KR20210150805A - Apparatus and method for generating High Definition Map - Google Patents

Abstract

The present application relates to a high precision map generation device and a high precision map generation method. A high precision map generation method according to one embodiment of the present invention may comprise: a step of recognizing road surface signs of a road included in an aerial image by using an area box detection neural network, and generating an area box which indicates location information and direction information of the road signs; a step of recognizing road objects included in the aerial image using a semantic extraction neural network, specifying lanes in the road according to input lane location information, and classifying the type of each lane; and a step of generating a high precision map corresponding to the aerial image by using the road signs and the lanes. According to the present invention, various road signs and lanes can be accurately recognized and classified.

Description

{Apparatus and method for generating High Definition Map}

The present application relates to a high-precision map generating apparatus and a high-precision map generating method, and more particularly, to a high-precision map generating apparatus and a high-precision map generating method, which can accurately recognize and classify various road signs or lanes from aerial images.

Unmanned autonomy of a vehicle (autonomous driving vehicle) can be largely composed of a stage of recognizing the surrounding environment, a stage of planning a driving route from the recognized environment, and a stage of driving along the planned path. What is required is a High Definition Map (HD-Map).

A precision road map (HD-Map) is a map in which information about the surrounding environment, such as roads, terrain, and curvature, is implemented in 3D. In particular, it includes various information necessary for driving on roads. For example, a precision road map (HD-Map) includes various types of information necessary for driving on the road, such as lanes, driving directions, intersections, signs, traffic lights, and speed limits.

Since autonomous vehicles drive on the road while recognizing the surrounding environment based on the HD-Map, the HD-Map can be the most basic in the field of autonomous driving.

An object of the present application is to provide a high-precision map generating device and a high-precision map generating method that can accurately recognize and classify various road signs or lanes from aerial images.

An object of the present application is to provide a high-precision map generating apparatus and a high-precision map generating method that can more accurately determine the direction of a road sign recognized from an aerial image.

An object of the present application is to provide a high-precision map generating device and a high-precision map generating method that can quickly classify various types of lanes included in an aerial image and provide precise location information of the lanes.

A high-precision map generation method according to an embodiment of the present invention uses an area box detection neural network to recognize a road sign of a road included in an aerial image, and generates an area box indicating location information and direction information of the road sign. to do; recognizing road objects included in the aerial image using a semantic extraction neural network, specifying a lane in the road according to the input lane location information, and classifying the type of each lane; and generating a high-precision map corresponding to the aerial image by using the road sign and lanes.

A high-precision map generating apparatus according to an embodiment of the present invention recognizes a road sign of a road included in an aerial image by using an area box detection neural network, and generates an area box indicating location information and direction information of the road sign. a road sign recognition unit; a lane recognition unit for recognizing road objects included in the aerial image by using a semantic extraction neural network, specifying a lane on the road according to the input lane location information, and classifying the type of each lane; and a map generator that generates a high-precision map corresponding to the aerial image by using the road sign and lanes.

Incidentally, the means for solving the above problems do not enumerate all the features of the present invention. Various features of the present invention and its advantages and effects may be understood in more detail with reference to the following specific embodiments.

According to an apparatus for generating a high-precision map and a method for generating a high-precision map according to an embodiment of the present invention, it is possible to accurately recognize and classify various road signs or lanes from an aerial image.

According to the high-precision map generating apparatus and high-precision map generating method according to an embodiment of the present invention, the direction of the old sign recognized from the aerial image is the direction of the road where the road sign is located or the direction of the road signs included in the same group. Since it can be corrected taking into account

According to the high-precision map generating apparatus and high-precision map generating method according to an embodiment of the present invention, it is possible to quickly classify various types of lanes included in an aerial image, and at the same time precisely provide location information for the lanes. It is possible.

However, the effects achievable by the high-precision map generating apparatus and the high-precision map generating method according to the embodiments of the present invention are not limited to the above-mentioned effects, and other effects not mentioned above can be seen from the description below. It will be clearly understood by those of ordinary skill in the art.

1 is a block diagram showing an apparatus for generating a high-precision map according to an embodiment of the present invention.
2 is a schematic diagram illustrating road signs and lanes extracted by the high-precision map generating apparatus according to an embodiment of the present invention.
3 is a schematic diagram showing direction correction for road signs according to an embodiment of the present invention.
4 is a schematic diagram illustrating lane recognition according to an embodiment of the present invention.
5 is a flowchart illustrating a method for generating a high-precision map according to an embodiment of the present invention.

