WO2020248371A1 - 道路损伤检测方法、装置、计算机设备及存储介质 - Google Patents

道路损伤检测方法、装置、计算机设备及存储介质 Download PDF

Info

Publication number
WO2020248371A1
WO2020248371A1 PCT/CN2019/102670 CN2019102670W WO2020248371A1 WO 2020248371 A1 WO2020248371 A1 WO 2020248371A1 CN 2019102670 W CN2019102670 W CN 2019102670W WO 2020248371 A1 WO2020248371 A1 WO 2020248371A1
Authority
WO
WIPO (PCT)
Prior art keywords
image
detected
road damage
threshold
value
Prior art date
Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
Ceased
Application number
PCT/CN2019/102670
Other languages
English (en)
French (fr)
Inventor
石磊
王健宗
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Ping An Technology Shenzhen Co Ltd
Original Assignee
Ping An Technology Shenzhen Co Ltd
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by Ping An Technology Shenzhen Co Ltd filed Critical Ping An Technology Shenzhen Co Ltd
Publication of WO2020248371A1 publication Critical patent/WO2020248371A1/zh
Anticipated expiration legal-status Critical
Ceased legal-status Critical Current

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T7/00—Image analysis
    • G06T7/0002—Inspection of images, e.g. flaw detection
    • G06T7/0004—Industrial image inspection
    • G06T7/0008—Industrial image inspection checking presence/absence
    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T7/00—Image analysis
    • G06T7/10—Segmentation; Edge detection
    • G06T7/11—Region-based segmentation
    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T7/00—Image analysis
    • G06T7/10—Segmentation; Edge detection
    • G06T7/136—Segmentation; Edge detection involving thresholding
    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T2207/00—Indexing scheme for image analysis or image enhancement
    • G06T2207/10—Image acquisition modality
    • G06T2207/10004—Still image; Photographic image
    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T2207/00—Indexing scheme for image analysis or image enhancement
    • G06T2207/20—Special algorithmic details
    • G06T2207/20081—Training; Learning
    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T2207/00—Indexing scheme for image analysis or image enhancement
    • G06T2207/20—Special algorithmic details
    • G06T2207/20084—Artificial neural networks [ANN]
    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T2207/00—Indexing scheme for image analysis or image enhancement
    • G06T2207/30—Subject of image; Context of image processing
    • G06T2207/30108—Industrial image inspection
    • G06T2207/30132—Masonry; Concrete

Definitions

  • This application relates to the field of intelligent decision-making, in particular to a road damage detection method, device, computer equipment and storage medium.
  • the embodiments of the present application provide a road damage detection method, device, computer equipment and storage medium to solve the problem of inaccurate road damage detection results.
  • a road damage detection method including:
  • a road damage detection device including:
  • the resolution judgment module is used to obtain the image to be detected and determine whether the image resolution of the image to be detected is greater than the resolution threshold;
  • a segmentation processing module configured to perform segmentation processing on the image to be detected when the image resolution of the image to be detected is greater than the resolution threshold to obtain N image blocks to be detected, where N is a positive integer;
  • the detection module is configured to input each of the image blocks to be detected into a preset road damage detection model for detection, and obtain detection information of each image block to be detected;
  • the merging module is used for merging the detection information of each image block to be detected to obtain road damage information of the image to be detected.
  • a computer device including a memory, a processor, and computer-readable instructions stored in the memory and capable of running on the processor, and the processor implements the above-mentioned road damage detection method when the processor executes the computer-readable instructions .
  • One or more non-volatile readable storage media storing computer readable instructions, when the computer readable instructions are executed by one or more processors, the one or more processors realize the above-mentioned road damage detection method.
  • FIG. 1 is a schematic diagram of an application environment of a road damage detection method in an embodiment of the present application
  • Fig. 2 is an example diagram of a road damage detection method in an embodiment of the present application
  • FIG. 3 is another example diagram of a road damage detection method in an embodiment of the present application.
  • Fig. 4 is another example diagram of a road damage detection method in an embodiment of the present application.
  • FIG. 5 is another example diagram of a road damage detection method in an embodiment of the present application.
  • FIG. 6 is another example diagram of a road damage detection method in an embodiment of the present application.
  • FIG. 7 is another example diagram of a road damage detection method in an embodiment of the present application.
  • FIG. 8 is another example diagram of a road damage detection method in an embodiment of the present application.
  • FIG. 9 is a principle block diagram of a road damage detection device in an embodiment of the present application.
  • Fig. 10 is a schematic diagram of a computer device in an embodiment of the present application.
  • the road damage detection method can be applied in the application environment as shown in FIG. 1.
  • the road damage detection method is applied in a road damage detection system.
  • the road damage detection system includes a client and a server as shown in Figure 1.
  • the client and the server communicate through the network to solve road damage detection. Inaccurate results.
  • the client is also called the client, which refers to the program that corresponds to the server and provides local services to the client.
  • the client can be installed on, but not limited to, various personal computers, laptops, smart phones, tablet computers, and portable wearable devices.
  • the server can be implemented with an independent server or a server cluster composed of multiple servers.
  • a road damage detection method is provided, and the method is applied to the server in FIG. 1 as an example for description, including the following steps:
  • S10 Obtain an image to be detected, and determine whether the image resolution of the image to be detected is greater than a resolution threshold.
  • the image to be detected refers to the original image to be detected.
  • the image to be detected may be an image extracted after video framing processing is performed on the video data in advance; it may also be an image pre-collected and saved by the client, or an image directly uploaded or sent to the client locally.
  • the method is applied to the user's mobile terminal, the mobile terminal is placed on an ordinary car driving on the road, and the video data of the actual conditions of the road ahead is collected in real time by using the video capture software of the mobile terminal , And then perform video framing processing on the collected video data, extract at least one image, and send it to the server as the image to be detected, and the server can obtain the image to be detected.
  • the resolution threshold refers to a preset value used to check whether the image resolution of the image to be detected meets the requirements.
  • the resolution threshold can be 600*600, 600*750, 750*600, etc., and the user can customize the setting according to the actual situation.
  • the resolution threshold is set to 600*600.
  • the resolution threshold includes a length threshold and a width threshold, and the image resolution of the image to be detected includes the length and width values of the image to be detected.
