KR102497488B1 - Image recognition apparatus for adjusting recognition range according to driving speed of autonomous vehicle - Google Patents

Abstract

The present invention relates to an image recognition device of an autonomous vehicle capable of adjusting a recognition range depending on a driving velocity. The image recognition device of an autonomous vehicle according to the present invention comprises: a photography part having a lens with a certain angle of view and generating an image by photographing the front of the vehicle; and an image recognition part containing image recognition AI algorithm and performing image recognition for recognizing information on an object to be recognized by the autonomous vehicle from the image provided by the photography part, wherein the image is downscaled to have reference resolution and image recognition is performed if the driving velocity of the autonomous vehicle is smaller than reference velocity, and the image is cropped into a size corresponding to the reference resolution and image recognition is performed from the cropped image if the driving velocity is equal to or greater than the reference velocity.

Description

Image recognition apparatus for adjusting recognition range according to driving speed of autonomous vehicle

The present invention relates to an image recognition device for an autonomous vehicle capable of adjusting a recognition range according to driving speed, and more specifically, by adjusting a recognition range according to driving speed, real-time recognition is possible while minimizing recognition errors. It relates to an image recognition device for an autonomous vehicle capable of adjusting a recognition range according to the present invention.

An autonomous vehicle refers to a vehicle that moves to a destination by controlling the driving of the vehicle itself, rather than by manipulating the driving of the vehicle by a driver. In order to drive an autonomous vehicle, various types of sensors are required to monitor the surrounding environment on behalf of the driver, and the autonomous vehicle uses this to detect objects such as people, animals, other vehicles, and traffic lights (hereinafter referred to as “object information”). ) must be accurately recognized, and the vehicle must operate in the set state after recognition.

In particular, interest in self-driving vehicles that perform self-driving by recognizing object information from images captured through cameras is increasing, and recent studies related to autonomous vehicles accurately recognize object information from images captured through cameras. Research on what to do is being actively conducted.

In this case, the range to be recognized may vary according to the driving speed of the vehicle in the image captured by the camera. In the case of an image captured through a camera, recognition of object information up to a distance of 100 m or more in the front is more important than the surroundings when driving on a highway, and recognition of object information around the vehicle is more important than long distances when driving slowly.

Accordingly, a need for an image recognition device capable of differently determining a recognition range of object information corresponding to a driving speed of an autonomous vehicle has emerged.

Republic of Korea Patent Registration No. 10-2188269 (2020.12.02)

Accordingly, an object of the present invention is to provide an image recognition device for an autonomous vehicle capable of adjusting a recognition range according to driving speed, which can overcome the above conventional problems.

Another object of the present invention is to provide an image recognition device for an autonomous vehicle that selects lenses in an order in which the angle of view decreases as the driving speed of the autonomous vehicle increases and performs image recognition based on an image of the selected lens.

Another object of the present invention is to downscale or crop an image to a reference resolution so that real-time recognition is possible by preventing a decrease in image recognition speed without causing a recognition error in response to the driving speed of an autonomous vehicle. ) to provide an image recognition device for an autonomous vehicle that performs image recognition.

According to the embodiment of the present invention for achieving some of the above technical problems, the image recognition device for an autonomous vehicle according to the present invention has a different angle of view, respectively, and photographs the front of the vehicle to generate images, respectively. a photographing unit having a plurality of lenses; An image recognition AI algorithm is embedded, and object information to be recognized in the self-driving vehicle is recognized from the image provided through the photographing unit, and as the driving speed of the self-driving vehicle increases, the angle of view among the plurality of lenses increases. An image recognizing unit which selects lenses in order of decreasing size and recognizes the object information from an image generated from the selected lenses is provided.

The photographing unit has a first lens that has a first angle of view and generates a first image by capturing the front of the vehicle, and has a second angle of view that is smaller than the first angle of view and photographs the front of the vehicle to generate a first image. A second lens for generating two images is provided, and the image recognition unit selects the first image to recognize the object information when the driving speed of the self-driving vehicle is less than the reference speed, and the driving speed is equal to or less than the reference speed. If the speed exceeds the speed, the object information may be recognized by selecting the second image.

