KR20190134303A - Apparatus and method for image recognition - Google Patents

Abstract

Disclosed is an image recognition method. The image recognition method comprises the steps of: generating a virtual 3D road image corresponding to a situation around a vehicle in real time by using vehicle location information, map information, and distance information from the vehicle to an object located around a road, which are collected by an information collection module; acquiring an actual road image by an image acquisition module in real time; and matching the actual road image with the 3D road image on a frame-by-frame basis to recognize an object which appears on the real road image using information included in the 3D road image.

Description

Image recognition device and method thereof {APPARATUS AND METHOD FOR IMAGE RECOGNITION}

The present invention relates to an image recognition apparatus and method, and more particularly, to an image recognition apparatus and method for recognizing an object in an image obtained from a driving vehicle.

Recently, with the rapid development of IT technology, there is a growing interest in intelligent vehicles integrated with vision systems.

Vision-based systems by image sensors have been widely used to this day because of the strong advantage of being able to extract a large amount of information at a low cost and utilizing various conventional image processing algorithms.

In particular, Advanced Safety Vehicle (ASV) technologies, such as lane departure information, lane keeping, and crash warning systems, which reduce the risk of traffic accidents, are the foundation technology for intelligent vehicle technology. It is being committed.

However, as is well known, the execution of an image processing algorithm acts as a heavy load of hardware, and there is a problem in that expensive hardware that provides high performance is required for smooth image processing.

Accordingly, an object of the present invention is to provide an image recognition apparatus and method for greatly reducing the computational load caused by image processing in a process of recognizing an object in an image acquired by a vehicle.

In accordance with an aspect of the present invention, an image recognition method for achieving the above object includes a vehicle surrounding situation using vehicle location information, map information, and distance information from an object located near a road to the vehicle. Generating a virtual 3D road image corresponding to the in real time; Acquiring a real road image in real time by an image acquisition module; And matching the real road image with the 3D road image in units of frames in order to recognize an object appearing in the real road image using information included in the 3D road image.

According to another aspect of the present invention, an image recognition device includes: an information collection module configured to collect vehicle location information, map information, and distance information from the vehicle to an object located around a road; A 3D road image generation module that generates a virtual 3D road image corresponding to a situation around the vehicle in real time using the collected vehicle location information, map information, and distance information; An image acquisition module for obtaining an actual road image in real time; And a matching module for matching the real road image with the 3D road image in units of frames in order to recognize an object appearing in the real road image using information included in the 3D road image.

According to the present invention, in order to recognize the objects appearing on the actual road image, 3D road image including information about the objects around the road without the computational load processing process, such as image-based object recognition algorithm to the actual road image Simply matching makes it easy to recognize the objects that appear in the actual road image.

1 is a block diagram of an image recognition device according to an exemplary embodiment.
2 to 5 are views for explaining a method of generating a 3D road image according to an embodiment of the present invention.
FIG. 6 is a block diagram illustrating the matching module shown in FIG. 1.
7 is a diagram conceptually illustrating a matching process according to an exemplary embodiment of the present invention.
8 to 11 are flowcharts illustrating an image recognition method according to an exemplary embodiment.

Embodiments of the present invention may be variously modified and have various embodiments, and specific embodiments are illustrated in the drawings and related detailed descriptions are described. However, this is not intended to limit the various embodiments of the present invention to specific embodiments, it should be understood to include all modifications and / or equivalents and substitutes included in the spirit and scope of the various embodiments of the present invention. In the description of the drawings, similar reference numerals are used for similar elements.

Expressions such as "including" or "can include" as used in embodiments of the present invention indicate the existence of the corresponding function, operation or component disclosed, and additional one or more functions, operations or configurations. Do not restrict elements. In addition, in various embodiments of the invention, the terms "comprise" or "have" are intended to indicate that there is a feature, number, step, action, component, part, or combination thereof described in the specification, one Or other features or numbers, steps, operations, components, parts or combinations thereof in any way should not be excluded in advance.

In an embodiment of the invention, the expression "or" includes any and all combinations of words listed together. For example, "A or B" may include A, may include B, or may include both A and B.

The terms used in the embodiments of the present invention are merely used to describe specific embodiments, and are not intended to limit the embodiments of the present invention. Singular expressions include plural expressions unless the context clearly indicates otherwise.

Hereinafter, an image recognition apparatus according to an embodiment will be described with reference to the accompanying drawings.

