JP2020112934A - Moving body, moving body control method, and computer program - Google Patents

Abstract

To provide a moving body, a moving body control method, and a computer program which can quickly recognize an avoidance object for which the execution of an avoidance operation is required, in a direction of movement to execute the avoidance operation.SOLUTION: The moving body includes: image acquisition means which photographs in a prescribed direction relating to a direction of movement of the moving body to capture an image; recognition means which recognizes an avoidance object in the image; and avoiding means which executes an avoidance operation on the basis of a recognition result of the recognition means. The recognition means is configured to execute overall recognition of taking the entire image as a recognition object and partial recognition of taking a part of the image in the direction of movement of the moving body or a lower part of the image in the direction of movement.SELECTED DRAWING: Figure 7

Description

The present invention relates to a moving body that detects an avoidance target such as an obstacle, a method for controlling the moving body, and a computer program.

Conventionally, a technique has been proposed in which an unmanned vehicle is automatically driven while detecting a marker arranged in the vicinity of a route in advance (for example, Patent Document 1). In addition, in such an unmanned vehicle, a technique of performing an avoidance operation such as stopping when the outside is photographed and a person is recognized in the image is performed.

Japanese Patent No. 5046310

For an unmanned vehicle, another vehicle, a human being, or the like is an object that needs to avoid a collision (hereinafter referred to as an "avoidance object"). However, even if a person is recognized in the image, avoidance action is not necessary if it is outside the route of the unmanned vehicle. Even if a person is present in the route of the unmanned vehicle, if the distance from the unmanned vehicle is long, the size of the person in the image may be small and the person may not be recognized as a person.

The present invention has attempted to solve such a problem, and a moving body and a moving body capable of promptly recognizing an avoidance target that needs to perform an avoidance operation and performing the avoidance operation in relation to a traveling direction. It is an object of the present invention to provide a control method and a computer program.

The first invention, by capturing a predetermined direction related to the traveling direction, an image acquisition unit that acquires an image, a recognition unit that recognizes an avoidance target in the image, and a recognition result of the recognition unit, Avoiding means for performing an avoiding operation, wherein the recognizing means targets the entire recognition of the entire image as a target of the recognition, and the target in the traveling direction or a portion below the traveling direction in the image. The moving object is configured to perform partial recognition.

According to the configuration of the first invention, the recognizing unit can perform the overall recognition and the partial recognition. Therefore, when the avoidance target is recognized in the overall recognition, the avoidance operation is performed and the avoidance target is detected in the partial recognition. Even when it is recognized, the avoidance operation can be performed. Here, the partial recognition does not target all the parts of the overall recognition, but a part in the traveling direction or a part below the moving direction. For this reason, in the shortest case, an appropriate avoidance operation can be performed by performing the entire recognition once and the partial recognition once. In particular, when the recognizing means uses a neural network, the avoidance target can be recognized with high accuracy and high speed by using both the overall recognition and the partial recognition targeting an image in a specific direction. Accordingly, it is possible to quickly recognize the avoidance target that needs to perform the avoidance operation in relation to the traveling direction, and to implement the avoidance operation.

A second invention is the moving body according to the configuration of the first invention, wherein the avoidance means is configured to perform the avoidance operation when the avoidance target recognized by the recognition means satisfies a predetermined condition. Is.

According to the configuration of the second invention, the avoidance unit does not always perform the avoidance operation when the avoidance target is recognized in the acquired image, but performs the avoidance operation only when the predetermined condition is satisfied. For example, when the avoidance target is recognized at a position that is completely deviated from the traveling direction, it is possible to avoid an unnecessary avoidance action such as not performing the avoidance action.

In a third aspect based on the configuration of the second aspect, the avoidance means refers to the result of the overall recognition and/or the result of the partial recognition, and a distance between the moving body and the avoidance target is a predetermined value. The moving body is configured to perform the avoiding operation when the predetermined condition of being in the distance range is satisfied.

