position of the moving object 100 at time i+1 is expressed as follows: where Δt is the current position of the moving object 100. i The next predicted position is O i+1 is the time interval until TIFF2026007432000002.tif18128 Based on the above formula, the path prediction unit 36 ​​can predict the path of the moving object 100.

[0041] The course evaluation unit 37 generates a bird's-eye view from the omnidirectional image. FIG. 8 is a diagram illustrating a method for generating a bird's-eye view from an omnidirectional image. The omnidirectional image is converted into a spherical image, and the spherical image is then converted into a perspective image, which is then generated as a bird's-eye view. FIG. 9 is a diagram illustrating an example of an omnidirectional image and a bird's-eye view generated from the omnidirectional image.

[0042] The path evaluation unit 37 further maps the predicted path of the moving body 100 onto a bird's-eye view map to evaluate the safety of the predicted path of the moving body 100. FIG. 10 is a diagram showing an example of a bird's-eye view map onto which the predicted path of the moving body is mapped. In the case of FIG. 10(a), the path evaluation unit 37 identifies that there is an obstacle on the predicted path of the moving body 100 and evaluates that the predicted path is dangerous. On the other hand, in the case of FIG. 10(b), the path evaluation unit 37 identifies that there is no obstacle on the predicted path of the moving body 100 and evaluates that the predicted path is safe.

[0043] If there is a danger on the predicted path of the moving body 100, the control unit 34 performs danger avoidance control such as issuing an alarm or forcibly stopping the moving body 100.

[0044] As described above, the control device 30C according to this embodiment can prevent the moving body 100 from colliding with an obstacle or the like on the local route.

[0045] <<Variations>> The moving body 100 may be a senior car, an electric cart, etc., other than an electric wheelchair. Furthermore, the moving body 100 does not need to be ridden by a person, and may be an unmanned moving robot.

[0046] The omnidirectional camera 20 does not need to capture a 360-degree field of view, but rather must be able to capture at least a 180-degree field of view in front of it. In that sense, it is also possible to use a smartphone with a fisheye lens attached as the omnidirectional camera 20, rather than a dedicated device.

[0047] The control devices 30A, 30B, and 30C may not be built into the moving body 100, but may be provided externally, for example, in the cloud, and may control the drive device 10 while communicating with the moving body 100 wirelessly.

[0048] As described above, the embodiments have been described as examples of the technology of the present invention. For this purpose, the accompanying drawings and detailed description have been provided. Therefore, the components described in the accompanying drawings and detailed description may include not only components essential for solving the problem, but also components that are not essential for solving the problem in order to exemplify the above technology. Therefore, the fact that these non-essential components are described in the accompanying drawings or detailed description should not be interpreted as immediately indicating that these non-essential components are essential. Furthermore, because the above-described embodiments are intended to exemplify the technology of the present invention, various modifications, substitutions, additions, omissions, etc. may be made within the scope of the claims or their equivalents. [Explanation of symbols]

[0049] 100 Mobile 30A, 30B, 30C control device 31 Image analysis unit 310 Multitask UniFuse (pre-trained model) 32 Pixel histogram calculation section 33 Route determination unit 34 Control Unit 35 Course direction estimation unit 350 Neural Networks (Pre-trained Models) 36 Career Prediction Department 37 Career Assessment Department

Claims

1. A control device for a moving body that causes the moving body to travel autonomously, an image analysis unit that performs semantic segmentation on an omnidirectional image captured in all directions of the moving object to identify a travelable area of ​​the moving object and estimates a distance from a current position of the moving object to an arbitrary position in the omnidirectional image; a pixel histogram calculation unit that calculates a pixel histogram representing a distribution of the number of pixels in the vertical direction of the image of the drivable area; a route determination unit that determines a local route from the current position of the moving object to a target point at a predetermined distance on a straight line that passes through the current position of the moving object and a position corresponding to a peak point of the pixel histogram in the omnidirectional image; a control unit that causes the moving body to travel along the local path; A control device for a moving body.

2. 2. The control device for a moving body according to claim 1, wherein the image analysis unit inputs an omnidirectional image of the moving body and a cube map image thereof into a trained model that has been trained in advance to output a segmentation map and a depth map from an input omnidirectional image and a cube map image, and simultaneously performs semantic segmentation and distance estimation on the omnidirectional image.

3. a traveling direction estimation unit that estimates an intended traveling direction of a passenger of the moving body from gaze point data of the passenger of the moving body and odometry of the moving body; The control device for a mobile body according to claim 1 or 2, wherein the control unit, when there are a plurality of local routes, selects a local route in the estimated direction and causes the mobile body to travel.

4. 4. The control device of a moving body as described in claim 3, wherein the traveling direction estimation unit inputs the gaze point data of the occupant of the moving body and the odometry of the moving body into a trained model that has been trained in advance to output the traveling direction intended by the occupant from the input gaze point data of the occupant and the odometry of the moving body, and estimates the traveling direction intended by the occupant.

5. a path prediction unit that predicts a path of the moving object from the odometry of the moving object; a path evaluation unit that generates a bird's-eye view from the omnidirectional image, maps a predicted path of the moving object on the bird's-eye view, and evaluates the safety of the predicted path of the moving object, The control device for a moving body according to claim 1 or 2, wherein the control unit performs danger avoidance control when there is a danger on the predicted course of the moving body.

6. a path prediction unit that predicts a path of the moving object from the odometry of the moving object; a path evaluation unit that generates a bird's-eye view from the omnidirectional image, maps a predicted path of the moving object on the bird's-eye view, and evaluates the safety of the predicted path of the moving object, The control device for a moving body according to claim 3 , wherein the control unit performs danger avoidance control when there is a danger on the predicted course of the moving body.

7. a path prediction unit that predicts a path of the moving object from the odometry of the moving object; a path evaluation unit that generates a bird's-eye view from the omnidirectional image, maps a predicted path of the moving object on the bird's-eye view, and evaluates the safety of the predicted path of the moving object, The control device for a moving body according to claim 4 , wherein the control unit performs danger avoidance control when there is a danger on the predicted course of the moving body.

8. A method for controlling a moving body that causes the moving body to travel autonomously, comprising: an image analysis unit performing semantic segmentation on an omnidirectional image obtained by capturing an image of the moving object in all directions to identify a travelable area of ​​the moving object; a step in which the image analysis unit estimates a distance from a current position of the moving object to an arbitrary position in the omnidirectional image; a pixel histogram calculation unit calculating a pixel histogram representing a distribution of the number of pixels in the vertical direction of an image of the drivable area; a route determination unit determining a local route from the current position of the moving object to a target point at a predetermined distance on a straight line passing through the current position of the moving object and a position corresponding to a peak point of the pixel histogram in the omnidirectional image; a step in which a control unit causes the moving body to travel along the local path; A method for controlling a moving object.

9. a step in which a course direction determination unit estimates a traveling direction intended by a passenger of the moving body from gaze point data of the passenger of the moving body and odometry of the moving body; The method for controlling a moving body according to claim 8 , further comprising a step in which the control unit, when there are a plurality of local routes, selects a local route in the estimated direction and causes the moving body to travel.

10. a course prediction unit predicting a course of the moving object from odometry of the moving object; a path evaluation unit generating a bird's-eye view from the omnidirectional image, mapping a predicted path of the moving object on the bird's-eye view, and evaluating the safety of the predicted path of the moving object; 10. The method for controlling a moving body according to claim 8, further comprising the step of: the control unit performing danger avoidance control when there is a danger on the predicted course of the moving body.

Images