1 is a block diagram of an image recognition device according to an embodiment of the present invention.

Referring to FIG. 1, a target of an image recognition apparatus 100 according to an exemplary embodiment may be a vehicle. However, the present invention is not limited thereto, and the image recognition apparatus according to an exemplary embodiment of the present invention may be applied to all objects requiring object recognition in an image, for example, an airplane, a train, a mobile robot, a ship, and the like.

Image recognition apparatus 100 according to an embodiment of the present invention may be referred to as a computing device having a communication function. The computing device may comprise a processor (e.g., a CPU, a graphics processor, etc.) mounted on a main board, memory (e.g., volatile and nonvolatile memory, etc.), storage (e.g., a hard disk, etc.), a network interface, Chipset and system bus connecting them.

Although a plurality of blocks are shown in FIG. 1, this is merely a functional division of the computing device and is not intended to be limiting. However, among the blocks shown in FIG. 1, modules indicated by reference numerals 110, 120, 130, 140, and 150 may be mounted in a device called a “processor” in the computing device.

In detail, the image recognition apparatus 100 may include the communication interface 110, the information collection module 120, the 3D road image generation module 130, the image acquisition module 140, the matching module 150, and the display module 160. And a storage module 170.

The communication interface 110 communicates with the electronic device 10 in a vehicle through vehicle network communication. Here, the vehicle network communication may include, for example, controller area network (CAN) communication or local interconnect network (LIN) communication.

The electronic device 10 in a vehicle may include a GPS receiver 10 providing real-time vehicle position information of a driving vehicle, a distance sensor 13 providing distance information to objects around a vehicle, and a navigation providing map information. Unit 15, an engine control unit that provides vehicle speed information 17, and the like.

The GPS receiver 10 may provide vehicle location information in the form of a coordinate value consisting of latitude and longitude.

The distance sensor 13 may be formed of an ultrasonic sensor, a radar, a lidar, or a combination thereof to provide distance information to an object around the vehicle. Here, the object around the vehicle includes a static object and a dynamic object, and the static object may include, for example, a traffic structure such as a building, a roadside tree, a traffic light, and the like. For example.

The map information may include information (hereinafter, road section information) on a road section of a driving vehicle, information on surrounding buildings located in the road section, information on a traffic signal, and the like.

The road section information may include information indicating whether the road section in which the vehicle is currently driving is a general road or a highway (road section information), and information on the number of lanes included in the road section.

The information collection module 120 collects information received through the communication interface 110, for example, vehicle location information, distance information, map information, vehicle speed information, and the like, and may be implemented as a kind of buffer memory. have.

The 3D image generation module 130 generates a virtual 3D (3D) road image corresponding to a situation around the vehicle in real time using the vehicle location information, the distance information, the map information, the vehicle speed information, and the like.

The image acquisition module 140 generates a real road image corresponding to the situation around the vehicle of the currently driving vehicle in real time. To this end, the image acquisition module 140 may be configured to include any one of a color camera, an infrared camera, a stereo camera, or a combination thereof, for example.

The matching module 150 matches the virtual 3D road image generated by the 3D image generation module 130 with the actual road image generated by the image acquisition module 140.

By the matching of the matching module 150, it is possible to recognize the objects appearing in the actual road image based on the information included in the 3D road image. That is, as in the related art, in order to recognize objects appearing in a real road image, a 3D road image including information about objects around a road without a high computational load processing process, such as an image-based object recognition algorithm, is a real road image. Just by matching to, you can easily recognize the objects appearing on the actual road image.

Detailed description of the matching module 150 will be described in detail below with reference to FIG.

Hereinafter, a method of generating a 3D road image in the 3D image generating module 130 will be described in detail with reference to FIGS. 2 to 5.

2 to 5, the 3D image generating module 130 uses the vehicle location information and the map information collected by the information collecting module to determine how many lanes the road section on which the vehicle is currently driving includes. .

For example, the 3D image generation module 130 matches the vehicle location information to a map included in the map information to identify a road section in which the vehicle is currently driving, and identifies lane information of the identified road section, that is, information on the number of lanes. Acquire.

When the number information of the lanes is obtained, the reference 3D road image corresponding to the obtained number of lane information is read from the storage module 170. An example of a reference 3D road image is shown in FIG. 2.

The reference 3D road image may be a road image of a 3D type previously modeled and may include a plurality of road images classified according to the number of lanes.