In a fourth aspect based on the configuration of the second aspect, the avoidance means refers to a result of a plurality of times of the overall recognition and/or the partial recognition, and the avoidance target approaches the traveling direction or the downward direction. The moving body is configured to perform the avoiding operation when the predetermined condition that it is determined that the moving body is present is satisfied.

A fifth aspect of the present invention is the configuration of the second aspect, wherein the avoidance means is configured such that, based on a result of the partial recognition and a speed of the moving body, the moving body is associated with the avoidance target within a predetermined time. The moving body is configured to perform the avoiding operation when it is determined that the position is reached.

A sixth invention is an image acquisition step in which a moving body captures an image by capturing a predetermined direction related to a traveling direction of the moving body, a recognition step of recognizing an avoidance target in the image, and the recognition. An avoidance step of performing an avoidance operation based on the recognition result in the step, and in the recognition step, a whole recognition step in which the whole of the image is a target of the recognition, and a progress of the moving body in the image A method for controlling a moving body, which carries out a partial recognition step targeting a lower portion in a traveling direction or a traveling direction.

A seventh invention is a computer that controls a moving body, an image acquiring unit that captures an image by capturing a predetermined direction associated with the traveling direction of the moving body, a recognition unit that recognizes an avoidance target in the image, It is a computer program for functioning as an avoidance means for performing an avoidance operation based on a recognition result of the recognition means, wherein the recognition means is a whole recognition in which the whole of the image is the target of the recognition, The computer program is configured to carry out a partial recognition targeting a traveling direction of the moving body or a portion below the moving direction in the image.

ADVANTAGE OF THE INVENTION According to this invention, the avoidance target which needs to perform an avoidance operation|movement can be recognized quickly regarding an advancing direction, and an avoidance operation can be implemented.

It is a schematic diagram showing an operation of a mobile concerning a first embodiment of the present invention. It is a schematic diagram showing the composition of a mobile. It is a schematic diagram showing the functional composition of a mobile. It is a conceptual diagram for explaining a neural network. It is a schematic flow chart for explaining machine learning. It is a conceptual diagram for demonstrating the whole recognition. It is a conceptual diagram for demonstrating a whole image and a partial image. It is a conceptual diagram for demonstrating partial recognition. It is a flow chart which shows operation of a mobile. It is a figure which shows the modification of 1st embodiment. It is a flow chart which shows operation of a mobile. It is a schematic diagram showing an operation of a mobile concerning a second embodiment of the present invention. It is the schematic which shows the effect|action of a mobile body. It is a schematic diagram showing the composition of a mobile. It is a schematic diagram showing the functional composition of a mobile. It is a flow chart which shows operation of a mobile.

Hereinafter, modes for carrying out the present invention (hereinafter, embodiments) will be described in detail. In the following description, the same components are designated by the same reference numerals, and the description thereof will be omitted or simplified. It should be noted that description of configurations that can be appropriately performed by those skilled in the art will be omitted, and only the basic configuration of the present invention will be described. The essence of the present invention is that when a moving body recognizes an avoidance target using artificial intelligence, it performs recognition processing (entire recognition) performed on the entire acquired image and recognition processing (partial recognition) performed on a part of the image. Together, they are intended to improve the accuracy and speed of recognition processing by artificial intelligence.

<First embodiment>
The unmanned vehicle 1 (hereinafter, referred to as "unmanned aerial vehicle 1") illustrated in FIG. 1 is an example of a moving body. The unmanned aerial vehicle 1 is capable of autonomous movement along a planned route (hereinafter referred to as “scheduled route”), and operates by a radio signal from a control device (also called “propo”) that controls the unmanned aerial vehicle 1. It is supposed to be implemented. Note that, unlike the present embodiment, the moving body is not limited to an unmanned moving body such as an unmanned vehicle, but may be a manned moving body such as a vehicle or an air vehicle such as a manned heavy machine operated by a human being. Good.