Hereinafter, the embodiments disclosed in the present specification will be described in detail with reference to the accompanying drawings, but the same or similar components are assigned the same reference numbers regardless of reference numerals, and redundant description thereof will be omitted. The suffixes "module" and "part" for components used in the following description are given or mixed in consideration of only the ease of writing the specification, and do not have distinct meanings or roles by themselves. That is, the term 'unit' used in the present invention means a hardware component such as software, FPGA, or ASIC, and 'unit' performs certain roles. However, 'part' is not limited to software or hardware. The 'unit' may be configured to reside on an addressable storage medium or it may be configured to refresh one or more processors. Thus, as an example, 'part' refers to components such as software components, object-oriented software components, class components and task components, processes, functions, properties, procedures, subroutines, segments of program code, drivers, firmware, microcode, circuitry, data, databases, data structures, tables, arrays and variables. Functions provided within components and 'units' may be combined into a smaller number of components and 'units' or further divided into additional components and 'units'.

In addition, in describing the embodiments disclosed in the present specification, if it is determined that detailed descriptions of related known technologies may obscure the gist of the embodiments disclosed in this specification, the detailed description thereof will be omitted. In addition, the accompanying drawings are only for easy understanding of the embodiments disclosed in the present specification, and the technical idea disclosed herein is not limited by the accompanying drawings, and all changes included in the spirit and scope of the present invention , should be understood to include equivalents or substitutes.

Recently, as interest in autonomous driving has increased, the need for high-definition maps for autonomous vehicles is emerging. The high-precision map may include lanes and road signs necessary for driving an autonomous vehicle, and in this case, very high accuracy is required for location information such as lanes or road signs.

Conventionally, in order to increase the accuracy of a high-precision map, a high-precision map is generated by an operator directly annotating road signs and lanes of roads from aerial images. However, in order to accurately classify the location and classification of road signs and lanes included in the aerial image, skilled workers are required. have. That is, there were problems such as annotating work for generating high-precision maps requiring a long time and a lot of manpower.

In relation to the production of high-precision maps, a method using images taken from a moving vehicle has also been proposed. However, in this case, since the measured road sign or the position of the lane is described with the measurement vehicle as the center, it may be used directly while driving the vehicle, but may not be suitable for making a high-precision map. That is, in order to find out the road sign or the geographic coordinates of the lane, accurate location information of the measurement vehicle is required, but the location of the vehicle in an area where the Global Positioning System (GPS) is inaccurate (for example, an area where vehicles are densely populated). Since it is difficult to accurately grasp In addition, in the case of being obscured by other vehicles or the angle of view of the photographing device does not appear, there is a problem such as not being able to find a road sign or a location of a lane regardless of the performance of the detection device.

Meanwhile, according to the apparatus for generating a high-precision map according to an embodiment of the present invention, since it is possible to automate the annotation operation of the high-precision map, it is possible to reduce the time and cost required for the annotation operation. In addition, since the high-precision map generating apparatus according to an embodiment of the present invention recognizes a road sign or a lane based on an aerial image, it is possible to solve problems that may occur when using an image captured by a driving vehicle.

That is, since the high-precision map generating apparatus according to an embodiment of the present invention uses an aerial image that knows in advance the one-to-one correspondence between image coordinates and geographic coordinates, it is easy to grasp the actual geographic coordinate value of the road sign detected in the aerial image. it is possible to do Also, in the case of aerial video, it is created by synthesizing images taken at various points of time and from various shooting angles. A corresponding road sign or lane may be displayed in the aerial image synthesized with Therefore, since the conventional occlusion phenomenon of road signs can be reduced by using the aerial image, it is possible to easily find the positions of road signs or lanes. Hereinafter, a high-precision map generating apparatus according to an embodiment of the present invention will be described.

1 is a block diagram showing an apparatus for generating a high-precision map according to an embodiment of the present invention. Referring to FIG. 1 , a high-precision map generating apparatus 100 according to an embodiment of the present invention may include a road sign recognition unit 110 , a lane recognition unit 120 , and a map generating unit 130 .