  • the length of the image to be detected must be compared with the corresponding length threshold to determine whether the length of the image to be detected is greater than the length threshold; and the width of the image to be detected The value is compared with the corresponding width threshold to determine whether the width of the image to be detected is greater than the width threshold.
  • the length value of the image to be detected is greater than the length threshold, or the width value of the image to be detected is greater than the width threshold, it means that the image resolution of the image to be detected is greater than the resolution threshold. Conversely, if the length value of the image to be detected is less than the length threshold and the width value is less than the width threshold, it means that the image resolution of the image to be detected is less than the resolution threshold.
  • the image to be detected is segmented to obtain N image blocks to be detected, and N is a positive integer.
  • the image block to be detected refers to an image obtained after segmentation processing of the image to be detected.
  • the segmentation processing of the image to be detected mainly includes image segmentation and image scaling of the image to be detected.
  • the image segmentation of the image to be detected refers to a process of segmenting the image to be detected with an image resolution greater than the resolution threshold into several sub-images according to a preset image segmentation method.
  • the image segmentation method may include: segmenting the image to be detected from the top left, bottom left, top right, bottom right, and center of the image to be detected to obtain 5 image blocks to be detected; or from the left, center, and right of the image to be detected
  • the position of the image to be inspected is segmented to obtain 3 image blocks to be inspected; or the image to be inspected is segmented from the upper, middle and lower positions of the image to be inspected to obtain 3 image blocks to be inspected.
  • the images to be detected with different image resolutions are segmented using different image segmentation methods.
  • the segmented image block to be detected is scaled to the image block to be detected with the same size as the image resolution and the resolution threshold.
  • the image block to be detected refers to an image after the image to be detected is segmented.
  • an image binarization processing method or an image processing tool may be used to implement segmentation processing of the image to be detected, and N image blocks to be detected are obtained, where N is a positive integer.
  • the image resolution of the image to be detected is 1200*1300 and the resolution threshold is 600*600, that is, the length and width of the image to be detected are greater than twice the length and width thresholds, respectively, according to the preset
  • the image segmentation method is performed from the upper left, lower left, upper right, lower right and center of the image to be detected, and the image resolutions are 600*650, 600*650, 600*650, 600*650, and 600*600, respectively
  • the image to be detected can be directly scaled to The image block to be detected whose image resolution is the same as the resolution threshold.
  • S30 Input each image block to be detected into a preset road damage detection model for detection, and obtain detection information of each image block to be detected.
  • the road damage detection model refers to a pre-trained model for detecting image blocks to be detected.
  • the road damage detection model can be obtained by training various network models.
  • the road damage detection model is obtained through convolutional neural network training, and the road damage detection model can determine the road damage category and corresponding location information corresponding to each input image block to be detected.
  • Convolutional Neural Network is a locally connected network. Compared with the fully connected network, its biggest feature is local connectivity and weight sharing. For a certain pixel p in an image, the closer the pixel p is, the greater the impact on it (local connectivity).
  • the weight of a certain area can also be used in another area, that is, the sharing of weights.
  • Weight sharing can be understood as convolution kernel sharing.
  • CNN convolutional neural network
  • a convolution kernel and a given image can be convolved to extract an image feature, and different convolution kernels can be extracted Different image characteristics. Due to the local connectivity of the convolutional neural network, the complexity of the model is reduced, and the efficiency of model training can be improved; and, due to the sharing of weights of the convolutional neural network, the convolutional neural network can learn in parallel to further improve the model training effectiveness.
  • the detection information of each image block to be detected can be obtained.
  • the detection information includes the road damage category and corresponding location information of each image block to be detected.
  • the road damage category refers to the category name obtained after the road damage category is classified according to the road damage situation.
  • road damage categories can be divided into eight types: straight cracks, straight cracks on new and old joints, straight cracks at equal intervals, straight cracks on new and old joints, tortoise cracks, grooves, blurred pedestrian crossings, and blurred white lines.
  • Location information refers to the specific location where road damage occurs in the image block to be detected.
  • the location information is represented by a rectangular area.
  • position information is expressed as (x min , y min , x max , y max ); x min represents the minimum value in the horizontal direction of the rectangular area, y min represents the minimum value in the vertical direction of the rectangular area, and x max represents The maximum value in the horizontal direction of the rectangular area, and y max represents the maximum value in the vertical direction of the rectangular area.
  • each image block to be detected is input into a preset road damage detection model for detection, if there is no road damage in the image to be detected, a default message can be output, or, The road damage information output for the image to be detected without road damage is empty or other default values.
  • S40 Combine the detection information of each image block to be detected to obtain road damage information of the image to be detected.
  • the road damage information refers to the information obtained by combining the detection information of each image block to be detected.
  • merging the detection information of each image block to be detected includes a process of merging image blocks to be detected of the same road damage category, and merging image blocks to be detected with associated location information.
  • the regular matching method or the string matching method may be used to match the road damage category of each image block to be detected, and then the successfully matched image blocks to be detected Determine the to-be-detected image blocks of the same road damage category, and determine the to-be-detected image blocks that fail to match as to-be-detected image blocks of different road damage categories.
  • the image blocks to be detected that are associated with position information are merged.
  • determining whether the position information of each image block to be detected is associated can be determined by pre-setting a position information association standard.
  • the location information association standard may be to first determine the area block composed of the location information of each image block to be detected, and then merge the image blocks to be detected with overlapping parts of the area block, or combine the area blocks.
  • the image blocks to be detected whose linear distance is less than the threshold distance are merged.
  • the threshold distance refers to the distance for judging whether the image blocks to be detected need to be merged.
  • the threshold distance can be 1cm, 3cm or 5cm. Users can customize the settings according to the actual situation.
  • the openCV method may be used to detect whether the position information of each image block to be detected is associated.
  • the image to be detected by acquiring the image to be detected, it is determined whether the image resolution of the image to be detected is greater than the resolution threshold; if the image resolution of the image to be detected is greater than the resolution threshold, the image to be detected is segmented to obtain N Image blocks to be detected; then input each image block to be detected into the preset road damage detection model for detection, and obtain the detection information of each image block to be detected; finally merge the detection information of each image block to be detected , To obtain road damage information of the image to be detected; not only solves the problem of inaccurate road damage detection results, but also improves the efficiency of road damage detection.