Unmanned aerial vehicle 1, a tractor used in the field, is being tested on road 102 in front of university building 100. Since the unmanned aerial vehicle 1 travels on the road 102 in the direction of the arrow X1, there is no possibility of the unmanned aerial vehicle 1 colliding when there are no obstacles such as humans or other vehicles on or near the road 102. On the other hand, if there is a person on the road 102 within a predetermined distance range, or if there is a person who is near the road 102 without being on the road 102, a collision occurs. There is a possibility of doing so, so it is designed to avoid this. In the present specification, a target to be avoided by the unmanned aerial vehicle 1, such as a human being or another vehicle, is referred to as an “avoidance target”. Hereinafter, it is assumed that the “avoidance target” is a human. The traveling direction of the drone 1 is the traveling direction. In the present embodiment, the predetermined direction related to the traveling direction is not the direction of the sky but the direction centered on the planned route such as the road 102.

As shown in FIG. 2, the drone 1 has a vehicle body 10. Tires are arranged in the vehicle body 10. A motor is arranged in the vehicle body 10, and the unmanned aerial vehicle 1 can run by rotating the rear wheels by the rotation of the motor. The traveling direction of the unmanned aerial vehicle 1 can be implemented by rotating the front wheels.

In the vehicle body 10, electric power is supplied to a computer that controls each part of the unmanned aerial vehicle 1, an autonomous traveling device, a wireless communication device, a positioning device that uses a GPS (Global Positioning System), an inertial sensor, a barometric sensor, and each part of the unmanned aerial vehicle 1. The battery etc. for supplying are stored. Further, the vehicle body 10 is provided with an antenna 12 for wireless communication and an antenna 14 for receiving positioning radio waves from a navigation satellite.

Further, the vehicle body 10 is provided with a camera 16 for photographing the front and acquiring an image.

FIG. 3 is a diagram showing a functional configuration of the unmanned aerial vehicle 1. The unmanned aerial vehicle 1 includes a CPU (Central Processing Unit) 50, a storage unit 52, a wireless communication unit 54, a satellite positioning unit 56, an inertial sensor unit 58, a drive control unit 60, an image processing unit 62, and a power supply unit 64.

The unmanned aerial vehicle 1 can communicate with the outside through the wireless communication unit 54. The unmanned aerial vehicle 1 receives, via the wireless communication unit 54, an instruction such as starting from a control device (propo) 70 operated by a user.

The unmanned aerial vehicle 1 can measure the position of the unmanned aerial vehicle 1 itself by the satellite positioning unit 56. The satellite positioning unit 56 basically receives positioning radio waves from four or more navigation satellites and measures the position of the unmanned aerial vehicle 1. The satellite positioning unit 56 is an example of a positioning unit that measures the current position. The inertial sensor unit 58 detects the movement of the unmanned aerial vehicle 1 by using, for example, an acceleration sensor and a gyro sensor. The position information of the unmanned aerial vehicle 1 itself is used not only for determining the movement route of the unmanned aerial vehicle 1 and for autonomous movement, but also for associating the image data captured by the camera 16 with the coordinates (position).

The drive control unit 60 controls the rotation of the motor connected to the rear wheels of the tire and the rotation of the front wheels to control the traveling speed and the traveling direction.

The drone 1 operates the camera 16 (see FIG. 2) by the image processing unit 62 to acquire an external image, and processes the acquired shooting data. The external image is mainly an image in the traveling direction of the drone 1.

The power supply unit 64 is, for example, a replaceable rechargeable battery, and supplies power to each unit of the drone 1.

The storage unit 52 stores the following programs in addition to various data and programs necessary for autonomous movement such as data indicating a movement plan for autonomous movement from a starting point to a target position.

An image capturing program, an image recognition program, and an avoidance program are stored in the storage unit 52. The CPU 50 and the image capturing program are an example of an image acquisition unit. The CPU 50 and the image recognition program are examples of recognition means. The CPU 50 and the avoidance program are an example of avoidance means.

The unmanned aerial vehicle 1 captures an image by capturing an image of the outside including the planned route along which the unmanned aerial vehicle 1 travels according to the image capturing program. The direction of the optical axis of the camera 16 is fixed to the front, which is the traveling direction of the drone 1.