The road sign recognition unit 110 may recognize a road sign of a road included in the aerial image A by using the area box detection neural network, and generate an area box indicating location information and direction information of the road sign. .

Here, the area box may be an Oriented Bounding Box (OBB), as shown in Fig. 2(a), and the area box detection neural network may generate an area box using the SCRDet algorithm. have.

Specifically, when using the SCRDet algorithm, a feature map for the aerial image (A) can be calculated using a convolutional network, and then the feature map is input to the area suggestion neural network, Candidate locations (regions of interest, ROIs) in which road signs may exist in the image A may be calculated. The feature map including the candidate positions is input back into the convolutional neural network to perform region box regression and class classification. In this process, correction of each candidate position region and direction prediction of road signs are performed. That is, the location information and direction information of the road sign may be determined to generate an area box.

However, when recognizing a road sign in the aerial image (A), as shown in FIG. 3( a ), even though the object has the same direction, some road signs may be mistakenly recognized as having a different direction. there was. In order to solve this problem, the road sign recognition unit 110 may correct the direction of the road sign to recognize the correct direction.

Specifically, as shown in FIG. 3(b), in general, a direction may exist in a road, and the direction of the road may have a high correlation with the direction of a road sign. Accordingly, the road sign recognition unit 110 may correct the direction of the road sign as shown in FIG. 3( c ) in consideration of the direction of the road included in the aerial image A .

Here, the direction information of the road may be generated using a road prediction neural network, and the road prediction neural network may output a matrix including a predicted direction angle of each pixel in the aerial image A. In this case, the road prediction neural network may be trained to reduce the L2 loss with respect to the ground truth direction for each pixel in the road area using sample aerial images. Accordingly, the direction of the road included in the aerial image A may be confirmed, and the road sign recognition unit 110 may correct the direction of the road sign in consideration of the direction of the road.

According to an embodiment, it is also possible for the road sign recognition unit 110 to correct the direction of the corresponding road sign in consideration of the direction of the road and the directions of the adjacent road sign. That is, as shown in FIG. 3( c ), a plurality of road signs may be arranged adjacent to each other for each lane, and they may be grouped into a group of road signs. Specifically, when the distance between adjacent road signs is less than or equal to a predetermined reference value and the difference in direction angles between adjacent road signs is less than or equal to a predetermined reference value, the road sign recognition unit 110 includes a plurality of road signs belonging to the same group. can be seen as

Thereafter, by using the average value of direction information of road signs included in the same group and the average value of road direction information, representative direction information of each grouped road sign may be generated. Thereafter, the representative direction information may be set as direction information of each grouped road sign. That is, all road signs included in the same group may be set to have the same direction.

Specifically, as described in Equation 1 below _{, after obtaining the average value (θ b ) of the direction information of the road signs and the average value (θ r} ) of the road direction information of the road, the average value of the direction information of the road signs (θ _{b )} ) and the weight (1-λ) of the average value (θ _r ) of the road direction information of the road, the representative direction information (θ) can be calculated.

[Equation 1]

θ = λ*θ _b + (1- λ)* θ _r (where 0 ≤ λ ≤ 1)

Here, the weight λ may be a value that varies depending on the characteristics of the aerial image A, and accordingly, representative direction information may be calculated while adjusting the weight λ. Furthermore, by constructing a neural network through learning of the weight (λ), it is possible to calculate the optimum representative direction information by applying the optimum weight (λ) according to the input aerial image (A).

The lane recognition unit 120 uses a semantic extraction neural network to recognize road objects included in the aerial image A, and specifies a lane in the road according to the input lane location information P to determine the type of each lane can be classified.

Specifically, as shown in FIG. 4 , the lane recognition unit 120 may classify each pixel included in the aerial image A into any one of a plurality of road objects by using a semantic extraction neural network. A semantic map M including objects can be created. Here, the road object may include a background, a road, a lane, a road sign, and the like, and in addition, various objects that may be included in the aerial image A may also be included in the road object. According to an embodiment, it is also possible to further subdivide the road object according to the type of lane.

The meaning extraction neural network can generate a feature vector for each pixel by calculating the probability that the pixels correspond to each road object. can be classified.