The first lens is a wide-angle lens, the second lens is a narrow-angle lens, and the reference speed may be 60 to 100 km/h.

The first angle of view may be 90 degrees to 180 degrees, and the second angle of view may be 0 to 60 degrees (not including 0 degrees).

The object information may include at least one of target road information, vehicle information, moving object information including people, obstacle information, traffic light information, sign information, and event information.

According to another embodiment of the present invention for achieving some of the above technical problems, an image recognition device for an autonomous vehicle according to the present invention has a lens having a certain angle of view and captures an image in front of the vehicle. a photographing unit to generate; An image recognition AI algorithm is embedded to perform image recognition for recognizing object information to be recognized in the self-driving vehicle from the image provided through the photographing unit, but when the driving speed of the self-driving vehicle is less than the reference speed Image recognition is performed by downscaling the image so that it has a standard resolution, and if the traveling speed is greater than or equal to the standard speed, the image is cropped to a size corresponding to the standard resolution. An image recognition unit that performs image recognition from the pinged image is provided.

The lens may be a wide-angle lens having an angle of view of 90 degrees to 180 degrees.

The reference resolution may be a resolution that can be recognized in real time without causing an image recognition error when the image recognition unit performs image recognition, and may be a preset resolution.

The reference resolution may be 640X480.

The reference speed may be 60 to 100 km/h.

The image recognizing unit, when the traveling speed is equal to or greater than the reference speed, performs downscaling at a rate corresponding to the traveling speed, crops the downscaled image to a size corresponding to the reference resolution, and crops the image. Image recognition can be performed from the pinged image.

According to another embodiment of the present invention for achieving some of the above technical problems, an image recognition device for an autonomous vehicle according to the present invention includes a photographing unit having a wide-angle lens and generating an image by photographing the front of the vehicle; An image recognition AI algorithm is embedded to perform image recognition for recognizing object information to be recognized in the self-driving vehicle from the image provided through the photographing unit, corresponding to the driving speed of the self-driving vehicle. An image recognizing unit for performing image recognition by performing at least one of downscaling ratio adjustment and cropping image size adjustment with respect to .

The image recognizing unit may perform cropping by making the size of the cropped image gradually decrease as the traveling speed increases.

The image recognizing unit may downscale the image before performing the cropping, perform image recognition on the cropped image by performing the cropping on the downscaled image.

The downscaling performed before performing the cropping may be performed at a fixed ratio regardless of the traveling speed, and the cropping performed on the downscaled image may be adjusted to a size corresponding to the traveling speed.

The downscaling performed before the cropping is performed at a rate corresponding to the traveling speed, and the cropping performed on the downscaled image may be performed at a fixed size regardless of the traveling speed.

The image recognizing unit performs image recognition using the entire image provided through the photographing unit when the traveling speed is less than the reference speed, and when the traveling speed is greater than or equal to the reference speed, the image is cropped as the traveling speed increases. An image recognizing unit may be provided to perform cropping by gradually decreasing the size of , and perform image recognition from the cropped image.

According to the present invention, it is possible to select lenses in an order in which the angle of view decreases as the driving speed of the autonomous vehicle increases, and perform image recognition based on the image of the selected lens. By adjusting the range and resolution, there is an advantage that real-time recognition is possible by preventing a decrease in image recognition speed without causing recognition errors.

1 is a schematic diagram of an image recognition device for an autonomous vehicle according to a first embodiment of the present invention;
Figure 2 shows an example of a first image and a second image captured through the photographing unit of Figure 1,
3 is a schematic diagram of an image recognition device for an autonomous vehicle according to a second embodiment of the present invention;
4 is a schematic diagram for calculating an image recognition range according to a lens constituting a camera and a distance;
Figure 5 shows the process of downscaling and cropping the image recognition unit of Figure 3,
6 shows examples of downscaled (b), cropped (c), and upscaled (d) images of an actual captured image (a);
7 is a schematic diagram of an image recognition device for an autonomous vehicle according to a third embodiment of the present invention;
8 is a schematic diagram illustrating a cropping process of the image recognition unit of FIG. 7 .