The drone 1 recognizes the avoidance target in the image by the image recognition program. The unmanned aerial vehicle 1 stores, for example, the neural network shown in FIG. 4 in the storage unit 52, and the avoidance target is recognized by the neural network.

An outline of the learning method (machine learning process) will be described with reference to FIGS. 4 and 5. When a large number of image data (Image data) are input to the model of the neural network of the computer (see FIG. 4) as learning data (step S1 of FIG. 5), learning is performed by the algorithm of the neural network (step S2). The learning data is data in which various types of human beings are reflected.

The model of the neural network is, for example, the neural network conceptually shown in FIG. It should be noted that neural networks and deep learning are well known, and therefore will be only outlined in this specification. The neural network includes an input layer (Input layer), at least one hidden layer (Hidden layer), and an output layer (Output layer). Although only one hidden layer is shown in FIG. 4, there are actually a plurality of hidden layers. That is, the learning of the present embodiment is machine learning (deep learning) by a multilayer neural network (deep neural network, Deep Neural Network).

In machine learning, the output from the output layer is correct by adjusting the weights w1a of the input values input to the hidden layer and the output layer and the bias (threshold in each layer) B1 in the input layer, the hidden layer, and the output layer. To get closer to. As a result of learning, a neural network in which weights and biases are adjusted is generated as learning result data (step S3 in FIG. 5). By this learning, a neural network for accurately recognizing that the object in the image is a human is generated. The storage unit 52 stores the neural network generated in this way.

The unmanned aerial vehicle 1 uses the image recognition program to recognize the entire acquired image (hereinafter referred to as “entire image”) as a recognition target, and a specific part of the entire image (hereinafter referred to as “partial image”). .) is performed to perform partial recognition. The overall recognition and the partial recognition will be described below.

The unmanned aerial vehicle 1 performs image recognition with a resolution of horizontal Pxs pixels and vertical Pys pixels (see FIG. 6 and the like) (hereinafter, referred to as “recognition resolution”) in both overall recognition and partial recognition. Pxs and Pys are, for example, 416 pixels for Pxs and 208 pixels for Pys. That is, the target of image recognition is switched to the whole image and the partial image at the same recognition resolution.

Whole recognition is a recognition process for the whole image. For example, as shown in FIG. 6A, it is assumed that the pixels of the entire image D1 are horizontal Px1 pixels and vertical Py1 pixels. For example, Px1 is 3960 pixels and Py1 is 2048 pixels. In the overall recognition, the entire image D1 is processed with the above-described recognition resolution. Pxs is smaller than Px1, and Pys is smaller than Py1. As a result, humans 200A and 200B located near the unmanned aerial vehicle 1 can be recognized, for example, as shown by being surrounded by a rectangle in FIG. However, even if the human 200C located relatively far from the drone 1 is located on the planned route of the drone 1, since the size of the entire image is small and the features do not sufficiently appear in the image, Cannot be recognized as. The time required for overall recognition is t1. t1 is, for example, 1/30 second (s).

Partial recognition is a recognition process for partial images. The partial image is an image corresponding to a portion of the entire image in the traveling direction of the drone 1. In FIG. 6B, in the overall image D1, the planned route is shown as a position between the lines L1 and L2, and the traveling direction of the unmanned aerial vehicle 1 is shown by an arrow X1.

The whole image D1 can be divided into a plurality of partial images D1a to D1y, for example, as shown in FIG. Among the plurality of portions D1a to D1y, the traveling direction of the unmanned aerial vehicle 1 is the portions D1w, D1r, and D1m. However, the partial images D1w and D1r are relatively close to the unmanned aerial vehicle 1, and when a human is present in those parts, they should be recognizable in the overall recognition. Therefore, the target of partial recognition is the partial image D1m that is relatively far away. As shown in FIG. 7B, the drone 1 extracts the partial image D1m as a target of partial recognition.