Here, the meaning extraction neural network may be trained in advance to generate a semantic map M corresponding to the input aerial image A. Specifically, the deep neural network can output a pseudo semantic map corresponding to a sample aerial image input, and the deep neural network can be trained to reduce the error between the pseudo semantic map and the ground truth semantic map. have. Here, as a loss function for learning, cross-entropy or focal loss may be used. Accordingly, the deep neural network may classify pixels included in the input aerial image A into a background, road, lane, road sign, etc., and give each meaning.

Thereafter, the lane recognition unit 120 may receive lane position information P indicating the positions of the lanes, set pixels corresponding to the lane position information P in the semantic map M as lanes, and set the lane types can be classified.

Here, the lane location information P may be generated by specifying the location of the corresponding lane displayed in the aerial image A after the operator identifies each lane included in the aerial image A. That is, in order to increase the accuracy of the lane position, the operator may directly specify the position of each lane to generate the lane position information P.

In this case, the lane location information P may be generated by the operator separately from the high-precision map generating device 100, and the lane recognition unit 120 receives the lane location information P for the aerial image A from the outside. can receive However, according to an embodiment, it is also possible for the high-precision map generating apparatus 100 to provide an interface capable of generating the lane location map P to the operator. For example, after the received aerial image (A) is displayed to the operator, the operator may provide an interface for inputting to specify the positions of the corresponding lanes displayed in the aerial image (A).

That is, as shown in Fig. 2(a), it is possible to classify lanes using the semantic map M, but the reliability of location information for pixels classified as lanes in the semantic map M is relatively can be as low as Accordingly, the lane recognition unit 120 may use the lane position information P input by the operator to specify the position of the lane in the semantic map M.

Here, when the lane location information P is provided as a set of points, the lane location information P can be converted into a mask form, and a mask is applied to the semantic map M to each lane. A set M' of the corresponding pixels may be extracted.

Thereafter, the lane recognition unit 120 may generate a lane feature vector corresponding to each lane by sampling the extracted feature vectors of the pixels. That is, the set of pixels corresponding to the lane in the semantic map M can be specified by using the lane feature vector. Here, the lane recognition unit 120 may generate the lane feature vector by applying an average of the extracted feature vectors of the pixels or various sampling techniques for them.

Here, since the lane feature vectors are generated corresponding to the number of lanes included in the aerial image A, when the operator generates the lane location information P for n lanes, a total of n lane feature vectors are generated can be

Thereafter, the lane recognition unit 120 may distinguish the types of each lane included in the semantic map M by using the lane feature vector. Here, the types of lanes may be variously classified into a center line, a stop line, a course change restriction line, and the like, and the lane recognition unit 120 may determine which category each of the lanes corresponding to the lane feature vector corresponds to. In this case, the type of lane may be preset according to laws or rules such as the Road Traffic Act.

Meanwhile, according to an embodiment, n lane feature vectors may be input to the lane recognition unit 120 . For example, when there are n lanes in the aerial image A, n lane feature vectors may be generated. In this case, the lane recognition unit 120 outputs the n lane feature vectors, each It is possible to output a probability vector including a probability distribution that may correspond to the suboptimal classification of . That is, a probability vector each including a probability corresponding to the center line, a probability corresponding to a stop line, a probability corresponding to a course change limit line, etc. may be output.

Here, the lane recognition unit 120 may further include a neural network, which may have been previously learned for lane classification. That is, when sample suboptimal feature vectors are input, a corresponding pseudo random vector is output, which is to be learned to reduce the error between the ground truth random vector and the pseudo random vector corresponding to each sample suboptimal feature vector. can Accordingly, when a lane feature vector is input, the lane recognition unit 120 may output a corresponding probability vector, and select the classification with the highest probability value among the probability of each classification included in the probability vector for the lane. It can be selected as the final classification.

Additionally, depending on the embodiment, there may be a case where the lane location information P is not input to the lane recognition unit 120. In this case, the lane recognition unit 120 may specify a lane from the semantic map M. can Specifically, when the lane location information P is not input, the lane recognition unit 130 may extract target pixels classified into lanes from the semantic map M as shown in FIG. 2(b), The target pixels may be grouped and specified as a lane. That is, a pixel group may be formed by grouping target pixels located within a set distance from among a plurality of target pixels, and vector simplification may be performed on the corresponding pixel group to specify a linear lane. Here, for simplification, a Douglas-peucker technique or the like may be used.