Hereinafter, preferred embodiments of the present invention will be described in detail with reference to the accompanying drawings, without any intention other than to provide a thorough understanding of the present invention to those skilled in the art to which the present invention pertains.

1 is a schematic diagram of an image recognition device for an autonomous vehicle according to a first embodiment of the present invention. FIG. 2 illustrates an example of a first image and a second image captured through the photographing unit of FIG. 1 .

As shown in FIGS. 1 and 2 , the image recognition device 100 for an autonomous vehicle according to the first embodiment of the present invention includes a photographing unit 110 and an image recognition unit 120 .

The photographing unit 110 includes a plurality of lenses 112 and 114 each having a different angle of view and generating images by photographing the front of the vehicle.

For example, the photographing unit 110 includes a first lens 112 that has a first angle of view and generates a first image by photographing the front of the vehicle, and a second angle of view smaller than the first angle of view. A second lens 114 that has a two angle of view and generates a second image by photographing the front of the vehicle may be obtained. In addition, a third lens having a third angle of view may be added, and a fourth lens having a fourth angle of view may be further added as needed.

Here, the first lens 112 is a wide-angle lens, the second lens 114 is a narrow-angle lens, the first angle of view is 90 degrees to 180 degrees, and the second angle of view is 0 to 60 degrees (0 degrees not included).

Here, a third lens having a third angle of view may be added in addition to the first lens 112 and the second lens 114, and a fourth lens having a fourth angle of view may be further added as needed. . In this case, the ranges of the first and second angles of view, which are the angles of view of the first lens 112 and the second lens 114, may be reduced.

In addition to the plurality of lenses 112 and 114, the photographing unit 110 may include all essential components well known to those skilled in the art that are necessary or advantageous for image capturing, such as an image sensor.

When the first lens 112 provided in the photographing unit 110 is a wide-angle lens and the second lens 114 is a narrow-angle lens, the first image captured by the wide-angle lens is wider than the long-distance image. Information on a short distance of the field of view is further included, and the second image captured by the narrow-angle lens further includes information on a long distance in front of the vehicle rather than information on a short distance. Accordingly, it is possible to perform image recognition by selecting the first image when more information on a short distance is needed and by selecting the second image when more information on a long distance is needed.

The image recognition unit 120 has a built-in image recognition AI algorithm, recognizes object information to be recognized in the self-driving vehicle from the image provided through the photographing unit 110, and drives the self-driving vehicle. As the speed increases, lenses are selected from among the plurality of lenses in order of decreasing angle of view, and the object information is recognized from an image generated from the selected lenses.

The image recognition unit 120 may be built into an electronic control unit (ECU) of an autonomous vehicle, and is equipped with an AI algorithm (deep learning algorithm, deep neural network algorithm, etc.) for image recognition to provide a deep learning learning model, etc. Recognition of object information in an image is performed through a learning process, and object information is recognized from an image provided from the photographing unit 110 .

The image recognizing unit 120 may include a memory (not shown) and an image processor (not shown) for storing images provided from the photographing unit 110. The image processor is included in the input image. It is possible to perform image processing for improving the recognition rate of object information.

And, the object information may include information about an object to be recognized in an autonomous vehicle for autonomous driving. The object information may include at least one of target road information to be recognized, vehicle information, moving object information including people, obstacle information, traffic light information, sign information, and event information. The moving object information may include all moving objects including people and animals, and the road information may include road surface information or lane information. In addition, the vehicle information may include information on various vehicles of two or more wheels, including two-wheeled vehicles such as electric quick boards, bicycles, and motorcycles. In addition, the object information may include target road information, vehicle information, moving object information including people, obstacle information, traffic light information, sign information, and location information or color information of event information. In addition, information known to require recognition from a photographed image for autonomous driving may be included. For example, abnormal object information may be additionally included. Abnormal object information may include, for example, objects such as people or animals performing abnormal behaviors other than generally known normal behaviors, such as a person sitting, a person lying down, a carcass of an animal, and a person using an umbrella. .