As shown in FIG. 8A, the unmanned aerial vehicle 1 targets the partial image D1m having the horizontal Px2 pixels and the vertical Py2 pixels by the partial recognition with the above-described recognition resolution. Px2 is smaller than Px1 and Py2 is smaller than Py1. Therefore, in the whole image, the human 200C displayed relatively small in relation to the recognition resolution described above is displayed larger in the partial image than in the whole image. Therefore, even with the above-described recognition resolution, the drone 1 can recognize the human 200C, as shown in FIG. 8B. The time required for partial recognition is t1 as in full recognition. The unmanned aerial vehicle 1 alternately executes the entire recognition and the partial recognition by the image recognition program.

If there are k partial images in the entire image, the time required for partial recognition of all partial images is t1×k. However, a partial image that does not correspond to a portion of the unmanned aerial vehicle 1 in the traveling direction is a partial image of a portion that the unmanned aerial vehicle 1 will not pass in the future, and thus it is not necessary to perform image recognition. In this respect, according to the present embodiment, the partial image to be partially recognized is limited to the partial image in the traveling direction, so that necessary and sufficient image recognition can be performed in a short time.

The unmanned aerial vehicle 1 performs the avoidance operation by the avoidance program based on the recognition result by the recognition program. The unmanned aerial vehicle 1 is configured to perform the avoidance operation when the avoidance target recognized by the recognition program satisfies a predetermined condition.

For example, the unmanned aerial vehicle 1 refers to the result of the overall recognition, and when the distance between the unmanned aerial vehicle 1 and the avoidance target satisfies a predetermined condition of being within a predetermined distance range, performs the avoidance operation. The distance in the predetermined distance range is, for example, 5 meters (m).

The unmanned aerial vehicle 1 also refers to the result of the partial recognition, and when the distance between the unmanned aerial vehicle 1 and the avoidance target satisfies a predetermined condition of being within a predetermined distance range, performs the avoidance operation. The predetermined distance range is, for example, 10 meters (m).

Since the human height is within a predetermined range, for example, 140 cm (cm) to 200 cm (cm), the drone 1 can be defined by, for example, the number of vertical pixels of the human in the image. The approximate distance from the unmanned aerial vehicle 1 can be determined.

The operation of the unmanned aerial vehicle 1 will be described below with reference to the flowchart of FIG. When the drone 1 receives the start signal from the control device 70, it starts (step ST1 in FIG. 9), continuously acquires external images (step ST2), measures the current position, and performs a predetermined schedule. The route is advanced (step ST3). Step ST2 is an example of an image acquisition step.

The unmanned aerial vehicle 1 image-recognizes the entire image by the neural network (step ST4), and when recognizing the avoidance target in the image (step ST5), determines whether or not the avoidance distance is a distance that needs to be avoided. (Step ST6) If the distance is the avoidance distance, the progress is stopped as the avoidance operation.

When the avoidance target is not recognized in the overall recognition in step ST5, or when it is determined that the avoidance target is not the avoidance distance in step ST6, a partial image in the traveling direction is image-recognized by the neural network (step ST7). When it is judged that the object is an avoidance target (step ST8), it is judged whether or not the distance is the avoidance distance (step ST9). On the other hand, if the avoidance target is not recognized in the partial recognition in step ST8, or if the avoidance target is recognized and it is determined that the distance does not require avoidance, it is determined whether the mission is completed (step ST10), if it is determined that the mission is not completed, the above steps ST2 to ST9 are repeated. Steps ST4 and ST7 are examples of recognition steps. In step ST4, whole recognition is performed, and in step ST7, partial recognition is performed.

<Modification>
Next, a modified example of the first embodiment will be described with reference to FIGS. 10 and 11. In a modification, the drone 1 refers to the results of the overall recognition and/or the partial recognition a plurality of times by the avoidance program, and when the predetermined condition that it is determined that the avoidance target is approaching in the traveling direction is satisfied. , Is configured to perform the avoidance operation.

