KR101926561B1 - Road crack detection apparatus of patch unit and method thereof, and computer program for executing the same - Google Patents

Abstract

The present invention relates to an apparatus for detecting a patch unit road crack based on deep learning using a black box image, a method thereof, and a computer program stored in a computer-readable recording medium to execute the method. The apparatus for detecting a road crack according to the present invention detects a road crack through model training based on deep learning in a patch unit by using a black box image for a vehicle. The apparatus for detecting a road crack extracts an image from a black box image for a vehicle to detect a road crack. A class of the extracted image is classified into a road crack, a lane, and a non-damaged road. The classified image is subjected to model training based on deep learning in a patch unit, and the trained model is applied to a black box image to detect a crack of a road surface in a patch unit. According to the present invention, the accuracy of detection of a road crack can be improved by detecting the road crack in a patch unit on a road surface. Since a black box image is used, it is possible to easily construct a system for detecting a road crack. The system construction costs can also be reduced.

Description

Field of the Invention [0001] The present invention relates to a patch crack detection apparatus and a patch crack detection apparatus based on deep-learning using a black box image, and a computer program stored in a computer-readable recording medium for executing the method. PROGRAM FOR EXECUTING THE SAME}

More particularly, the present invention relates to a road crack detecting apparatus for detecting a crack on a road surface in units of patches based on a deep running based on a black box image for a vehicle, .

The present invention also relates to a computer program stored in a computer-readable recording medium for executing a road crack detection method of a patch-unit road crack detection apparatus based on a deep-run based on a black box image.

With the development of the industry, the construction of roads, which are social overhead capital, has been widely carried out. Due to the construction of a wide range of roads, the cost of maintenance and the time required also have a significant impact on national operations. Therefore, efficient maintenance is required in the repair of pavement roads.

Conventionally, there is no way to reliably measure the cracks on the pavement, so that a specific investigator who is highly experienced in the field of crack inspection can visually check the pavement road and then determine the cracked state of the pavement road . However, since this method lacks objectivity of crack detection, there is a limit to systematically and integratively managing pavement roads.

Due to such a problem, various attempts have been made to detect road cracks by installing a pavement road survey vehicle or a camera for photographing the surface of the road at a specific position and photographing the road surface using the camera.

For example, techniques for extracting a road surface using a deep running technique on a road taken image, and a technique for detecting a crack using a deep running technique on road surfaces such as concrete, metal, and asphalt have already been disclosed variously. In particular, existing techniques for detecting road cracks require a great deal of effort in data collection because they are taken at a constant distance from the road cracked surface for image acquisition.

For example, a method of managing roads by camera images is disclosed in 'Method for detecting cracks in pavement' in Patent No. 10-0443667. This technology allows the vehicle to detect the cracked portion of the package by photographing the surface of the road while moving along the packed road and imaging the photographed image.

In order to detect cracks in the road from the image of the applied surface, these methods generally capture the road surface image only for roads and acquire road surface images, and through this, a system for detecting road cracks requires precision, Etc., it takes a lot of cost to construct the system. In addition, monitoring of road pavement status is performed for some road sections, and it is difficult to prevent preventive maintenance of defective roads.

In addition, when the cracks of the road surface are extracted using the existing CCTV camera, the road pavement status is monitored for some road sections. Therefore, preventive maintenance of defective roads is difficult. In addition, this method consumes a lot of cost for system construction, and when a crack of a road surface is detected from a high resolution image by utilizing a deep learning technology, it is analyzed through a complex image recognition technology and a training process, have.

In addition, when a black box image is directly applied to a technique for detecting an existing road crack, there arises a problem of recognizing other vehicles, lanes, buildings, and the like included in the black box image as cracks.

Therefore, the present invention aims at designing a crack detection model suitable for an image using a black box image which is easier to acquire data than existing road crack detection techniques.

Korean Patent Laid-Open Publication No. 10-2016-0039316 (Publication date 2016.04.11.) Korean Registered Patent No. 10-1543342 (Published on Aug. 20, 2015) Korean Registered Patent No. 10-1772916 (Published on Aug. 31, 2017) Korean Registered Patent No. 10-1178772 (Published on September 07, 2012)

The apparatus and method for patch cracking based on deep running using a black box image according to the present invention have the following problems.

First, we want to detect the road crack using the black box image of the vehicle.