For example, when the road section currently being driven is identified as one lane according to the obtained number of lane information, the 3D image generating module 130 reads the reference 3D road image composed of one lane from the storage module 170. .

As another example, when the road section currently being driven is identified as three lanes according to the obtained number of lane information, the 3D image generating module 130 calls the reference 3D road image composed of three lanes from the storage module 170. come.

When the reference 3D road image is prepared according to the number of lane information, surroundings of the road in the prepared reference 3D road image by using structure information such as traffic information and information about nearby buildings located in the road section included in the map information. Place the structures 42, 44 in. An example in which a structure is placed in a reference 3D road image is shown in FIG. 3.

When the generation of the reference 3D road image on which the structure is disposed is completed, the driving lane of the current vehicle is determined, and the viewpoint of the reference 3D road image on which the structure is disposed is determined based on the determined driving lane. Convert to a viewpoint.

First, a method of determining how many lanes a vehicle is currently driving may be determined based on, for example, the number of lanes of a currently driving road section and distance information to objects located near a road.

Referring to FIG. 5, a method of determining how many lanes a vehicle is currently driving can be sufficiently predicted by a statistical method from the number of lanes.

Lane width is a value that is designed according to the size of the vehicle and can be known in advance. Therefore, the total road width W1 can be predicted relatively accurately as the sum of the road widths of each lane. In FIG. 5, three lanes are illustrated, and a vehicle driving in two lanes of the three lanes is illustrated.

The distance d1 from the current position of the vehicle to the object 62 can be known from the distance information collected from the distance sensor 13.

In addition, when the distance sensor 13 is capable of forward scanning from left to right or right to left and uses a sensor designed to know the scan angle, the emission light and the object emitted from the vehicle in the traveling direction of the vehicle (the vehicle front) ( The angle θ between the exiting lights emitted in the direction 62) can be known.

When the distance d1 and the angle θ are applied to the trigonometric formula, the distance d2 of the vehicle away from the object 62 in the road width direction D2 may be calculated.

By analyzing the total road width W1 and the calculated distance d2 , it is possible to predict that the vehicle is currently driving two lanes.

As such, when the prediction of the driving lane of the current vehicle is completed, the image acquisition module 140 may describe the viewpoint of the reference 3D road image including the structures 42 and 44 of FIG. 3 based on the predicted driving lane. To camera's viewpoint.

This viewpoint conversion is necessary to match the actual road image acquired by the image acquisition module 140 with the reference 3D road image including the structure.

That is, the image acquisition module 140 provides the actual road image of the point in time at which the current driving lane is disposed at the center of the screen, whereas the 3D road image generation module 130 displays the virtual reference 3D road image, that is, the current driving lane. Provide an image of a time point not considered. Therefore, it is difficult to smoothly match between the actual road image at the time when the current driving lane is disposed at the center of the screen and the reference 3D road image at the time when the current driving lane is not considered.

Therefore, in order to match the actual road image and the reference 3D road image, it is necessary to convert the viewpoint of the reference 3D road image to a viewpoint in consideration of the previously predicted current driving lane. That is, as shown in FIG. 4, the viewpoint transformation of the reference 3D road image is required so that the current driving lane is disposed in the center of the screen in the screen showing the reference 3D road image.

For this viewpoint conversion, the 3D road image generation module 130 may convert the viewpoint of the reference 3D road image based on the viewpoint information of the actual image acquired by the image acquisition module 140 when the vehicle is driving a specific lane. Can be.

The viewpoint information of the actual image related to the driving lane may be information stored in advance in the storage module 170 and may be information related to an angle representing the pose of the camera as a roll, a pitch, and a yaw.

The angles represented by Roll, Pitch, and Yaw can be calculated from the correlation between the lines separating one lane and the lines separating the peripheral lanes of the first lane when the vehicle is driving one lane. have.

As shown in FIG. 4, when the viewpoint transformation of the reference 3D road image is completed such that the current driving lane is disposed at the center of the screen in the screen showing the reference 3D road image, generation of the final 3D road image is completed.

Generation of the final 3D road image is performed in real time whenever the information collection module 120 collects information.

FIG. 6 is a block diagram illustrating the matching module shown in FIG. 1.

Referring to FIG. 6, the matching module 150 matches the 3D road image generated by the 3D road image generation module 130 with the actual road image acquired by the image acquisition module 140 in units of frames.