  • the image resolution of the image to be detected includes the length and width of the image to be detected, and the resolution threshold includes the length threshold and the width threshold. If the image resolution of the image to be detected is greater than the resolution Rate threshold, the segmentation processing of the image to be detected is performed to obtain N image blocks to be detected, which specifically includes the following steps:
  • the first reference rectangle refers to a preset reference area.
  • the first length value and the first width value of the first reference rectangle are determined according to the comparison relationship between the length value and the length threshold value of the image to be detected, and the width value and the width threshold value.
  • the length value of the image to be detected is greater than the length threshold and the width value is less than the width threshold, it is determined that the first length value of the first reference rectangle is equal to the length threshold, and the first width value is equal to the width value of the detected damage image; If the width value of the detected image is greater than the width threshold and the length value of the image to be detected is less than the length threshold, it is determined that the first width value of the first reference rectangle is equal to the width threshold, and the first length value is equal to the length value of the detected damage image.
  • S202 Determine three first vertices from the image to be detected, and determine three first target rectangles according to the three first vertices and the first length value and the first width value of the first reference rectangle.
  • the first vertex refers to a point determined on the image to be detected.
  • the image to be detected is an image whose length value is greater than the length threshold and the width value is less than the width threshold, or the length value is less than the length threshold and the width value is greater than the width threshold. Specifically, if the length value of the image to be detected is greater than the length threshold, and the width value of the image to be detected is less than the width threshold, or the width value of the image to be detected is greater than the width threshold and the length value of the image to be detected is less than the length threshold, then Determine the three first vertices in the image to be detected.
  • a coordinate system is established, with the lower left corner of the image to be detected as the origin, the resolution threshold is expressed as (M, N), and the length and width values of the image to be inspected are expressed as (x, y), if If the length value of the image to be detected is greater than the length threshold, and the width value of the image to be detected is less than the width threshold, the three first vertices determined from the image to be detected are set as vertices A 1 , A 2 and A 3 .
  • the coordinates of the vertex A 1 determined from the image to be detected are (M, y)
  • the coordinates of the vertex A 2 determined from the image to be detected are (xM, y)
  • the vertex determined from the image to be detected The coordinates of A 3 are ((xM)/2,y).
  • the binding 1 according to the above step of determining the first vertex A and the length of a first value and a width value and a left direction to be detected determines the first target rectangle image; determined according to the above steps with the first vertex A 2 a first length value and a width value binding, and the right direction to be detected determines the first target rectangle image; and to the above-described step of determining the image to be detected vertices a 3 in combination with a first length and a first width value according to the value of the right The direction determines the first target rectangle, thereby determining three first target rectangles on the image to be detected. Understandably, the length values of the three first target rectangles determined are the same as the length threshold value, and the width value is less than the width threshold value.
  • the three first vertices determined from the image to be detected are set as vertices B 1 , vertices B 2 and Vertex B 3 ; the coordinate of vertex B 1 determined from the image to be detected is (x, yN), and the coordinate of vertex B 2 determined from the image to be detected is (x, N), which is determined from the image to be detected The coordinates of vertex B 3 are ((x, (yM)/2). Further, the vertex B 1 determined according to the above steps is combined with the first length value and the first width value, and the first length value is determined toward the upper direction of the image to be detected.
  • the vertex B 2 determined according to the above steps is combined with the first length value and the first width value, and the first target rectangle is determined downward of the image to be detected; the vertex B 3 determined according to the above steps is combined with the first A length value and a first width value are combined and the first target rectangle is determined upwards of the image to be detected, thereby determining three first target rectangles on the image to be detected. Understandably, the determined three first targets
  • the width of the rectangle is the same as the width threshold, and the length is less than the length threshold.
  • S203 Segment the image to be detected according to the three first target rectangles to obtain three first rectangular image blocks.
  • the image to be detected may be segmented according to the three first target rectangles to obtain three first rectangular image blocks.
  • image binarization processing methods or image processing tools can be used to achieve segmentation of the image to be detected.
  • S204 Perform scaling processing on the three first rectangular image blocks to obtain three image blocks to be detected.
  • performing scaling processing on the three first target rectangles refers to a process of scaling the obtained three first rectangular image blocks into images with the same image resolution as the resolution threshold.
  • the image scaling algorithm can be used to implement the scaling processing on the three first rectangular image blocks; or the image scaling tool can be used to implement the scaling processing on the three first rectangular image blocks to obtain three image blocks to be detected.
  • the image scaling algorithm may be a bilinear interpolation algorithm or a trilinear convolution interpolation algorithm.
  • the image zoom tool can be photoshop, iResizer or FastStone Photo Resizer.
  • the first reference rectangle is determined A length value and a first width value; determine three first vertices from the image to be detected, and determine three first target rectangles according to the first length value and the first width value of each first vertex and the first reference rectangle; According to the three first target rectangles, the image to be detected is segmented to obtain three first rectangular image blocks, and the three first rectangular image blocks are scaled to obtain three image blocks to be detected; the image to be detected first Performing segmentation processing, and then zooming the cut image, while ensuring that all road damage features contained in the image to be detected can be fully obtained, it also improves the definition of the image block to be detected.
  • the image resolution of the image to be detected includes the length and width of the image to be detected, and the resolution threshold includes the length and width thresholds. If the image resolution of the image to be detected is greater than the resolution Rate threshold, the image to be detected is segmented to obtain N image blocks to be detected, which specifically includes the following steps:
  • S205 If the length value of the image to be detected is greater than the length threshold value, and the width value is greater than the width threshold value, determine the second length value and the second width value of the second reference rectangle.
  • the length value of the image to be detected is greater than the length threshold
  • the width value is greater than the width threshold includes: the length of the image to be detected is greater than the length threshold but less than twice the length threshold, and the width value is greater than the width threshold but less than twice the width threshold ; The length value is greater than the length threshold but less than twice the length threshold, and the width value is greater than twice the width threshold; the width value is greater than the width threshold but less than twice the width threshold, and the length value is greater than twice the length threshold; and the length value And the width values are both greater than the length threshold and the width threshold are more than twice the four cases.
  • the second reference rectangle refers to a preset reference area.
  • the second length value and the second width value of the second reference rectangle are determined according to the comparison relationship between the length value and the length threshold value of the image to be detected, and the width value and the width threshold value. Specifically, if the length value of the image to be detected is greater than the threshold length value but less than twice the length threshold, and the width value is greater than the width threshold but less than twice the width threshold, it is determined that the second length value of the second rectangle is equal to the length threshold, And the second width is equal to the width threshold.