For example, as shown in FIG. 10A, it is assumed that the humans 200A and 200B and their positions are recognized in the image recognition of the entire image D1. Although the humans 200A and 200B are actually moving in the direction of the arrow Y1, it is not possible to determine the moving state of the humans 200A and 200B with a single image recognition. The drone 1 recognizes the humans 200A and 200B and their positions in the image recognition of the entire image D2 at a slightly later time, and compares them with their positions in the entire image D1 to detect the humans 200A and 200B. The movement state can be determined. Since the human 200A is moving in the arrow Y1 direction, it is moving in a direction away from the movement route of the unmanned aerial vehicle 1, and the unmanned aerial vehicle 1 does not need to perform the avoidance operation. On the other hand, since the human 200B is moving in the direction of the arrow Y1, it is moving in the direction approaching the movement route of the unmanned aerial vehicle 1, and the unmanned aerial vehicle 1 needs to perform the avoidance operation.

The unmanned aerial vehicle 1 also performs the above-described processing in partial recognition. That is, when the whole image D1 is acquired, the whole recognition and the partial recognition are performed, the position of the recognized human is stored, and the whole image D2 at a slightly later time is acquired to perform the whole recognition and the partial recognition. Then, by recognizing the human being and its position, the moving state of the human being is judged.

Hereinafter, the operation of the unmanned aerial vehicle 1 in the modified example will be described with reference to the flowchart in FIG. When the unmanned aerial vehicle 1 receives the start signal from the control device 70, the unmanned aerial vehicle 1 starts (step ST1 in FIG. 11), continuously acquires external images (step ST2), measures the current position, and performs a predetermined schedule. The route is advanced (step ST3).

The unmanned aerial vehicle 1 recognizes the entire image by the neural network (step ST4), recognizes the avoidance target in the image (step ST5), and determines that the avoidance distance is a distance that needs to be avoided and the approaching distance is approached. It is determined whether or not (step ST6A), and if the distance is the avoidance distance and the vehicle is approaching, the progress is stopped as an avoidance operation.

If the avoidance target is not recognized in the overall recognition in step ST5, or if the avoidance target is recognized but is not the avoidance distance or is not approaching in step ST6A, a partial image in the traveling direction is detected by the neural network. When the image is recognized (step ST7) and it is determined that the target is an avoidance target (step ST8), it is determined whether the avoidance distance is approaching or not (step ST9A) and the avoidance distance is approaching. If so, the progress is stopped as an avoidance operation. On the other hand, if the avoidance target is not recognized in the partial recognition in step ST7, or if the avoidance target is recognized but the avoidance distance is not reached or it is determined that the avoidance target is not approaching, whether or not the mission is completed is determined. If it is determined (step ST10) that the mission is not completed, the above steps ST2 to ST9A are repeated.

<Second embodiment>
Next, a second embodiment will be described with reference to FIGS. 12 to 16. It should be noted that description of items that are common to the first embodiment will be omitted, and the description will focus on the parts that differ from the first embodiment.

In the second embodiment, the moving body is an unmanned aerial vehicle 20 (hereinafter referred to as "unmanned aerial vehicle 20"). The unmanned aerial vehicle 20 needs to avoid flying over humans and moving vehicles, primarily to prevent harm to humans. In the unmanned aerial vehicle 20, the avoidance target is a person or a car below. The predetermined direction related to the traveling direction is a direction centered on the lower side of the traveling direction, and more specifically, a direction centered on the diagonally lower side. Then, the unmanned aerial vehicle 20 targets the image directly below the region in which the aircraft will fly in the future for partial recognition. The basic functions and control program of the unmanned aerial vehicle 20 are the same as those in the first embodiment.

FIG. 12 is a schematic plan view of the unmanned aerial vehicle 20 and a region in which the unmanned aerial vehicle 20 flies, as seen from above. The unmanned aerial vehicle 20 flies along the planned route indicated by the arrow X2 in the region where the roads 120A and 120B and the buildings 122A to 122D exist. An automobile 130A is traveling on the road 120A, and its traveling route intersects with the planned route of the unmanned aerial vehicle 20.