After specifying the lane, the method of classifying the type of the lane by extracting the lane feature vector is the same as described above, and thus a detailed description thereof will be omitted.

The map generator 130 may generate a high-definition map B corresponding to the aerial image A using road signs and lanes.

Here, each road sign and lane may be specific according to the image coordinate system in the aerial image A. However, since the high-precision map B must be provided according to the actual geographic coordinate system, the map generator 130 can convert the coordinate values of each road sign and lanes into an actual geographic coordinate system, and use this to convert the high-precision map B ) can be created. Accordingly, the autonomous vehicle and the like can perform precise self-positioning even in a GPS shaded area, etc. by using road signs or lanes displayed on the high-precision map (B). For example, when correcting the positioning of the autonomous vehicle, lane position information may be used to correct the accuracy in the horizontal axis direction, and road signs may be used to correct the accuracy in the vertical axis direction.

In addition, since the autonomous vehicle recognizes the current location, even when the recognition device such as the sensor of the autonomous vehicle does not recognize the road sign or lane on the road, using a high-precision map, the surrounding road sign or lane, etc. Able to know. Therefore, even in this case, it is possible to drive according to traffic laws.

Additionally, since the road sign recognizing unit 110 and the lane recognizing unit 120 individually recognize a road sign and a lane, a case in which a road sign and a lane overlap in a high-precision map may occur. That is, any one of a road sign and a lane may be incorrectly recognized.

In this case, the map generator 130 may leave an object with high reliability among the recognized road sign and lane, and delete an object with low reliability. According to an embodiment, the reliability of each recognized road sign and lane may be provided from the area box extraction neural network or the meaning extraction neural network, respectively. Accordingly, the map generator 130 may compare and select an object to be deleted.

However, since the lane recognition unit 120 specifies the position of the lane by using the lane position information P, which precisely specifies the position of the lane by the operator, it is generally considered that the reliability of the lane is higher than the reliability of the road sign. can Accordingly, depending on the embodiment, when the lane and the road sign overlap, it is possible to set the lane to be left and only the road sign to be deleted.

Furthermore, according to an embodiment, it is also possible for the map generator 130 to further improve the reliability of the road sign by mixing the semantic map and the area box. Specifically, the map generator 130 may project the area box provided from the road sign recognition unit 110 into the semantic map M. Thereafter, a feature map of a region corresponding to the region box in the semantic map M may be extracted, and a feature vector of each pixel included in the feature map may also be extracted. In this case, the map generator 130 may check whether the road object corresponding to the characteristic map corresponds to the road sign. If the road object corresponding to the characteristic map does not correspond to the road sign, the road sign recognition unit 110 ), the recognized road sign can be deleted. That is, since it is determined that the corresponding area is not a road sign in the semantic map M, the area box generating neural network may consider it to be mistakenly recognized and delete it.

5 is a flowchart illustrating a method for generating a high-precision map according to an embodiment of the present invention. Here, each step of the high-definition map generating method may be performed by the high-definition map generating apparatus.

Specifically, the high-precision map generating apparatus may recognize a road sign of a road included in an aerial image by using the area box detection neural network, and generate an area box indicating location information and direction information of the road sign (S10). Here, the area box may be an Oriented Bounding Box (OBB) having directionality, and the area box detection neural network may use the SCRDet algorithm.

Here, there may exist a case where the direction of the road sign is incorrectly recognized. In this case, the high-precision map generating apparatus may correct the direction of the road sign. Specifically, the high-precision map generating apparatus may generate road direction information by recognizing the direction of the road included in the aerial image, and may correct direction information of road signs using the road direction information.

According to an embodiment, it is also possible to correct the direction of the corresponding road sign by considering the direction of the adjacent road sign along with the direction of the road. That is, the high-precision map generating apparatus can group adjacent road signs by using the location information and direction information of the road signs, and grouping using the average value of the direction information of each grouped road sign and the average value of the road direction information It is possible to generate representative direction information of each road sign. Thereafter, the representative direction information may be set as direction information of each of the grouped road signs.

In addition, the high-precision map generating apparatus can recognize road objects included in the aerial image by using a semantic extraction neural network, and classify the types of each lane by specifying a lane in the road according to the input lane location information (S20). ).