  • the second width value of the second rectangle is equal to one-half of the width value of the image to be detected, and the second length value is equal to the value of the detected damage image. One half of the length value. If the length value of the image to be detected is greater than the length threshold but less than twice the length threshold, and the width value is greater than twice the width threshold, then it is determined that the second length value of the second rectangle is equal to the length threshold and the second width value is equal to the length to be detected One half of the width of the image.
  • the width value of the image to be detected is greater than the width threshold but less than twice the width threshold, and the length value is greater than twice the length threshold, it is determined that the second width value of the second rectangle is equal to the width threshold and the second length value is equal to the to-be-detected image One half of the length of the image.
  • S206 Determine five second vertices from the image to be detected, and determine five second target rectangles according to the five second vertices and the second length and second width values of the second reference rectangle.
  • the second vertex refers to a point determined on the image to be detected.
  • the image to be detected is an image whose length value is greater than the length threshold and width value is greater than the width threshold.
  • the five first vertices are determined from the image to be detected for the image to be detected whose length value is greater than the length threshold and the width value is greater than the width threshold.
  • a coordinate system is established, with the lower left corner of the image to be detected as the origin; the resolution threshold is expressed as (A, B), and the length and width values of the image to be inspected are expressed as (c, d); The length and width of the detected image are greater than the corresponding length threshold and width threshold, but less than twice the corresponding length and width threshold, then the five second vertices determined on the image to be detected are set as vertices C 1. Vertex C 2 , vertex C 3 , vertex C 4 and vertex C 5 .
  • the coordinates of the vertex C 1 determined from the image to be detected are (A, d), the coordinates of the vertex C 2 determined from the image to be detected are (cA, d), and the coordinates of the vertex C 3 determined from the image to be detected Is (A, 0), the coordinate of the vertex C 4 determined from the image to be detected is (cA, 0), and the coordinate of the vertex C 5 determined from the image to be detected is ((c–A)/2, d+ B/2).
  • the five second vertices determined from the image to be detected are set as vertices D 1 , vertices D 2 , and vertices D 3 , Vertex D 4 and vertex D 5
  • the coordinate of vertex D 1 determined from the image to be detected is (c/2, d)
  • the coordinate of vertex D 2 determined from the image to be detected is (c/2, d)
  • the coordinate of the vertex D 3 determined from the image to be detected is (c/2, 0)
  • the coordinate of the vertex D 4 determined from the image to be detected is (c/2, 0), which is determined from the image to be detected
  • the coordinates of the vertex D 5 of is ((c–A)/2, d+B/2).
  • the second vertex coordinates determined for the image to be detected in the other two cases are the same as the method for determining the second vertex coordinates according to the relationship between the length and width of the image
  • the vertex C 1 /D 1 determined according to the above steps is combined with the second length value and the second width value, and the second target rectangle is determined toward the upper left direction of the image to be detected;
  • the vertex C 2 determined according to the above steps /D 2 is combined with the second length value and the second width value, and the second target rectangle is determined in the upper right direction of the image to be detected;
  • combine the vertex C 4 /D 4 determined according to the above steps with the second length value and the second width value, and to the lower right direction of the image to be detected Determine the second target rectangle;
  • the determined length value of the second target rectangle is equal to the length threshold
  • the width value is equal to the width threshold. If the length and width values of the image to be detected are both greater than twice the length threshold and width threshold, the determined length value of the second target rectangle is one-half of the length of the image to be detected, and the width value is equal to the image to be detected One-half of the width value.
  • S207 Segment the image to be detected according to the five second target rectangles to obtain five second rectangular image blocks.
  • the image to be detected may be segmented according to the five second target rectangles to obtain five second rectangular image blocks.
  • image binarization processing methods or image processing tools can be used to achieve segmentation of the image to be detected.
  • S208 Perform scaling processing on the five second rectangular image blocks to obtain five image blocks to be detected.
  • the five second rectangular image blocks determined according to step S205 are scaled to obtain N image blocks to be detected. Understandably, in this step, N of the N image blocks to be detected is 5. Specifically, the zooming process and method of the five second rectangular image blocks are the same as the specific process and method of zooming the three first target rectangles in step S203, and will not be redundantly described here.
  • the second length value and the second width value of the second reference rectangle are determined; the five first values are determined from the image to be detected Two vertices, and determine five second target rectangles according to the five second vertices and the second length value and second width value of the second reference rectangle, and divide the image to be detected according to the five second target rectangles to obtain Five second rectangular image blocks; the five second rectangular image blocks are scaled to obtain N image blocks to be inspected; the image to be inspected is segmented first, and then the cut image is scaled to ensure completeness While acquiring all the road damage features contained in the image to be detected, the clarity of the acquired image block to be detected is also improved.
  • the detection information includes road damage category and location information. Combining the detection information of each image block to be detected to obtain road damage information of the image to be detected includes the following steps:
  • S401 Classify the detection information according to the road damage category to obtain M detection information classification sets, where M is a positive integer.
  • the detection information classification set refers to the classification set obtained by classifying the detection information according to the road damage category.
  • classifying the detection information according to the road damage category specifically includes: adopting a regular matching method to respectively match the road damage categories in each detection information one by one. If the matching is successful, it indicates the road damage in the corresponding detection information. If the categories are the same, if the matching fails, it means that the road damage categories in the corresponding detection information are different; then the detection information corresponding to the same road damage category that is successfully matched is classified into the same category, and the different road damage categories that fail to match The detection information is classified separately, and finally M detection information classification sets are obtained. Understandably, if the road damage categories in the acquired detection information are all the same, then the detection information classification set obtained after classification is one. If there are different road damage categories in the acquired detection information, the detection obtained after classification is performed The information classification set is at least two.
  • each detection information classification set In each detection information classification set, combine the detection information based on the location information to obtain a detection information combined set.
  • merging detection information based on location information refers to a process of merging image blocks to be detected with associated location information.
  • the position information of the image block to be detected is represented by a specific coordinate value
  • determining whether the position information of each image block to be detected is associated can be determined by pre-setting a position information association standard.
  • the location information association standard may be to first determine the area block composed of the location information of each image block to be detected, and then merge the image blocks to be detected with overlapping parts of the area block, and combine the area blocks. The image blocks to be detected whose linear distance is less than the threshold distance are merged.