As shown in FIG. 12, the unmanned aerial vehicle 20 continuously acquires the entire image D10 below the traveling direction. Then, the entire recognition is performed on the entire image D10, and the partial recognition is performed on a specific partial image D10m of the entire image D10. The partial image D10m is a portion directly below the flight path of the drone 20. That is, the unmanned aerial vehicle 20 will fly right above the region corresponding to the partial image D10m in the future.

As shown in FIG. 13, the unmanned aerial vehicle 20 continuously acquires the entire images D10 to Dx, performs the entire recognition, and performs the partial recognition of the partial images D10m to Dxm. As a result, partial recognition can be performed on the portion directly below the area where the unmanned aerial vehicle 20 will fly.

Hereinafter, the configuration of the unmanned aerial vehicle 20 will be described. As shown in FIG. 14, the drone 20 has a housing 22. The housing 22 is provided with an autonomous flight device including a computer, a wireless communication device, a positioning device that uses positioning radio waves from a navigation satellite system such as GPS (Global Positioning System), and a battery. The autonomous flight device includes an inertial sensor, an electronic compass, and a barometric pressure sensor. Further, a camera 34 is arranged in the housing 22 via a fixing device 32.

The unmanned aerial vehicle 20 acquires an external image with the camera 34. The fixing device 32 is a triaxial fixing device (so-called gimbal) that can minimize the blurring of the image captured by the camera 34 and control the optical axis of the camera 34 in any direction.

A round bar-shaped arm 24 is connected to the housing 22. A motor 26 is connected to each arm 24, and a propeller 28 is connected to each motor 26. Each motor 26 is a DC motor (brushless DC motor). The rotation speed of each motor 26 is controlled by a motor driver (not shown). The motor drivers are independently controlled by the autonomous flight device. The motors 26 are independently controlled so that the drone 1 can be freely moved in the vertical and horizontal directions, stopped in the air (hovering), and controlled in attitude. A protection frame 30 is also arranged on the arm 24.

FIG. 15 is a diagram showing a functional configuration of the unmanned aerial vehicle 20. The unmanned aerial vehicle 20 includes a CPU 150, a storage unit 152, a wireless communication unit 154, a satellite positioning unit 156, an inertial sensor unit 158, a drive control unit 160, an image processing unit 162, and a power supply unit 164.

The unmanned aerial vehicle 20 can communicate with the radio 70 by the wireless communication unit 154. The wireless communication unit 154 of the unmanned aerial vehicle 20 receives an instruction such as a start from the transmitter 70.

The unmanned aerial vehicle 20 can measure the position of the unmanned aerial vehicle 20 itself by the satellite positioning unit 156. The satellite positioning unit 156 basically receives positioning radio waves from four or more navigation satellites and measures the position of the unmanned aerial vehicle 20. The satellite positioning unit 156 is an example of a positioning unit that measures the current position. The inertial sensor unit 158 detects the movement of the unmanned aerial vehicle 20 based on the outputs from the acceleration sensor and the gyro sensor. The position information of the unmanned aerial vehicle 20 itself is used for determining the movement route of the unmanned aerial vehicle 20 and autonomous movement.

The drive control unit 160 controls the unmanned aerial vehicle 20 to control the rotation of the propeller 28 connected to each motor 26, and to control the posture such as vertical and horizontal movement, air stop, and tilt.

The drone 1 operates the camera 34 by the image processing unit 162 to acquire an external image, and processes the acquired shooting data.

The power supply unit 164 is, for example, a replaceable rechargeable battery and supplies power to each unit of the unmanned aerial vehicle 20.

The storage unit 152 stores various data and programs necessary for autonomous movement, such as data indicating a movement plan for autonomous movement from a starting point to a destination position, an image capturing program, an image, as in the first embodiment. A recognition program and an avoidance program are stored.