Second, we try to extract the road cracks by training and analyzing the deep - run - based model of the images from the vehicle black box images in patch units.

Third, it is aimed to improve the accuracy of road crack detection according to various time of day, road condition, lighting condition, etc. by using a black box image of a vehicle.

The solution of the present invention is not limited to those mentioned above, and other solutions not mentioned can be clearly understood by those skilled in the art from the following description.

In order to achieve the above-mentioned objects, the road crack detecting apparatus of the present invention is characterized in that road cracks are detected in units of patches based on deep running using a vehicle black box image. The road crack detecting apparatus of the present invention can detect a road crack using an image of a black box for a vehicle, thereby making it possible to photograph a road surface including an area having a large resolution of a black box image, No additional work is required to acquire, which saves cost and time.

The road crack detecting apparatus of the present invention according to this feature comprises: a data collecting unit for collecting image data photographed from at least one black box; An image extracting unit for extracting a plurality of images from the image data collected by the data collecting unit; A patch image generation unit for classifying the classes of the road crack, the lane and the undamaged road by selecting the image extracted by the image extracting unit for each pixel of a predetermined size, and generating a patch image for each of the classified classes; A model training unit for training each patch image generated from the patch image generating unit to extract a road crack using a neural network model; And a crack detector for extracting an image by receiving image data from an arbitrary black box and applying a neural network model trained from the model training unit to the extracted image to extract road cracks.

In one embodiment of this aspect, the patch image generating unit may include: A patch image is generated for each road crack, lane, and undamaged road in a patch unit of 40 × 40 pixels.

In another embodiment, the patch image generator comprises: An image extracted from the image extracting unit is extracted as an image having a predetermined size to limit the crack detection region and the patch image is generated in a limited crack detection region.

In yet another embodiment, the patch image generator may include: In the case of the patch image of the road crack, the crack detection area is overlapped to detect a road crack at the edge.

In another embodiment, the model training unit includes: We use VGG 16 network among neural network models.

In another embodiment, the model training unit includes: A patch image is trained for each road crack, lane, and undamaged road to calculate and output the probability of detecting road cracks.

In another embodiment, the model training unit includes: Determine the weights for each of the road cracks, lanes and undamaged roads and apply them to the model training. Determine the layer while gradually reducing the size of the patch image to calculate the probability for each crack, lane and undamaged road And outputs it.

According to another aspect of the present invention, there is provided a road crack detection method of a road crack detection apparatus.

According to another aspect of the present invention, there is provided a road crack detection method for a road crack detection apparatus, comprising the steps of: collecting image data from a plurality of black boxes and extracting a plurality of frames from the collected image data; When a desired point is selected from the frame extracted by the operator, generating a patch image of a predetermined size for each road crack, lane, and undamaged road centered on the selected point; Training patch images for road cracks, lanes and undamaged roads using a neural network model; Extracting a plurality of frames when an arbitrary black box image is input, and extracting an image of the road surface in the extracted frame; Extracting an image of a road surface with a predetermined size image to limit a crack detection area; Dividing an image of a road surface with the crack detection area limited by patch units of a predetermined size; And detecting a road crack by applying a trained neural network model to an image of a road surface divided into patch units.

In one embodiment of this aspect, the step of generating the patch image comprises: A patch image is generated for each road crack, lane, and undamaged road in a patch unit of 40 × 40 pixels.

In another embodiment, training using the neural network model comprises: A probability for determining whether a road crack is a road crack is calculated and outputted, and when the calculated probability is equal to or larger than a certain size, .

In yet another embodiment, extracting an image for the road surface comprises: In the extracted frame, the image of the road surface is extracted by excluding the portion not the road surface and the image of the front portion of the vehicle.

In another embodiment, the step of limiting the crack detection region comprises: Limit the crack detection area to 720 x 240 pixel size image.

In yet another embodiment, dividing the region comprises: The image of 720x240 pixels is divided into 40x40 pixels so that the images are overlapped by 20 pixels each in the up, down, left and right directions.

According to still another aspect of the present invention, there is provided a computer program stored in a computer-readable recording medium in combination with hardware for executing a road crack detection method.

The apparatus and method for patch crack detection based on deep running using a vehicle black box image according to the present invention have the following effects.

First, by detecting a road crack using an image of a vehicle black box, it is possible to photograph a road surface including an area having a large resolution of a black box image, to easily acquire data, and to perform additional operations to acquire data It is possible to save cost and time.