  • a detection information classification set with a road damage category of DO contains two different location information, namely A 1 (45,77,65,90) and A 2 (55,90,80,100)
  • the detection can be If the area blocks composed of positions A 1 (45,77,65,90) and A 2 (55,90,80,100) overlap each other, then the detection information corresponding to A 1 and A 2 is combined to obtain a detection information combination set.
  • S403 Combine each detection information to form road damage information of the image to be detected.
  • the target detection set refers to the detection information formed by the combination of each detection information combined set. Specifically, all the detection information combined sets obtained according to step S402 are combined to obtain the road damage information of the image block to be detected.
  • the detection information can be combined according to the location information first; then in each detection information location set, each detection information location set is classified according to the road damage category, and the detection information of the same road damage category is grouped into one Class, separate the detection information of different road damage categories to obtain a combined set of detection information; finally, each combined set of detection information is composed of road damage information of the image block to be detected.
  • the detection information is classified according to the road damage category to obtain M detection information classification sets; in each detection information classification set, the detection information with adjacent or overlapping location information is merged, Obtain a combined set of detection information; each combined set of detection information forms the road damage information of the image to be detected, which further ensures the integrity of the road damage detection result.
  • the road damage detection method before each image block to be detected is input into a preset road damage detection model for detection, and the detection information of each image block to be detected is obtained, the road damage detection method also include:
  • the original sample image refers to a sample image containing road damage features
  • L is a positive integer.
  • Each original sample image has been sample labeled means that each original sample image has been labeled with road damage category and location information in advance.
  • the road damage category is divided into eight grades, namely: straight cracks in the driving part of the vehicle, straight cracks in the new and old joints, and equal spacing Straight cracks, straight cracks in new and old joints, tortoise cracks, grooves, blurred pedestrian crossings and blurred white lines; and the position information is expressed through a rectangular area. Specifically, it can be represented by the coordinates of four points on the rectangular area. For example: the position information is expressed as (x min , y min , x max , y max ).
  • the grade identifier refers to an identifier for distinguishing different damage grades corresponding to different road damage categories, and can be represented by any combination of Arabic numerals, uppercase letters or lowercase letters.
  • the grade identifier of the linear crack in the driving part of the vehicle is DO
  • the grade identifier of the linear crack in the old and new joints is D1
  • the grade identifier of the equidistant linear crack is D2
  • the grade identifier of the linear crack in the new and old joints is D3.
  • the grade identifier of the crack is D4
  • the identifier of the groove grade is D5
  • the grade identifier of the crosswalk is D6
  • the grade identifier of the white line is D7.
  • the target sample image refers to the image obtained by preprocessing the original sample image.
  • performing sample preprocessing on the original sample image includes performing image cropping and image scaling on the original sample image.
  • the image processing algorithm of opencv can be used to implement image cropping and image scaling of the original sample image to obtain the target sample image.
  • opencv is a cross-platform computer vision library released under the BSD license, consisting of a series of C functions and a small number of C++ classes; opencv also provides interfaces for languages such as Python, Ruby, MATLAB, etc., which can realize image processing and computer vision Many general algorithms.
  • the dimension of the input vector of the fully connected layer in the road damage detection model is a fixed size, in order to avoid dynamic changes of the network, the purpose of parameter training cannot be achieved, and each original sample image is sampled
  • the size of the target sample image obtained after preprocessing is also a fixed size.
  • S43 Input each target sample image into the convolutional neural network for training to obtain a road damage detection model.
  • the road damage detection model refers to a model that is generated after training through a large number of target sample images and can be used to detect images to be detected. Specifically, each target sample image that has been labeled with road damage category and location information is input into the convolutional neural network for end-to-end training. When the training reaches a certain number of iterations, the training stops, and the final road damage is obtained. Detection model.
  • the road damage detection model can be trained using the TensorFlow calculation framework.
  • the bazel tool can also be used to compile the model trained using the tensorflow framework into a .so file and jar package, and then proceed Android configuration, to realize the porting of road damage detection model from PC to Android.
  • the Bazel tool is an open source build and test tool. It uses a high-level build language that is easy for people to understand and supports multiple development languages. It can be built on multiple platforms.
  • each original sample image has been sample-labeled, and L is a positive integer; then sample preprocessing is performed on each original sample image to obtain a target sample image; Finally, each target sample image is input into the convolutional neural network for training to obtain a road damage detection model; the accuracy of the obtained road damage detection model is guaranteed.
  • sample preprocessing is performed on each original sample image to obtain the target sample image, which specifically includes the following steps:
  • S421 Obtain original sample images, perform detection on each original sample image, and obtain detection information of each original sample image.
  • the detection information refers to marking information that divides the road damage image part and the background image part in each original sample image.
  • an edge detection algorithm can be used to detect each original sample image, and obtain the detection information of each original sample image.
  • the edge detection algorithm can be any one of opencv's Canny edge detection algorithm, Sobel algorithm or Laplacian algorithm.
  • S422 According to the detection information, perform cropping processing on each original sample image to obtain a training sample image.
  • the training sample image refers to the sample image obtained after image cropping processing.
  • Cutting the original sample image refers to the process of cutting out the background image part from the original sample image according to the detection information in each original sample image.
  • an image cropping tool is used to implement cropping processing on each original sample image.
  • the image cropping tool can be jQuery Jcrop image cropping tool or FOTOE image cropping tool, etc.
  • the image segmentation algorithm of opencv can also be used to automatically implement the cropping process of each original sample image to obtain the training sample image.
  • S423 Obtain a preset image ratio, and perform scaling processing on each training sample image according to the preset image ratio to obtain a target sample image.
  • the preset image ratio refers to a preset standard image size.
  • the preset image ratio may be 600*600, 750*750, 800*800, or the like.
  • the preset image ratio needs to be the same as the resolution threshold in step S10.
  • the preset image ratio is set to 600*600.
  • the target sample image refers to the sample image that can be directly trained after scaling.
  • performing scaling processing on each training sample image according to a preset image ratio refers to a process of scaling each training sample image into a target sample image with a preset image ratio size.
  • an image scaling processing algorithm can be used to implement scaling processing on each training sample image to obtain a target sample image.
  • the image scaling processing algorithm can be a neighboring interpolation algorithm, a bilinear interpolation algorithm, or a cubic convolution method.
  • the image scaling API tool of opencv can also be used to automatically perform scaling processing on the cropped training sample image to obtain the target sample image.