Although the 3D road image generation module 130 performs the viewpoint conversion of the 3D road image according to the current driving lane for the matching success rate of the 3D road image and the actual road image, the matching module 150 ) Performs the objects (lanes, traffic structures, buildings around roads) included in the actual road image acquired by the image acquisition module, and compensates for errors between the extracted objects and the objects (virtual objects) included in the 3D road image. .

The important point is that when the object is extracted every frame unit from the actual road image as in the related art, the hardware weighted load increases, so to reduce this, only the actual road image of the first frame among the actual road images configured in a certain frame unit is used. , Calculate the error between the object extracted from the real road image of the first frame and the object of the 3D road image corresponding to the real road image of the first frame, and use the calculated error The matching process is performed in a manner that compensates for errors between 3D road images.

As such, the matching is performed by using only the error calculated in the first frame in a manner that compensates for the error between the actual road image and the 3D road image of the next frame, so that there is no need to extract an object from the actual road image every frame. Therefore, it is possible to significantly reduce the computational load due to object extraction and at the same time achieve precise matching between the actual road image and the 3D road image.

In detail, the matching module 150 may include the frame extractor 150-1, the object extractor 150-3, the 3D road image extractor 150-5, the comparator 150-7, and the 3D road image. The correction unit 150-9 and the matching unit 150-11 may be included.

The frame extractor 150-1 configures the actual road images acquired by the image acquisition module 140 in a predetermined frame unit (F _n to F _{n + k} ), and in a predetermined frame unit (F _n to F _{n + k} ). Extract the actual road image of the first frame (F _n ) from the actual road images consisting of.

The object extractor 150-3 extracts an object from an actual road image of the first frame F _n extracted by the frame extractor 150-1. Here, the object may include a lane, a traffic light, a building, a structure around a road, and the like.

As a method of extracting an object, various image processing algorithms may be used, and since the present invention is not characterized in defining the image processing algorithm, the description thereof is replaced by a known technique. However, as an example of an image processing algorithm, an image-based lane detection algorithm, an object extraction algorithm, and the like may be exemplified.

The 3D road image extracting unit 150-5 is 3D in the same frame unit (M _n to M _{n + k} ) as a predetermined frame unit (F _n to F _{n + k} ) configured in the above-described frame extracting unit 150-1. The road images are constructed, and the 3D road image of the first frame M _n is extracted from the 3D road images in the same frame unit (M _n to M _{n + k} ).

Meanwhile, prior to the process of extracting the 3D road image, the synchronization process between the actual road image and the 3D road image may be preceded. The synchronization method may be performed by synchronizing the operation time of the image acquisition module 140 and the information collection time of the information acquisition module 120. Although not illustrated in FIGS. 1 and 2, the operation of the image acquisition module 140 is performed. A synchronization module for synchronizing the time and information collection time of the information collection module 120 may be shown at a suitable location in FIG. 1 or 2.

The comparator 150-7 is an object extracted from the actual road image of the first frame F _n by the object extractor 150-3 and the first frame by the 3D road image extractor 150-5. M _n ) compares the objects included in the 3D road image to calculate the error corresponding to the difference. Here, the objects corresponding to the comparison object may be represented by feature vectors, and the error may be a difference value between feature vectors.

Meanwhile, since the 3D road image extracted by the 3D road image extractor 150-5 includes information about a virtual object, the object is extracted from the actual road image, as in the object extractor 150-3. The process does not have to be performed.

The 3D road image correction unit 150-9 corrects 3D road images configured in frame units M _n to M _{n + k} using the error calculated by the comparison unit 150-7.

The matching unit 150-11 is corrected by the 3D road image correcting unit 150-9 and the actual road images in a predetermined frame unit F _n to F _{n + k} input to the frame extracting unit 150-1. Matched 3D road images in a predetermined frame unit (M _n to M _{n + k} ) every frame.

7 conceptually illustrates a matching process performed by the matching unit 150-11.

According to the matching process of the matching unit 150-11, image processing for object recognition is performed only on the actual road image of the first frame among the actual road images in a predetermined frame unit, and thus, the matching unit 150-11 should be performed every frame as in the prior art. This greatly reduces the computational load on the object recognition process, and enables accurate matching between actual road images and 3D road images.

By such a precise matching process, it is possible to easily recognize the objects appearing on the actual road image based on the information included in the 3D road image without complicated image processing.