The operation of the unmanned aerial vehicle 20 will be described below with reference to the flowchart of FIG. When the unmanned aerial vehicle 20 receives the start signal from the transmitter 70, the unmanned aerial vehicle 20 starts to move (step ST1 in FIG. 16) and continuously acquires images of a region centered on a lower portion in the traveling direction (step ST2A). The vehicle moves along a predetermined planned route while positioning the position (step ST3).

The unmanned aerial vehicle 20 recognizes the entire image by the neural network (step ST4), and when the avoidance target is recognized in the image (step ST5), the avoidance distance is a distance that needs to be avoided, and is approached. It is determined whether or not there is (step ST6A), and if it is the avoidance distance and is approaching, as the avoidance operation, for example, the progress is stopped. Here, the “stop of the progress” is an aerial stop (hovering).

If the avoidance target is not recognized in the overall recognition in step ST5, or if the avoidance target is recognized but is not within the avoidance distance or is not approaching in step ST6A, the neural network determines a lower portion in the traveling direction. When the image is recognized as an image (step ST7) and it is determined that it is an avoidance target (step ST8), it is an avoidance distance and it is determined whether or not it is approaching (step ST9A), an avoidance distance, and If it is approaching, the progress is stopped as an avoidance operation. On the other hand, if the avoidance target is not recognized in the partial recognition in step ST7, or if the avoidance target is recognized but the avoidance distance is not reached or it is determined that the avoidance target is not approaching, whether or not the mission is completed is determined. If it is determined (step ST10) that the mission is not completed, the above steps ST2 to ST9A are repeated.

It should be noted that the present invention is not limited to the above-described embodiments, and modifications, improvements, etc. within the scope of achieving the object of the present invention are included in the present invention.

1 Unmanned vehicle (unmanned aerial vehicle)
10 vehicle body 12 antenna 14 antenna 16 camera 20 unmanned aerial vehicle (unmanned aerial vehicle)
22 housing 34 camera

Claims (7)

An image acquisition unit that captures an image by capturing a predetermined direction related to the traveling direction,
Recognition means for recognizing an avoidance target in the image,
And an avoidance unit that performs an avoidance operation based on the recognition result of the recognition unit,
The recognizing unit is configured to perform a total recognition targeting the entire image, and a partial recognition targeting the traveling direction or a portion below the traveling direction in the image, Mobile. The avoidance means is
When the avoidance target recognized by the recognition means satisfies a predetermined condition, the avoidance operation is performed.
The moving body according to claim 1. The avoidance means refers to the result of the overall recognition and/or the result of the partial recognition, and when the distance between the moving body and the avoidance target satisfies the predetermined condition of being a predetermined distance range, The moving body according to claim 2, which is configured to perform an avoidance operation. The avoidance means refers to a result of the entire recognition and/or the partial recognition of a plurality of times, and when the predetermined condition that it is determined that the avoidance target approaches the traveling direction or the downward direction, The moving body according to claim 2, wherein the moving body is configured to perform the avoidance operation. The avoiding means performs the avoiding operation when it is determined that the moving body reaches a predetermined position related to the avoidance target within a predetermined time based on the result of the partial recognition and the speed of the moving body. The moving body according to claim 2, which is configured to: The moving body
An image capturing step of capturing an image by capturing a predetermined direction related to the traveling direction of the moving body,
A recognition step of recognizing an avoidance target in the image;
Based on the recognition result in the recognition step, performing an avoidance step of performing an avoidance operation,
In the recognition step, a moving body which carries out a whole recognition step of targeting the entire image as the recognition target and a partial recognition step of targeting a traveling direction of the moving body or a lower portion in the traveling direction of the moving body in the image, Control method. A computer that controls the mobile
An image acquisition unit that captures an image by capturing a predetermined direction related to the traveling direction of the moving body,
Recognition means for recognizing an avoidance target in the image,
Avoidance means for performing an avoidance operation based on the recognition result of the recognition means,
A computer program for functioning as
The recognizing means is configured to perform a total recognition targeting the entire image, and a partial recognition targeting a traveling direction of the moving body or a lower portion in the traveling direction of the moving body in the image. Is a computer program.