  • each original sample image is detected to obtain the detection information of each original sample image; then, according to the detection information, each original sample image is cropped to obtain the training sample image ; Finally, the preset image ratio is obtained, and each training sample image is scaled according to the preset image ratio to obtain the target sample image, which further ensures the accuracy and validity of the obtained target sample image.
  • inputting each target sample image into a convolutional neural network for training to obtain a road damage detection model specifically includes the following steps:
  • initializing the convolutional neural network refers to pre-initializing the model parameters of the convolutional neural network model (that is, the convolution kernel and the bias).
  • the convolution kernel refers to the weight of the convolutional neural network. When the training data is input, it will be multiplied by a weight, the convolution kernel, and then get the output of the neuron, which reflects the importance of the training data.
  • Bias is the linear component used to change the range of the weight multiplied by the input.
  • S442 Use a convolutional neural network to perform feature extraction on each target sample image to obtain road damage features.
  • the road damage feature is the damage feature obtained by using the convolutional neural network to extract the feature of the target sample image in the training set.
  • a convolutional neural network is used to perform feature extraction on the target sample image.
  • the calculation formula for convolution operation includes * Represents the convolution operation; x j represents the j-th input feature map; y j represents the j-th output feature map; w ij is the convolution kernel between the i-th input feature map and the j-th output feature map (weight Value); b j represents the bias term of the j-th output feature map.
  • the maximum pooling down-sampling is used to down-sampling the convolved feature map to achieve dimensionality reduction of the feature map.
  • y j represents the i-th output spectrum in the down-sampling process (ie the feature map after down-sampling), and each neuron in the down-sampling process is from the i-th input spectrum (the feature map after convolution)
  • the S*S down-sampling frame is locally sampled, that is, the input spectrum is down-sampled by S times, where the specific value of S can be set according to actual sampling needs; m and n respectively represent the step size of the down-sampling frame.
  • the road damage features obtained according to step S442 are input into the convolutional neural network model for training, and the road damage detection model can be obtained.
  • the model parameters of the convolutional neural network model are initialized; then the convolutional neural network is used to extract features of each target sample image to obtain the road damage feature; finally the road damage feature is input to the convolutional neural network model
  • the road damage detection model is obtained by training in the middle, which further improves the accuracy of the road damage detection model.
  • a road damage detection device is provided, and the road damage detection device corresponds to the road damage detection method in the above-mentioned embodiment one-to-one.
  • the road damage detection device includes a resolution judgment module 10, a segmentation processing module 20, a detection module 30 and a merging module 40.
  • the detailed description of each functional module is as follows:
  • the resolution judgment module 10 is used to obtain the image to be detected and determine whether the image resolution of the image to be detected is greater than the resolution threshold;
  • the segmentation processing module 20 is configured to perform segmentation processing on the image to be detected when the image resolution of the image to be detected is greater than the resolution threshold to obtain N image blocks to be detected, where N is a positive integer;
  • the detection module 30 is configured to input each image block to be detected into a preset road damage detection model for detection, and obtain detection information of each image block to be detected;
  • the merging module 40 is used to merge the detection information of each image block to be detected to obtain road damage information of the image to be detected.
  • the segmentation processing module 20 includes:
  • the first determining unit is configured to determine the first reference rectangle when the length value of the image to be detected is greater than the length threshold and the width value is less than the width threshold, or the length value of the image to be detected is less than the length threshold and the width value is greater than the width threshold.
  • the second determining unit is configured to determine three first vertices from the image to be detected, and determine three first target rectangles according to the three first vertices and the first length value and the first width value of the first reference rectangle;
  • the first segmentation unit is configured to segment the image to be detected according to the three first target rectangles to obtain three first rectangular image blocks;
  • the first zoom processing unit is configured to perform zoom processing on the three first rectangular image blocks to obtain three to-be-detected image blocks.
  • the segmentation processing module 20 further includes:
  • the third determining unit is configured to determine the second length value and the second width value of the second reference rectangle when the length value of the image to be detected is greater than the length threshold and the width value is greater than the width threshold;
  • a fourth determining unit configured to determine five second vertices from the image to be detected, and determine five second target rectangles according to the five second vertices and the second length value and the second width value of the second reference rectangle;
  • the second segmentation unit is configured to segment the image to be detected according to the five second target rectangles to obtain five second rectangular image blocks;
  • the second scaling processing unit is configured to perform scaling processing on the five second rectangular image blocks to obtain five image blocks to be detected.
  • the merging module 40 includes:
  • the classification unit is used to classify the detection information according to the road damage category to obtain M detection information classification sets, where M is a positive integer;
  • the merging unit is used to merge the detection information based on the location information in each detection information classification set to obtain a detection information combined set;
  • the composition unit is used to combine each detection information to form road damage information of the image to be detected.
  • the road damage detection device further includes:
  • the original sample image acquisition module is used to obtain L original sample images, where each original sample image has been sample-labeled, and L is a positive integer;
  • the preprocessing module is used to perform sample preprocessing on each original sample image to obtain the target sample image
  • the training module is used to input each target sample image into the convolutional neural network for training to obtain a road damage detection model.
  • the preprocessing module includes:
  • the detection unit is used to obtain original sample images, detect each original sample image, and obtain detection information of each original sample image
  • the cropping processing unit is used to perform cropping processing on each original sample image according to the detection information to obtain a training sample image
  • the third scaling processing unit is configured to obtain a preset image scale, and perform scaling processing on each training sample image according to the preset image scale to obtain a target sample image.
  • the training module includes:
  • the initialization unit is used to initialize the model parameters of the convolutional neural network model
  • the feature extraction unit is used to extract features of each target sample image by using a convolutional neural network to obtain road damage features;
  • the training unit is used to input the road damage feature into the convolutional neural network model for training to obtain a road damage detection model.
  • Each module in the above road damage detection device can be implemented in whole or in part by software, hardware and a combination thereof.
  • the foregoing modules may be embedded in the form of hardware or independent of the processor in the computer device, or may be stored in the memory of the computer device in the form of software, so that the processor can call and execute the operations corresponding to the foregoing modules.
  • a computer device is provided.
  • the computer device may be a server, and its internal structure diagram may be as shown in FIG. 10.
  • the computer equipment includes a processor, a memory, a network interface and a database connected through a system bus.