When the driving speed of the self-driving vehicle is less than the reference speed, the image recognition unit 120 selects the first image as shown in (a) of FIG. 2 and recognizes the object information through image recognition, and When the traveling speed is equal to or greater than the reference speed, control is performed to recognize the object information by selecting the second image as shown in FIG. 2(b). Here, the reference speed may be 60 to 100 km/h.

In general, when a vehicle is running at a low speed less than the reference speed, recognition of object information located in a short distance becomes more important than a distant object of 60 to 100 m or more. Accordingly, it is further required to recognize object information from the first image photographed with the wide-angle lens during low-speed driving. Here, in the case of bad weather, since it is inevitable to operate at low speed, object information is recognized from the first image by applying mutatis mutandis to the case of driving at low speed.

However, when the vehicle is operating at a high speed equal to or higher than the reference speed, recognition of object information located at a distance of 60 to 100 m or more is more required than short-distance information near the vehicle. Accordingly, it is further required to recognize object information from the second image captured with a narrow angle lens during high-speed driving. In particular, when driving on a highway, recognition of object information located at a distance of 60 to 100 m or more must be performed, and notification or braking is possible only when the recognition is fast. will come to recognize This is because images captured with a narrow-angle lens are more efficient in terms of recognition speed and recognition range than images captured with a wide-angle lens.

As described above, according to the first embodiment of the present invention, in recognizing an image photographed in front of an autonomous vehicle, an image of a wide-angle lens or a narrow-angle lens is selected and recognized according to the driving speed of the vehicle. , there is an advantage in that the efficiency of recognition can be increased because it is possible to select a necessary recognition range while solving the problem of a decrease in recognition speed according to a high-resolution image.

3 is a schematic diagram of an image recognition device for an autonomous vehicle according to a second embodiment of the present invention, and FIG. 4 is a schematic diagram for calculating an image recognition range according to a distance from a lens constituting a camera.

As shown in FIG. 3 , the image recognition device 200 for an autonomous vehicle according to the second embodiment of the present invention includes a photographing unit 210 and an image recognition unit 220 .

The photographing unit 210 includes at least one lens 212 each having a predetermined angle of view and generating images by photographing the front of the vehicle. The lens 212 may be a wide-angle lens having an angle of view of 90 degrees to 180 degrees, but a lens having a specific angle of view in the range of 0 degrees to 180 degrees (not including 0 degrees) regardless of a wide-angle lens or a narrow-angle lens is provided. It is also possible to be provided.

In addition to the lens 212 , the photographing unit 210 may include all essential components well known to those skilled in the art that are necessary or advantageous for image capturing, such as an image sensor.

As shown in FIG. 4, a general camera, including the camera constituting the photographing unit 210, is composed of a lens 212 and an image sensor 214, and "L2" is assigned to the camera according to the cell size of the sensor. It is decided. Here, “L1” is the size of the object, “d” is the distance between the lens and the object, and “f” is the distance between the lens and the image sensor, which may mean the focal length.

Here, as an example, in a 1280x720 1 mega camera, if the cell size is 3um, d = 100m, L1 (person) = 170cm, and f = 5.5, L2 is calculated as 0.09405mm and 31.35pixel. For reference, the visible ratio varies depending on the cell size and the focal length "f".

Based on this, in a 1 mega sensor, a 170 cm person at a distance (d) of 100 m is measured as 10 pixels in width and 30 pixels in height, and in this case, difficulties arise in recognizing images based on the current image recognition algorithm. . Here, when using the current state of the art image recognition algorithm, it is assumed that an image recognition range capable of recognizing a person of 170 cm at a distance (d) of 100 m without error is a range of 32 pixels in width and 92 pixels in height.