Specifically, the high-precision map generating apparatus may classify each pixel included in the aerial image into any one of a plurality of road objects by using a semantic extraction neural network, and may generate a semantic map including the road objects.

Thereafter, when lane location information indicating the location of the lanes is received, pixels corresponding to the lane location information on the semantic map may be set as lanes, and types of the set lanes may be classified. Here, the lane location information may be generated by specifying the location of the corresponding lane displayed in the aerial image after the operator identifies each lane included in the aerial image. That is, in order to increase the accuracy of the lane position, the operator may directly specify the position of each lane to generate lane position information.

In this case, the operator may generate the lane position information separately from the high-precision map generating device, and may receive lane position information for the corresponding aerial image from the outside. However, according to an embodiment, it is also possible for the high-precision map generating apparatus to provide an interface for generating a lane location map to the operator.

On the other hand, when lane location information indicating the location of lanes is input, feature vectors of pixels corresponding to the lane location information can be extracted from the semantic map, and the extracted feature vectors are sampled to generate a lane feature vector corresponding to the lane. can Thereafter, types of lanes may be distinguished using the lane feature vector.

Additionally, depending on the embodiment, there may be a case in which lane location information corresponding to some aerial images is not input. In this case, the high-precision map generating apparatus can specify a lane from the semantic map. That is, when lane location information is not input, target pixels classified into lanes may be extracted from the semantic map, and target pixels may be grouped to specify lanes.

Specifically, among a plurality of target pixels classified into lanes, target pixels positioned within a set distance may be bundled to form a pixel group. Since a plurality of lanes may exist in the aerial image, target pixels located within a set distance may be grouped into the same pixel group in order to form a pixel group for each lane.

Thereafter, each lane may be specified by performing vector simplification on the pixel group. Since the pixel group may not appear in a single straight line, simplification may be performed for linearization, and a Douglas-peucker technique may be used according to an embodiment.

Meanwhile, although FIG. 5 shows that the step of classifying the lane type (S20) is performed after the step (S10) of generating the area box, the present invention is not limited thereto, and the step of classifying the type of the lane (S20) is performed first, or it is also possible that the step of generating the area box (S10) and the step of classifying the type of the lane (S20) are simultaneously performed in parallel.

Thereafter, the high-definition map generating apparatus may generate a high-precision map corresponding to the aerial image by using the road sign and the lanes (S30). Here, each road sign and lane may be specific using an image coordinate system in an aerial image. However, since the high-precision map must be provided in the actual geographic coordinate system, the high-precision map generating apparatus can generate the high-precision map B by converting the coordinate values of each road sign and lanes into the actual geographic coordinate system.

On the other hand, a case may occur in which each road sign and lane recognized by the high-definition map generating apparatus overlap within the high-definition map. That is, any one of a road sign and a lane may be incorrectly recognized. In this case, the high-precision map generating apparatus may leave an object with high reliability among the recognized road sign and lane and delete an object with low reliability. Since the reliability of each road sign and lane can be provided from an area box extraction neural network or a meaning extraction neural network, an object to be deleted can be selected by comparing them. However, in the case of a lane, since the operator specifies the position of the lane by using the lane position information P, which precisely specifies the position of the lane, in general, the reliability of the lane is higher than the reliability of the road sign. Accordingly, depending on the embodiment, when the lane and the road sign overlap, it is possible to set the lane to be left and only the road sign to be deleted.

Furthermore, depending on the embodiment, it is possible to further improve the reliability of the road sign by mixing the semantic map and the area box. Specifically, by projecting the area box into the semantic map, it is possible to extract the feature map of the area corresponding to the area box from the semantic map. Thereafter, by extracting the feature vector of each pixel included in the feature map, it can be checked whether the road object corresponding to the feature map corresponds to the road sign. If the road object corresponding to the feature map does not correspond to the road sign, The relevant road sign may be deleted. That is, since it is determined that the corresponding area is not a road sign in the semantic map, the area box generating neural network may consider it to be incorrectly recognized and delete it.