8 to 11 are flowcharts illustrating an image recognition method according to an exemplary embodiment.

First, referring to FIG. 8, in step S811, a process of collecting vehicle location information, map information, and distance information from a vehicle to an object located around a road is performed.

Subsequently, in step S813, a process of generating a virtual 3D road image corresponding to a situation around a vehicle in real time using the collected vehicle location information, map information, and distance information is performed.

In an embodiment, as shown in FIG. 9, in step S813, a process of determining a driving lane in which the vehicle is driven using the vehicle location information, the map information, and the distance information (S813-1) and the The step S813-3 may generate the 3D road image based on the determined driving lane.

Here, the step (S813-3) of generating the 3D road image based on the determined driving lane is the same as the point in time at which the driving lane of the vehicle is displayed on the display screen in the actual road image acquired by the image acquisition module. The determined driving lane may have the 3D road image arranged at the center of the display screen to have a viewpoint.

In another embodiment, as illustrated in FIG. 10, in operation S813-5, acquiring the number of lanes included in the road section in which the vehicle is currently driving by matching the vehicle location information with map information (S813-5). (S813-7) reading a reference 3D road image corresponding to the obtained lane number information from a storage module (step S813-7), and using the structure information included in the map information, to the reference 3D road image. Disposing a structure (S813-9) and converting the viewpoint of the reference 3D road image on which the structure is disposed to a camera viewpoint of the image acquisition module to generate the 3D road image (S813-11). .

The process (S813-11) may include determining the driving lane of the vehicle and acquiring the image of the viewpoint of the reference 3D road image so that the determined driving lane is disposed at the center of a display screen displaying the reference 3D road image. The method may include converting to a camera viewpoint of the module.

The determining of the driving lane of the vehicle may include calculating the distance that the vehicle is away from an object in a road width direction by using the distance information collected from a distance sensor capable of knowing a scan angle, and driving the vehicle. The driving lane of the vehicle may be determined by analyzing the total road width of the road section and the calculated distance.

Subsequently, in step S815, a process of obtaining an actual road image from the image acquisition module in real time is performed.

Subsequently, in step S817, a process of matching the real road image with the 3D road image in units of frames is performed to recognize an object appearing in the real road image using the information included in the 3D road image.

In detail, in operation S817, an object is extracted from the actual road image of the first frame among the actual road images including N frames (S817-1), and the N frames. Extracting an object from the 3D road image of the first frame among the 3D road images consisting of (S817-3), the feature vector value of the object extracted from the actual road image of the first frame and the 3D of the first frame A process of calculating an error between feature vector values of an object extracted from a road image (S817-5), and a process of correcting 3D road images from the first frame to the Nth frame using the calculated error (S817-7) And matching the corrected 3D road images with the real road images every frame (S817-9).

Although the present invention has been described above with reference to the embodiments, these are only examples and are not intended to limit the present invention, and those skilled in the art to which the present invention pertains may have an abnormality within the scope not departing from the essential characteristics of the present invention. It will be appreciated that various modifications and applications are not illustrated. For example, each component specifically shown in the embodiment of the present invention can be modified. And differences relating to such modifications and applications will have to be construed as being included in the scope of the invention defined in the appended claims.

Claims (14)