Second, accurate road crack detection is possible by detecting a road crack from a black box image of a vehicle by a patch unit, and it is easy to construct a system for detecting a road crack, and it can be implemented at a low cost.

Third, since the state of the road can be quickly identified by observing cracks in many road sections within a short time, the product can be differentiated in the road crack detection device, and the marketability is very high.

The effects of the present invention are not limited to those mentioned above, and other effects not mentioned can be clearly understood by those skilled in the art from the following description.

1 is a block diagram showing the configuration of a patch unit road crack detection apparatus in a black box image utilizing deep running according to the present invention;
2 is a flowchart showing a patch unit road crack detection procedure of a patch unit road crack detection apparatus in a black box image utilizing deep running according to the present invention,
3 is a flow chart showing the neural network model training routine shown in FIG. 2,
4 is a view illustrating an image extracted from a black box image according to an embodiment of the present invention,
5 is a diagram illustrating a neural network model structure according to an embodiment of the present invention.
6 is a diagram showing a road surface extraction state according to an embodiment of the present invention,
FIG. 7 is a view showing a crack area division state according to an embodiment of the present invention, and FIG.
8 is a view showing a road crack detection state according to an embodiment of the present invention.

Hereinafter, embodiments of the present invention will be described with reference to the accompanying drawings so that those skilled in the art can easily carry out the present invention. It will be apparent to those skilled in the art that various modifications and variations can be made in the present invention without departing from the spirit or scope of the invention. Wherever possible, the same or similar parts are denoted using the same reference numerals in the drawings.

The terminology used herein is for the purpose of describing particular embodiments only and is not intended to be limiting of the invention. The singular forms as used herein include plural forms as long as the phrases do not expressly express the opposite meaning thereto.

Means that a particular feature, region, integer, step, operation, element and / or component is specified and that other specific features, regions, integers, steps, operations, elements, components, and / It does not exclude the existence or addition of a group.

All terms including technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention belongs. Predefined terms are further interpreted as having a meaning consistent with the relevant technical literature and the present disclosure, and are not to be construed as ideal or very formal meanings unless defined otherwise.

The present invention uses a black box image for a vehicle for a road crack detection, a deep training based model training on a patch unit basis, and a trained model is applied to a black box image to detect a crack on a road surface.

According to the present invention, it is possible to increase the accuracy of road crack detection, to easily construct a system for detecting road cracks, and to reduce the construction cost of a system.

Hereinafter, the present invention will be described in detail with reference to the drawings.

FIG. 1 is a block diagram showing the configuration of a patch unit road crack detection apparatus in a black box image utilizing deep learning according to the present invention.

Referring to FIG. 1, the road crack detecting apparatus 100 of the present invention detects a crack on a road surface by using a video of a vehicle black box 10, The model is trained and a trained model is applied from the image of the black box image to detect the road crack in pixel unit.

To this end, the road crack detection apparatus 100 of the present invention collects moving image data from a plurality of vehicle black boxes 10, extracts a plurality of images for detecting a road crack from the collected moving images, Classes are classified into road cracks, lanes, and undamaged roads in the image, and the images according to the classified classes are applied to the deep-run-based model in units of patches.

The road crack detecting apparatus 100 of the present invention extracts an image from the image of the black box 10 and applies a trained model to the extracted image to detect a road crack in units of patches.

Specifically, the road crack detecting apparatus 100 of the present invention includes a data collecting unit 110, an image extracting unit 120, a patch image generating unit 130, a model training unit 140, and a crack detecting unit 150 .

The data collecting unit 110 collects moving image data on the front road surface photographed from each of the plurality of vehicle black boxes 10. The data collecting unit 110 is connected to the black box 10 in real time via a wired or wireless communication network and collects the moving image data or transmits the image data through the data storage medium storing the image of the black box 10. [ Lt; / RTI >

The image extracting unit 120 receives the moving image data collected from the data collecting unit 110 and extracts a plurality of images for detecting a road crack. The image extracting unit 120 arbitrarily extracts a plurality of frames for detecting a crack in the road surface from the moving image data of the collected black box 10, for example. In this embodiment, hundreds of frames are extracted from the moving image data of the black box 10. These frames are then used to train deep-run-based CNN models.

In this embodiment, the image extraction is based on the Convolution Neural Network (CNN) model proposed by K. Simonyan and A. Zisserman in VGG 16 network proposed in 2014 to extract the road surface from the image of the vehicle black box 10 It uses python coding.