The present invention described above can be implemented as computer-readable codes on a medium in which a program is recorded. The computer-readable medium may continuously store a computer-executable program, or may be temporarily stored for execution or download. In addition, the medium may be various recording means or storage means in the form of a single or several hardware combined, it is not limited to a medium directly connected to any computer system, and may exist distributedly on a network. Examples of the medium include a hard disk, a magnetic medium such as a floppy disk and a magnetic tape, an optical recording medium such as CD-ROM and DVD, a magneto-optical medium such as a floppy disk, and those configured to store program instructions, including ROM, RAM, flash memory, and the like. In addition, examples of other media may include recording media or storage media managed by an app store for distributing applications, sites supplying or distributing other various software, and servers. Accordingly, the above detailed description should not be construed as restrictive in all respects but as exemplary. The scope of the present invention should be determined by a reasonable interpretation of the appended claims, and all modifications within the equivalent scope of the present invention are included in the scope of the present invention.

The present invention is not limited by the above embodiments and the accompanying drawings. For those of ordinary skill in the art to which the present invention pertains, it will be apparent that the components according to the present invention can be substituted, modified and changed without departing from the technical spirit of the present invention.

100: high-precision map generator
110: road sign recognition unit
120: lane recognition unit
130: map generation unit

Claims (13)

Recognizing a road sign of a road included in an aerial image by using an area box detection neural network, and generating an area box indicating location information and direction information of the road sign;
recognizing road objects included in the aerial image using a semantic extraction neural network, specifying a lane in the road according to the input lane location information, and classifying the type of each lane; and
and generating a high-precision map corresponding to the aerial image by using the road sign and lanes.
The method of claim 1, wherein generating the area box comprises:
A high-precision map generating method, characterized in that by recognizing the direction of the road included in the aerial image, generating road direction information, and correcting the direction information of the road sign using the road direction information.
The method of claim 2, wherein generating the area box comprises:
Grouping adjacent road signs by using the location information and direction information of the road sign, and using the average value of the direction information of each grouped road sign and the average value of the road direction information for each grouped road sign A method for generating a high-precision map, characterized in that representative direction information is generated, and the representative direction information is set as direction information of each of the grouped road signs.
The method of claim 1, wherein classifying the type of the lane comprises:
A method for generating a high-precision map, characterized in that by using the meaning extraction neural network, each pixel included in the aerial image is classified into any one of a plurality of road objects, and a semantic map including the road objects is generated.
5. The method of claim 4, wherein classifying the type of the lane comprises:
When lane position information indicating the positions of the lanes is received, pixels corresponding to the lane position information on the semantic map are set as the lanes, and types of the set lanes are classified.
The method of claim 5, wherein classifying the type of the lane comprises:
Extracting feature vectors of pixels corresponding to the lane location information from the semantic map, generating a lane feature vector corresponding to the lane by sampling the extracted feature vectors, and using the lane feature vector A method for generating a high-precision map, characterized in that the
5. The method of claim 4, wherein classifying the type of the lane comprises:
When the lane location information is not input, the target pixels classified as the lanes are extracted from the semantic map, and the target pixels are grouped to specify the lanes.
The method of claim 7, wherein classifying the type of the lane comprises:
Among a plurality of target pixels, a pixel group is formed by grouping target pixels located within a set distance from each other, and vector simplification is performed on the pixel group to specify each lane. Way.
The method of claim 1, wherein the generating of the high-precision map comprises:
The method for generating a high-definition map, characterized in that when the position of the road sign and the lane overlap each other in the high-precision map, the road sign is deleted.
5. The method of claim 4, wherein the generating of the high-precision map comprises:
By projecting the area box into the semantic map, a feature map of the area corresponding to the area box is extracted, and a road object corresponding to the feature map is obtained by using the feature vector of each pixel included in the feature map. A method for generating a high-precision map, characterized in that it is checked whether it corresponds to a road sign.
11. The method of claim 10, wherein the generating of the high-precision map comprises:
If the road object corresponding to the characteristic map does not correspond to the road sign, the road sign recognized by the road sign recognition unit is deleted.
A computer program stored in a medium in combination with hardware to execute the method for generating a high-precision map according to any one of claims 1 to 11.
a road sign recognition unit for recognizing a road sign of a road included in an aerial image by using an area box detection neural network, and generating an area box indicating location information and direction information of the road sign;
a lane recognition unit for recognizing road objects included in the aerial image by using a semantic extraction neural network, specifying a lane on the road according to the input lane location information, and classifying the type of each lane; and
and a map generator configured to generate a high-precision map corresponding to the aerial image by using the road sign and lanes.

Images