In the image recognition method for recognizing the object in the image acquired by the driving vehicle,
Generating, in real time, a virtual 3D road image corresponding to a situation around the vehicle by using the vehicle location information, the map information collected by the information collection module, and the distance information from the vehicle to an object located around the road;
Acquiring a real road image in real time by an image acquisition module; And
Matching the real road image with the 3D road image in units of frames in order to recognize an object appearing in the real road image using information included in the 3D road image.
Image recognition method comprising a.
The method of claim 1, wherein the generating of
Determining a driving lane on which the vehicle is driven by using the vehicle location information, the map information, and the distance information; And
Generating the 3D road image based on the determined driving lane
Image recognition method comprising a.
The method of claim 2, wherein the generating of the
Generating the 3D road image in which the determined driving lane is disposed at the center of the display screen such that the determined driving lane has the same view point as the time at which the driving lane of the vehicle is displayed on the display screen in the actual road image acquired by the image acquisition module; Image recognition method.
The method of claim 1, wherein the generating of
Acquiring the number of lanes included in a road section in which the vehicle is currently driving by matching the vehicle location information with map information;
Reading a reference 3D road image corresponding to the obtained number of lane information from a storage module;
Arranging a structure in the reference 3D road image by using the structure information included in the map information; And
Generating the 3D road image by converting the viewpoint of the reference 3D road image on which the structure is disposed to the camera view of the image acquisition module
Image recognition method comprising a.
The method of claim 4, wherein the converting step,
Determining a driving lane of the vehicle; And
Converting the viewpoint of the reference 3D road image to the camera viewpoint of the image acquisition module so that the determined driving lane is disposed at the center of a display screen displaying the reference 3D road image.
Image recognition method comprising a.
The method of claim 5, wherein the determining comprises:
Calculating a distance that the vehicle is away from an object in a road width direction using the distance information collected from a distance sensor capable of knowing a scan angle; And
Determining a driving lane of the vehicle by analyzing a total road width and the calculated distance of a road section in which the vehicle is driving;
Image recognition method comprising a.
The method of claim 1, wherein the matching comprises:
Constructing the actual road image and the 3D road image into N frames consecutive over time;
Extracting an object from the actual road image of the first frame among the actual road images composed of the N frames;
Extracting an object from the 3D road image of the first frame among the 3D road images composed of the N frames;
Calculating an error between the feature vector value of the object extracted from the actual road image of the first frame and the feature vector value of the object extracted from the 3D road image of the first frame;
Correcting 3D road images from the first frame to the Nth frame using the calculated error; And
Matching the corrected 3D road images and the actual road images every frame
Image recognition method comprising a.
In the image recognition device for recognizing the object in the image acquired by the driving vehicle,
An information collection module for collecting vehicle location information, map information, and distance information from the vehicle to an object located around a road;
A 3D road image generation module that generates a virtual 3D road image corresponding to a situation around the vehicle in real time using the collected vehicle location information, map information, and distance information;
An image acquisition module for obtaining an actual road image in real time; And
Matching module for matching the real road image and the 3D road image in units of frames in order to recognize an object appearing in the real road image using information included in the 3D road image.
Image recognition device comprising a.
The method of claim 8, wherein the 3D road image generation module,
And determining the driving lane in which the vehicle is driven by using the vehicle location information, the map information, and the distance information, and generating the 3D road image based on the determined driving lane.
The method of claim 8, wherein the 3D road image generation module,
Generating the 3D road image in which the determined driving lane is disposed at the center of the display screen such that the determined driving lane has the same view point as the time when the driving lane of the vehicle is displayed on the display screen in the actual road image acquired by the image acquisition module; Image recognition device.
The method of claim 8, wherein the 3D road image generation module,
Obtaining information on the number of lanes included in a road section in which the vehicle is currently driving by matching the vehicle location information with map information;
Reading a reference 3D road image corresponding to the obtained lane number information from a storage module;
A process of arranging the structure in the reference 3D road image by using the structure information included in the map information;
The process of converting the viewpoint of the reference 3D road image in which the structure is disposed into the camera viewpoint of the image acquisition module.
The image recognition device to process.
The converting process of claim 11, wherein the converting process is performed by the 3D road image generating module.
A process of determining a driving lane of the vehicle, and
A process of converting the viewpoint of the reference 3D road image into the camera viewpoint of the image acquisition module such that the determined driving lane is disposed at the center of a display screen displaying the reference 3D road image
Image recognition apparatus comprising a.
The determining process of claim 12, wherein the determining process performed in the 3D road image generation module comprises:
A process of calculating the distance the vehicle is away from an object in the road width direction using the distance information collected from a distance sensor capable of knowing a scan angle;
A process of determining a driving lane of the vehicle by analyzing a total road width and the calculated distance of a road section in which the vehicle is driving;
Image recognition apparatus comprising a.
The method of claim 8, wherein the matching module,
A process of constructing the actual road image and the 3D road image with N frames contiguous with time;
A process of extracting an object from the actual road image of the first frame among the actual road images composed of the N frames,
A process of extracting an object from the 3D road image of the first frame among the 3D road images composed of the N frames,
A process of calculating an error between the feature vector value of the object extracted from the actual road image of the first frame and the feature vector value of the object extracted from the 3D road image of the first frame;
Correcting 3D road images from the first frame to the Nth frame using the calculated error, and
A process of matching the corrected 3D road images and the actual road images every frame
The image recognition device to perform.

Images