And, under the same conditions, the 8-megapixel sensor comes into the image recognition range where a 170 cm tall person appears as 32 pixels in width and 92 pixels in height at a distance (d) of 100 m with a resolution of 3840x2160. 3840/1280=3, 2160/720=3 It has a difference of about 3 times, and when measuring a 170cm person from a distance of 100m, it appears in proportion to 32 pixels and 92 pixels. However, in this case, when recognizing an image having a resolution of 3840x2160 using the image recognition algorithm built into the image recognition unit 220, it is possible to recognize a person within a range of 100 m, but real-time recognition is not possible due to heavy load. A difficult problem arises.

As a result, in the case of an image captured through a camera having a 1280x720 1 mega sensor, object information that is a distant object of 100 m or more is not recognized, and in the case of an image captured through a camera having a 3840x2160 8 mega sensor, a distance of 100 m or more Recognition of object information is possible, but a problem arises in that real-time recognition is difficult due to a lot of load.

Accordingly, a high-resolution image is required for recognition of the entire range of short and long distances in front of the autonomous vehicle, but in this case, a lot of load occurs and real-time recognition is difficult. A configuration is required to reduce load generation and enable real-time recognition.

In order to overcome this problem, the image recognition unit 220 has an image recognition AI algorithm embedded therein to recognize object information, which is a target of recognition in the self-driving vehicle, from an image provided through the photographing unit 210. recognition will be performed. At this time, if the driving speed of the self-driving vehicle is less than the reference speed, the image is downscaled to a size having a reference resolution to perform image recognition, and if the driving speed is greater than or equal to the reference speed, image recognition is performed. Image recognition is performed from the cropped image by cropping to a size corresponding to the reference resolution. Here, the reference speed may be 60 to 100 km/h.

When the image recognition unit 220 performs image recognition, the reference resolution is a resolution at which image recognition errors do not occur and image recognition speed does not decrease and real-time recognition is possible with respect to object information of a short distance as well as a long distance. , it may be a preset resolution. Here, it is assumed that the reference resolution is 640X480. The reference resolution may be set to a higher resolution than 640X480 according to the development of image recognition technology.

5 shows the process of down-scaling and cropping the image recognition unit, and FIG. 6 shows examples of down-scaling (b), cropping (c), and up-scaling (d) images of an actual captured image (a). it is depicted

As shown in (a) of FIG. 5 , the image recognition unit 220 detects a high-resolution image 215a provided through the photographing unit 210 when the driving speed of the self-driving vehicle is less than the reference speed. Downscaling is performed to have the reference resolution, and object recognition is performed on the downscaled image 225a. In this case, in the case of a downscaled image (see FIG. 6(b)) from the original image (see FIG. 6(a)), a distant object may look smaller and deteriorate, resulting in a recognition error. In the case of driving at a low speed below the reference speed, recognition of object information located in a short distance is more important than a distant object of 60 to 100 m or more, so that the deteriorated part can be ignored, so that there is no problem in real-time recognition and object information recognition. don't

As shown in (b) of FIG. 5, the image recognition unit 220 is located at a distance of 60 to 100 m or more rather than the short distance information near the vehicle when the driving speed of the autonomous vehicle is running at a high speed higher than the reference speed. Recognition of object information becomes more necessary. Accordingly, during high-speed driving, in this case, cropping is performed according to the cropping size 215b on the image based on the front of the vehicle from the original image (215a or FIG. 6(a)), and the cropped image (225b or FIG. 6 (c) reference), object recognition is performed.