When a desired point is selected by the operator in the image extracted from the image extracting unit 120 in order to detect a crack in units of patches, the patch image generating unit 130 generates an image of a predetermined size, that is, a patch Create an image.

In this embodiment, the patch image has a size of 40 x 40 pixels. A patch image is created for each of the three classes. That is, the patch images are respectively generated as data for cracks, lanes and undamaged roads.

The model training unit 140 processes each of the patch images generated by the patch image generating unit 130 to train using the deep learning based model. The model training unit 140 calculates the probability of detecting a road crack by outputting a patch image for each of cracks, lanes and undamaged roads, and outputs the calculated probability.

In addition, the model training unit 140 receives the black box image and processes it so as to apply a trained model for cracks, lanes, and undamaged roads to detect road cracks on a patch-by-patch basis.

In this embodiment, the model training unit 140 utilizes a neural network (CNN) model. The CNN model includes, for example, three layers: a convolution layer (Conv layer), a pooling layer (Pool layer), and a fully-connected layer Hereinafter FC layer).

Conv The filter composing the Conv layer is subjected to a convolution operation with the result of the previous layer, which becomes the input value of the next layer. The Pool layer plays a role of downsampling the result of the previous layer to a small size. The FC layer is connected to all the nodes of the previous layer and mainly constitutes the rear part of the CNN model. The final output value of CNN model is composed of FC layer and it calculates the probability belonging to each class. When the CNN model is used as an image classifier, the class having the highest probability of the final output value is predicted as a class for the input image.

The present invention tests whether the model has high performance by adjusting parameters such as the depth and the size of a layer, and finally, a neural network (CNN) model structure is designed as shown in FIG. The CNN model determines the weights for the layers for each class, applies them to the model training, decides the layers while gradually reducing the size of the patch image, and finally calculates the probability for cracks, lanes, and unimpaired roads And outputs it. The image belongs to the class with the highest probability among the three classes. If the probability of cracking is the highest, it is judged as a crack.

When the image is input from the black box, the crack detecting unit 150 detects a road crack by applying a model trained by the model training unit 140. The crack detecting unit 150 calculates a predicted value by classifying each patch image by applying a model trained by the model training unit 140. If the probability of cracking among the three classes is the highest, it is judged as a crack.

Therefore, the road crack detection apparatus 100 of the present invention uses an image of the vehicle black box 10, extracts an image in units of patches, trains the model to a deep learning-based model, applies the trained model to a black box image, Detects road cracks in units.

FIG. 2 is a flowchart showing a patch unit road crack detection procedure of a patch unit road crack detection apparatus in a black box image utilizing deep learning according to the present invention, and FIG. 3 is a flowchart showing a neural network model training routine shown in FIG. to be.

Referring to FIGS. 2 and 3, the road crack detection apparatus 100 of the present invention collects image data from a plurality of black boxes in step S160, and extracts a plurality of images, that is, frames, from the image data in step S162.

If the extracted frame is trained by using the neural network model, a desired point is selected from the frame extracted by the operator in step S164, and an image of a predetermined size, that is, a patch image is generated centering on the selected point. In this embodiment, the patch image has a size of 40 x 40 pixels. The patch image is generated for three classes, crack, lane, and undamaged road, respectively.

In step S166, a patch image for each of the crack, lane and undamaged road is trained using a neural network model. Specifically, as shown in FIG. 3, the neural network (CNN) model training routine S166 classifies the class in step S190 and calculates a model predicted value for each class classified in step S192 to calculate a probability And outputs it. If the probability is greater than or equal to a certain size in step S194, it is judged as a crack. If the probability of cracking is the highest among the three classes, it is judged as a crack.

This trained model is applied to detect a road crack when a black box image is input later. That is, referring again to FIG. 2, in step S170, when a frame is extracted to detect a road crack from a black box image, an image of the road surface is extracted from the extracted frame. At this time, the extracted frame extracts the road surface, the portion not the road surface, and the front portion of the vehicle (that is, the bonnet) together.

In step S172, the road surface image is extracted as an image of a predetermined size to limit the crack detection area. In this embodiment, the road surface image is extracted with a size of 720 x 240 pixels. This road surface image includes areas along road cracks, lanes and undamaged roads.