  • the processor of the computer device is used to provide calculation and control capabilities.
  • the memory of the computer device includes a non-volatile storage medium and an internal memory.
  • the non-volatile storage medium stores an operating system, computer readable instructions, and a database.
  • the internal memory provides an environment for the operation of the operating system and computer-readable instructions in the non-volatile storage medium.
  • the database of the computer device is used to store the data used in the road damage detection method in the foregoing embodiment.
  • the network interface of the computer device is used to communicate with an external terminal through a network connection.
  • the computer readable instruction is executed by the processor to realize a road damage detection method.
  • a computer device including a memory, a processor, and computer-readable instructions stored in the memory and capable of running on the processor, and the processor implements the following steps when the processor executes the computer-readable instructions:
  • one or more non-volatile readable storage media storing computer readable instructions, when the computer readable instructions are executed by one or more processors, cause the one or more processing
  • the device performs the following steps:
  • Non-volatile memory may include read only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM), or flash memory.
  • ROM read only memory
  • PROM programmable ROM
  • EPROM electrically programmable ROM
  • EEPROM electrically erasable programmable ROM
  • Volatile memory may include random access memory (RAM) or external cache memory.
  • RAM is available in many forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), double data rate SDRAM (DDRSDRAM), enhanced SDRAM (ESDRAM), synchronous chain Channel (Synchlink) DRAM (SLDRAM), memory bus (Rambus) direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM (RDRAM), etc.

Landscapes

  • Engineering & Computer Science (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Physics & Mathematics (AREA)
  • General Physics & Mathematics (AREA)
  • Theoretical Computer Science (AREA)
  • Quality & Reliability (AREA)
  • Image Analysis (AREA)

Abstract

本申请公开了一种道路损伤检测方法、装置、计算机设备及存储介质；通过获取待检测图像，判断待检测图像的图像分辨率是否大于分辨率阈值；若待检测图像的图像分辨率大于分辨率阈值，则对待检测图像进行分割处理，得到N个待检测图像块；然后将每一待检测图像块输入到预设的道路损伤检测模型中进行检测，得到每一待检测图像块的检测信息；最后将每一待检测图像块的检测信息进行合并，得到待检测图像的道路损伤信息；不但解决了道路损伤检测结果不准确的问题，还进一步提高了道路损伤检测的效率。

Description

道路损伤检测方法、装置、计算机设备及存储介质
本申请以2019年6月14日提交的申请号为201910517637.0，名称为“道路损伤检测方法、装置、计算机设备及存储介质”的中国发明专利申请为基础，并要求其优先权。
技术领域
本申请涉及智能决策领域，尤其涉及一种道路损伤检测方法、装置、计算机设备及存储介质。
背景技术
随着高速公路通车里程的快速增长和道路服务年限的增加，高速公路逐渐进入建设与养护并重的时期。众所周知，车辆行驶道路路面的状况，会直接影响到道路的车辆行驶服务质量，我国经过多年来的快速基建建设，很多道路都开始出现各种各样的道路损伤问题。目前，为了能够及时发现年久或受损的路面，以确保道路能够保持良好的使用性能，需对道路进行大量的定期检测和养护，这对相关人力和财政费用都提出了巨大要求。传统的道路损伤检测方法大部分都是采用人工检测或使用特定的路面检测车进行检测，因此，使用传统的道路损伤检测方法对道路进行检测经常会出现检测结果不准确等现象。
申请内容
本申请实施例提供一种道路损伤检测方法、装置、计算机设备及存储介质，以解决道路损伤检测结果不准确的问题。
一种道路损伤检测方法，包括：
获取待检测图像，判断所述待检测图像的图像分辨率是否大于分辨率阈值；
若所述待检测图像的图像分辨率大于所述分辨率阈值，则对所述待检测图像进行分割处理，得到N个待检测图像块，N为正整数；
将每一所述待检测图像块输入到预设的道路损伤检测模型中进行检测，得到每一所述待检测图像块的检测信息；
将每一所述待检测图像块的所述检测信息进行合并，得到所述待检测图像的道路损伤信息。
一种道路损伤检测装置，包括：
分辨率判断模块，用于获取待检测图像，判断所述待检测图像的图像分辨率是否大于分辨率阈值；
分割处理模块，用于在所述待检测图像的图像分辨率大于所述分辨率阈值时，对所述待检测图像进行分割处理，得到N个待检测图像块，N为正整数；