In this case, the cropped image (see 225b or (c) of FIG. 6) is cropped to a size having the reference resolution. For example, if the original image shown in (a) of FIG. 6 is taken by a camera having an 8-megapixel sensor and has a resolution of 3840x2160, and there is a person at a distance of 170 cm at a distance of 100 m, the image of the person It is possible to perform cropping with an image of a size having a resolution of 640x480, which is a reference resolution, including a corresponding 32-pixelx92-pixel part. Accordingly, it is possible to set the image recognition speed for distant object information to a real-time recognition level. In particular, when driving on a highway, recognition of object information located at a distance of 60 to 100 m or more rather than a short distance must be performed without error and must be recognized in real time, so it is possible to recognize object information through cropping. In the case of cropping, since a corresponding size is cut out from an original captured image, unlike downscaling, deterioration of a distant object does not occur. Here, cropping is performed based on the size of the image, not the size of the subject.

The above-described cropping is generally performed based on the central part of the original image captured by the photographing unit 210, but if necessary (eg, when turning left or right), the left or right side of the central part Cropping may be performed based on the right side.

Here, if necessary, it is possible to perform cropping after downscaling the image 215a provided through the photographing unit 210 to a certain level before performing the cropping. In this case, there is an advantage in that the recognition range can be broadened compared to the case where cropping is performed without downscaling. Further, the downscaling may be performed at a rate corresponding to the traveling speed. That is, as the driving speed increases, the downscale ratio decreases, and as the driving speed decreases, the downscale ratio increases, so that the recognition range can be widened when the driving speed is slow.

The image recognition unit 220 may be built into an electronic control unit (ECU) of an autonomous vehicle, and is equipped with an AI algorithm (deep learning algorithm, deep neural network algorithm, etc.) for image recognition to provide a deep learning learning model, etc. Recognition of object information in an image is performed through a learning process, and object information is recognized in a state in which the image provided from the photographing unit 210 is downscaled or cropped.

The image recognizing unit 220 may include a memory (not shown) and an image processor (not shown) for storing images provided from the photographing unit 210. The image processor is included in the input image. It is possible to perform image processing for improving the recognition rate of object information.

And, the object information may include information about an object to be recognized in an autonomous vehicle for autonomous driving. The object information may include at least one of target road information to be recognized, vehicle information, moving object information including people, obstacle information, traffic light information, sign information, and event information. The moving object information may include all moving objects including people and animals, and the road information may include road surface information or lane information. In addition, the vehicle information may include information on various vehicles of two or more wheels, including two-wheeled vehicles such as electric quick boards, bicycles, and motorcycles. In addition, the object information may include target road information, vehicle information, moving object information including people, obstacle information, traffic light information, sign information, and location information or color information of event information. In addition, information known to require recognition from a photographed image for autonomous driving may be included. For example, abnormal object information may be additionally included. Abnormal object information may include, for example, objects such as people or animals performing abnormal behaviors other than generally known normal behaviors, such as a person sitting, a person lying down, a carcass of an animal, and a person using an umbrella. .

As described above, according to the second embodiment of the present invention, while capturing an image using a high-resolution camera, downscaling or cropping the image is performed in response to the driving speed of the autonomous vehicle so that the front of the vehicle It has the advantage of being able to recognize the entire range of near and far distances, as well as real-time recognition by reducing the load in recognizing high-resolution images and distant object information.

7 is a schematic diagram of an image recognition device for an autonomous vehicle according to a third embodiment of the present invention, and FIG. 8 is a schematic diagram illustrating a cropping process of the image recognition unit of FIG. 7 .

As shown in FIG. 7 , an image recognizing device 300 for an autonomous vehicle according to a third embodiment of the present invention includes a photographing unit 310 and an image recognizing unit 320 .

The photographing unit 310 includes at least one lens 312 each having a predetermined angle of view and generating images by photographing the front of the vehicle. The lens 312 may be a wide-angle lens having an angle of view of 90 degrees to 180 degrees, but a lens having a specific angle of view in the range of 0 degrees to 180 degrees (not including 0 degrees) regardless of a wide-angle lens or a narrow-angle lens is provided. It is also possible to be provided.

In addition to the lens 312, the photographing unit 310 may include all essential elements well known to those skilled in the art that are necessary or advantageous for image photographing, such as an image sensor.