In step S174, the road surface image is divided into patches of a predetermined size. The trained neural network model is applied to the road surface image divided into patch units in step S176, and the class is classified into road cracks, lanes and unimpaired roads in step S178. Then, a road crack is detected using the class classified in step S180.

The road crack detection method of the road crack detection apparatus 100 described above can be provided as a computer program. This computer program is provided as software for detecting road cracks and is used in conjunction with a plurality of vehicle black boxes 10 on a computer system provided with a road crack detecting apparatus 100, for example, a portable computer, a personal computer, .

Such a computer program may be recorded on a computer-readable recording medium for executing the road crack detection method of the road crack detecting apparatus 100 of the present invention described above.

For example, the road crack detection apparatus 100 of the present invention can be implemented by a computer, and the road crack detection method can be implemented by a computer-readable computer program on a computer readable recording medium. The computer-readable recording medium may include various types of recording devices in which data that can be read by a computer is stored. For example, the recording medium includes not only a ROM, a RAM, an optical disk, a magnetic tape, a floppy disk, a hard disk, a nonvolatile memory and the like but also distributed to computers interconnected via a network such as the Internet, A computer-readable computer program may be stored and executed.

4 is a view showing an image extracted from a black box image according to an embodiment of the present invention.

Referring to FIG. 4, in an embodiment of the present invention, three classes of data are created to train a deep learning model for detecting cracks in patch units of 40 × 40 pixels on the extracted road surface. After selecting, for example, 100 images 200 from which to extract data from the black box image, the image 200 is displayed on the screen through Pichuan coding and a desired point is selected. A 40x40 pixel image centered on the selected point is created. In this way, data is generated for three classes of road cracks, lanes, and undamaged roads. In FIG. 4, red portions 202 are selected at the crack portions.

5 is a diagram illustrating a neural network model structure according to an embodiment of the present invention.

Referring to FIG. 5, the neural network model 210 of the present invention includes, for example, three layers: a convolution layer (Conv layer), a pooling layer (Pool layer) And a fully-connected layer (FC layer). Conv The filter composing the Conv layer is subjected to a convolution operation with the result of the previous layer, which becomes the input value of the next layer. The Pool layer plays a role of downsampling the result of the previous layer to a small size. The FC layer is connected to all the nodes of the previous layer and mainly constitutes the rear part of the CNN model. The final output value of CNN model is composed of FC layer and it calculates the probability belonging to each class.

When the CNN model is used as an image classifier, the class having the highest probability of the final output value is predicted as a class for the input image. The present invention tests whether the model has high performance by adjusting parameters such as the depth and the size of a layer, and finally, a neural network (CNN) model structure is designed as shown in FIG. This neural network (CNN) model 210 determines weights for the layers for each class and applies them to the model training. The layer is determined while gradually reducing the size of the patch image, and finally, each of the cracks, lanes, And outputs the calculated probability. If the probability of cracking is the highest among the three classes, it is judged as a crack.

FIG. 6 is a view showing a road surface extraction state according to an embodiment of the present invention, FIG. 7 is a view showing a crack area division state according to an embodiment of the present invention, and FIG. 8 is a cross- These are views showing a crack detection state.

Referring to Figs. 6 to 8, in the present embodiment, the frame extracted from the black box image is extracted together with the road surface, the portion not on the road surface, and the front portion (bonnet) of the vehicle. The road surface image 220 thus excludes the non-road surface portion 204 and the bonnet portion 206 of the vehicle.

In order to train using the neural network model, a patch image 230 for the road surface area is created, as shown in FIG. 7, for the road surface image 220. At this time, the patch image 230 limits the crack detection area to the patch image 232 of 720 x 240 pixel size.

The designed CNN model 210 selects a patch having a size of 40 × 40 pixels within a patch image 232 having a size of 720 × 240 pixels, and classifies the patch as a road crack, a lane, and an unimpaired road surface class. In the case of a crack, since it may not be detected if it is at the edge of the patch image 232, it is designed so that a crack part at the edge in the patch image 232 can also be detected.

For this purpose, a rectangular image (234) located at a distance of 20 pixels in each of the upper, lower, left, and right directions from the original position is also constructed so as to be used as a test image. That is, the rectangular image 232 is a patch image generated by 20 pixels in the up, down, left, and right directions in the patch image 232. Thus, the patch image 232 and the rectangular image 234 for the road surface are superimposed to create an overlapping patch image 236, 238.

As a result of applying the CNN model to a patch having a size of 40 × 40 pixels, the road crack is detected (240) as shown in FIG. Here, patches 242 classified as road cracks are indicated by red borders.

The embodiments and the accompanying drawings described in the present specification are merely illustrative of some of the technical ideas included in the present invention. Accordingly, the embodiments disclosed herein are for the purpose of describing rather than limiting the technical spirit of the present invention, and it is apparent that the scope of the technical idea of the present invention is not limited by these embodiments. It will be understood by those of ordinary skill in the art that various changes in form and details may be made therein without departing from the spirit and scope of the invention as defined by the appended claims.

10: Black box
100: Road crack detector
110: Data collecting unit
120: image extracting unit
130: patch image generating unit
140: Model Training Department
150:

Claims (14)

A road crack detection apparatus comprising:
A data collecting unit for collecting image data photographed from at least one black box;
An image extracting unit for extracting a plurality of images from the image data collected by the data collecting unit;
A patch image generation unit for classifying the classes of the road crack, the lane and the undamaged road by selecting the image extracted by the image extracting unit for each pixel of a predetermined size, and generating a patch image for each of the classified classes;
A model training unit for training each patch image generated from the patch image generating unit to extract a road crack using a neural network model; And
Extracting a plurality of frames from the arbitrary black box image, extracting an image of the road surface from the extracted frame, applying a neural network model trained from the model training section to the extracted image, And a detector,
The model training unit includes: A patch image is trained for each road crack, lane, and undamaged road to calculate and output a probability for detecting a road crack,
Wherein the image of the road surface is extracted from the road surface by excluding the portion of the extracted road surface that is not the road surface and the front portion image of the vehicle. The method according to claim 1,
Wherein the patch image generating unit comprises:
And generates a patch image for each of the road crack, the lane, and the undamaged road in units of patches each having a size of 40 x 40 pixels. The method of claim 2,
Wherein the patch image generating unit comprises:
Extracting an image extracted from the image extracting unit as an image of a predetermined size to limit a crack detection area, and generating the patch image in a limited crack detection area. The method of claim 3,
Wherein the patch image generating unit comprises:
In the case of the patch image of the road crack, overlaps the crack detection area to detect a road crack at the edge. The method according to claim 1,
The model training unit includes:
Wherein the learning is performed using the VGG 16 network among the neural network models. delete The method according to claim 1,
The model training unit includes:
Determine the weights for each of the road cracks, lanes and undamaged roads, apply them to the model training, and calculate the probability for each of the cracks, lanes and undamaged roads by determining the layers while gradually reducing the size of the patch image And outputting the road cracks. A road crack detection method for a road crack detection apparatus comprising:
Collecting image data from a plurality of black boxes, and extracting a plurality of frames from the collected image data;
When a desired point is selected from the frame extracted by the operator, generating a patch image of a predetermined size for each road crack, lane, and undamaged road centered on the selected point;
Training patch images for road cracks, lanes and undamaged roads using a neural network model;
Extracting a plurality of frames when an arbitrary black box image is input, and extracting an image of the road surface in the extracted frame;
Extracting an image of a road surface with a predetermined size image to limit a crack detection area;
Dividing an image of a road surface with the crack detection area limited by patch units of a predetermined size; And
Detecting a road crack by applying a trained neural network model to an image of a road surface divided by a patch unit,
Training using the neural network model comprises: A probability for determining whether a road crack is a road crack is calculated and outputted, and when the calculated probability is equal to or larger than a certain size, However,
Wherein the step of extracting an image for the road surface comprises: Wherein an image of the road surface is extracted by excluding a portion of the extracted frame that is not a road surface and an image of a front portion of the vehicle. The method of claim 8,
Wherein the generating the patch image comprises:
And a patch image is generated for each of road cracks, lanes, and undamaged roads in units of patches each having a size of 40 x 40 pixels. delete delete The method of claim 8,
Limiting the crack detection region comprises:
And the crack detection area is limited to an image of 720 x 240 pixels in size. The method of claim 12,
Dividing the region comprises:
Wherein the image is divided into 40 × 40 pixel units so that the image of 720 × 240 pixels is overlapped by 20 pixels in the vertical and horizontal directions, respectively. A computer program stored in a computer-readable recording medium for executing the road crack detection method according to any one of claims 8, 9, 12 and 13, in combination with hardware.

Images