The image recognition unit 320 embeds an image recognition AI algorithm to perform image recognition for recognizing object information to be recognized in the self-driving vehicle from the image provided through the photographing unit. Corresponding to the driving speed, image recognition is performed by performing at least one of downscaling ratio adjustment and cropping image size adjustment with respect to the image provided through the photographing unit.

In general, when the driving speed is low, it is important to recognize object information in a short distance in front of the vehicle and in a wide range, and when the driving speed is high, it is important to recognize object information in a long distance in front of the vehicle.

Accordingly, as an example, as shown in FIG. 8, the image recognition unit 320 performs cropping by making the size of the image on which the cropping is performed gradually decrease as the driving speed of the autonomous vehicle increases can do. Specifically, as shown in FIG. 8 , with respect to the original image 315a photographed and provided by the photographing unit 310, as the traveling speed of the vehicle increases, the size of the image to which cropping is performed increases with the first cropping. Image 315b, second cropped image 315c, and third cropped image 315d are cropped in order to reduce the size of the cropped image, and image recognition is performed on the cropped image. can This is because the recognition of distant object information becomes more important than a wide range of short distances as the driving speed of the vehicle increases, so it is possible to increase the efficiency of image recognition by adjusting the size of the cropped image according to the driving speed. .

The above-described cropping is generally performed based on the central part of the original image captured by the photographing unit 310, but if necessary (eg, when turning left or right), the left side of the central part or Cropping may be performed based on the right side.

And, as another example, before performing the cropping, the image recognition unit 320 downscales the original image captured by the photographing unit 310, and performs the cropping on the downscaled image. Image recognition can be performed from the cropped image by performing (croping).

In this case, before performing the cropping, downscaling is performed at a rate corresponding to the traveling speed with respect to the original image captured by the photographing unit 310, and the cropping is performed on the downscaled image. It is possible to perform image recognition from the cropped image by performing In this case, the downscaling is performed such that the downscaling ratio decreases as the driving speed increases and the downscaling ratio increases as the driving speed decreases, so that the recognition range can be further widened when the driving speed is slow. In this case, the size of the image to be cropped can be gradually reduced according to the traveling speed, and it is also possible to have a fixed size regardless of the traveling speed. In this case, compared to the case of adjusting the downscaling ratio and cropping size in response to the driving speed, the downscaling ratio can be adjusted to a greater extent when the downscaling ratio is adjusted and the cropping size is fixed in response to the traveling speed.

In another case, the downscaling performed before performing the cropping on the original image captured by the photographing unit 310 is performed at a fixed ratio regardless of the driving speed, and is performed on the downscaled image The cropping may be performed to a size corresponding to the traveling speed. In this case, the size of the image to be cropped can be gradually reduced according to the traveling speed, and it is also possible to have a fixed size regardless of the traveling speed.

As another example, the image recognition unit 320 performs image recognition using the entire image provided through the photographing unit 310 when the driving speed is less than the reference speed, and when the driving speed is greater than or equal to the reference speed It is also possible to perform cropping by making the size of the cropped image gradually smaller as the traveling speed increases, and to perform image recognition from the cropped image. Here, the reference speed may be 60 to 100 km/h.

The image recognition unit 320 may be built into an electronic control unit (ECU) of an autonomous vehicle, and is provided with an AI algorithm (deep learning algorithm, deep neural network algorithm, etc.) for image recognition to provide a deep learning learning model, etc. Recognition of object information in an image is performed through a learning process, and object information is recognized in a state in which the image provided from the photographing unit 310 is three-cropped.

The image recognizing unit 320 may include a memory (not shown) and an image processor (not shown) for storing images provided from the photographing unit 310. The image processor is included in the input image. It is possible to perform image processing for improving the recognition rate of object information.

And, the object information may include information about an object to be recognized in an autonomous vehicle for autonomous driving. The object information may include at least one of target road information to be recognized, vehicle information, moving object information including people, obstacle information, traffic light information, sign information, and event information. The moving object information may include all moving objects including people and animals, and the road information may include road surface information or lane information. In addition, the vehicle information may include information on various vehicles of two or more wheels, including two-wheeled vehicles such as electric quick boards, bicycles, and motorcycles. In addition, the object information may include target road information, vehicle information, moving object information including people, obstacle information, traffic light information, sign information, and location information or color information of event information. In addition, information known to require recognition from a photographed image for autonomous driving may be included. For example, abnormal object information may be additionally included. Abnormal object information may include, for example, objects such as people or animals performing abnormal behaviors other than generally known normal behaviors, such as a person sitting, a person lying down, a carcass of an animal, and a person using an umbrella. .

As described above, according to the third embodiment of the present invention, it is possible to adjust the recognition range in response to the traveling speed of the front of the vehicle by cropping the image in response to the traveling speed of the autonomous vehicle. There are advantages.

The description of the above embodiment is merely an example with reference to the drawings for a more thorough understanding of the present invention, and should not be construed as limiting the present invention. In addition, it will be apparent to those skilled in the art that various changes and modifications are possible within a range that does not deviate from the basic principles of the present invention.

110,210,310: photographing unit 120,220,320: image recognition unit

Claims (17)

delete delete delete delete delete In the image recognition device of an autonomous vehicle:
a photographing unit having a lens having a certain angle of view and generating an image by photographing the front of the vehicle;
An image recognition AI algorithm is embedded to perform image recognition for recognizing object information to be recognized in the self-driving vehicle from the image provided through the photographing unit,
When the driving speed of the self-driving vehicle is less than the reference speed, image recognition is performed by downscaling the image so that the image has a reference resolution, and when the driving speed is greater than or equal to the reference speed, the reference for the image is performed. An image recognition device for an autonomous vehicle comprising an image recognition unit that performs image recognition from a cropped image by cropping to a size corresponding to a resolution.
The method of claim 6,
The lens is an image recognition device for an autonomous vehicle, characterized in that a wide-angle lens having a view angle of 90 degrees to 180 degrees.
The method of claim 6,
The reference resolution is a resolution capable of real-time recognition without an image recognition error when the image recognition unit performs image recognition, characterized in that it is a preset resolution.
The method of claim 8,
The image recognition device for an autonomous vehicle, characterized in that the reference resolution is 640X480.
The method of claim 6,
The image recognition device for an autonomous vehicle, characterized in that the reference speed is 60 to 100 km / h.
The method of claim 6,
The image recognizing unit, when the traveling speed is equal to or greater than the reference speed, performs downscaling at a rate corresponding to the traveling speed, crops the downscaled image to a size corresponding to the reference resolution, and crops the image. An image recognition device for an autonomous vehicle, characterized in that it performs image recognition from a pinged image.
delete delete In the image recognition device of an autonomous vehicle:
a photographing unit having a wide-angle lens and generating an image by photographing the front of the vehicle;
An image recognition AI algorithm is embedded to perform image recognition for recognizing object information to be recognized in the self-driving vehicle from the image provided through the photographing unit,
An image recognition unit configured to perform image recognition by performing at least one of a downscaling rate adjustment and a size adjustment of a cropped image of the image in response to the driving speed of the autonomous vehicle,
Characterized in that the image recognition unit performs downscaling on the image before performing the cropping, and performs image recognition on the cropped image by performing the cropping on the downscaled image. Image recognition device for autonomous vehicles.
The method of claim 14,
The downscaling performed before performing the cropping is performed at a fixed ratio regardless of the driving speed, and the cropping performed on the downscaled image is adjusted to a size corresponding to the driving speed. Image recognition device of driving vehicle.
The method of claim 14,
The downscaling performed before performing the cropping is performed at a rate corresponding to the driving speed, and the cropping performed on the downscaled image is performed at a fixed size regardless of the driving speed. image recognition device.
delete

Images