检测模块，用于将每一所述待检测图像块输入到预设的道路损伤检测模型中进行检测，得到每一所述待检测图像块的检测信息；
合并模块，用于将每一所述待检测图像块的所述检测信息进行合并，得到所述待检测图像的道路损伤信息。
一种计算机设备，包括存储器、处理器以及存储在所述存储器中并可在所述处理器上运行的计算机可读指令，所述处理器执行所述计算机可读指令时实现上述道路损伤检测方法。
一个或多个存储有计算机可读指令的非易失性可读存储介质，所述计算机可读指令被一个或多个处理器执行时，使得所述一个或多个处理器实现上述道路损伤检测方法。
本申请的一个或多个实施例的细节在下面的附图和描述中提出，本申请的其他特征和优点将从说明书、附图以及权利要求变得明显。
附图说明
为了更清楚地说明本申请实施例的技术方案，下面将对本申请实施例的描述中所需要使用的附图作简单地介绍，显而易见地，下面描述中的附图仅仅是本申请的一些实施例，对于本领域普通技术人员来讲，在不付出创造性劳动性的前提下，还可以根据这些附图获得其他的附图。
图1是本申请一实施例中道路损伤检测方法的一应用环境示意图；
图2是本申请一实施例中道路损伤检测方法的一示例图；
图3是本申请一实施例中道路损伤检测方法的另一示例图；
图4是本申请一实施例中道路损伤检测方法的另一示例图；
图5是本申请一实施例中道路损伤检测方法的另一示例图；
图6是本申请一实施例中道路损伤检测方法的另一示例图；
图7是本申请一实施例中道路损伤检测方法的另一示例图；
图8是本申请一实施例中道路损伤检测方法的另一示例图；
图9是本申请一实施例中道路损伤检测装置的一原理框图；
图10是本申请一实施例中计算机设备的一示意图。
具体实施方式
下面将结合本申请实施例中的附图，对本申请实施例中的技术方案进行清楚、完整地描述，显然，所描述的实施例是本申请一部分实施例，而不是全部的实施例。基于本申请中的实施例，本领域普通技术人员在没有作出创造性劳动前提下所获得的所有其他实施例，都属于本申请保护的范围。
本申请实施例提供的道路损伤检测方法，该道路损伤检测方法可应用如图1所示的应用环境中。具体地，该道路损伤检测方法应用在道路损伤检测系统中，该道路损伤检测系统包括如图1所示的客户端和服务端，客户端与服务端通过网络进行通信，用于解决道路损伤检测结果不准确的问题。其中，客户端又称为用户端，是指与服务端相对应,为客户提供本地服务的程序。客户端可安装在但不限于各种个人计算机、笔记本电脑、智能手机、平板电脑和便携式可穿戴设备上。服务端可以用独立的服务器或者是多个服务器组成的服务器集群来实现。
在一实施例中，如图2所示，提供一种道路损伤检测方法，以该方法应用在图1中的服务端为例进行说明，包括如下步骤：
S10：获取待检测图像，判断待检测图像的图像分辨率是否大于分辨率阈值。
其中，待检测图像指待进行检测的原始图像。可选地，待检测图像可以是预先对视频数据进行视频分帧处理后所提取出来的图像；也可以是客户端预先采集并保存的图像，或者是本地直接上传或者发送到客户端的图像。例如：在一个应用场景中，该方法应用在用户的移动终端中，该移动终端放置于行驶在道路上的普通小汽车上，通过使用移动终端的视频采集软件实时采集前方道路实际状况的视频数据，然后对采集的视频数据进行视频分帧处理，提取出至少一幅图像，作为待检测图像发送到服务端，服务端即可获取到该待检测图像。
其中，分辨率阈值指预先设定的、用来检验待检测图像的图像分辨率是否符合要求的数值。可选地，分辨率阈值可以为600*600、600*750或750*600等，用户可根据实际情况自定义设置。优选地，为了保证所得图像的清晰度，在本实施例中，将分辨率阈值设置为600*600。具体地，分辨率阈值包括长度阈值和宽度阈值，待检测图像的图像分辨率包括待检测图像的长度值和宽度值。判断待检测图像的图像分辨率是否大于分辨率阈值，需分别将待检测图像的长度值与对应的长度阈值进行对比，判断待检测图像的长度值是否大于长度阈值；和将待检测图像的宽度值与对应的宽度阈值进行对比，判断待检测图像的宽度值是否大于宽度阈值。
在本实施例中，若待检测图像的长度值大于长度阈值，或待检测图像的宽度值大于宽度阈值，则表示该待检测图像的图像分辨率大于分辨率阈值。反之，若待检测图像的长度值小于长度阈值且宽度值小于宽度阈值，则表示该待检测图像的图像分辨率小于分辨率阈值。
S20：若待检测图像的图像分辨率大于分辨率阈值，则对待检测图像进行分割处理，得到N个待检测图像块，N为正整数。
若根据步骤S10判断得到待检测图像的图像分辨率大于分辨率阈值，则对该待检测图像进行分割处理，得到N个待检测图像块，N为正整数。其中，待检测图像块指对待检测图像进行分割处理后所得的图像。
在本实施例中，对待检测图像进行分割处理主要包括对待检测图像的图像分割和图像缩放。具体地，对待检测图像的图像分割指根据预设的图像分割方式，将图像分辨率大于分辨率阈值的待检测图像分割成若干个子图像的过程。具体地，图像分割方式可以包括：从待检测图像的左上、左下、右上、右下和正中位置对待检测图像进行分割，得到5个待检测图像块；或者从待检测图像的左边、正中和右边位置对待检测图像进行分割，得到3个待检测图像块；或者从待检测图像的上边、正中和下边位置对待检测图像进行分割，得到3个待检测图像块。在一具体实施例中，对不同图像分辨率的待检测图像，采用不同的图像分割方式进行分割。
然后，再对分割后的待检测图像块缩放成图像分辨率与分辨率阈值相同大小的待检测图像块。其中，待检测图像块指对待检测图像进行分割处理后的图像。可选地，可采用图像二值化处理方法或采用图像处理工具实现对待检测图像的分割处理，得到N个待检测图像块，N为正整数。
示例性地，若待检测图像的图像分辨率为1200*1300，分辨率阈值为600*600，即待检测图像的长度值和宽度值分别大于长度阈值和宽度阈值的两倍，则根据预设的图像分割方式，从该待检测图像的左上、左下、右上、右下和正中进行图像分割，得到图像分辨率分别为600*650、600*650、600*650、600*650和600*600的五个初始待检测图像块；然后再对这五个初始待检测图像块进行图像缩放，得到五个图像分辨率都为600*600的待检测图像块。
在一具体实施例中，若待检测图像的图像分辨率小于分辨率阈值，即待检测图像的长度值和宽度值都小于对应的长度阈值和宽度阈值，则可直接将该待检测图像缩放成图像分辨率与分辨率阈值相同的待检测图像块。
S30：将每一待检测图像块输入到预设的道路损伤检测模型中进行检测，得到每一待检测图像块的检测信息。
其中，道路损伤检测模型是指预先训练的用于对待检测图像块进行检测的模型。道路损伤检测模型可以通过对个各种网络模型进行训练得到。可选地，道路损伤检测模型为采用卷积神经网络训练得到的，道路损伤检测模型可以判断输入的每一待检测图像块对应的道路损伤类别和对应的位置信息。其中，卷积神经网络(Convolutional Neural Network，CNN))是局部连接网络。相对于全连接网络其最大的特点就是局部连接性和权值共享性。对于一副图像中的某个像素p来说，离像素p越近的像素对其影响也就越大(局部连接性)。另外，根据自然图像的统计特性，某个区域的权值也可以用于另一个区域，即权值共享性。权值共享可以理解为卷积核共享，在卷积神经网络(CNN)中，将一个卷积核与给定的图像做卷积运算就可以提取一种图像特征，不同的卷积核可以提取不同的图像特征。由于卷积神经网络的局部连接性，使得模型的复杂度降低，可以提高模型训练的效率；并且，由于卷积神经网络的权值共享性，因此卷积神经网络可以并行学习，进一步提高模型训练效率。
具体地，将每一待检测图像块输入到预设的道路损伤检测模型中进行检测，即可得到每一待检测图像块的检测信息。在本实施例中，检测信息包括每一待检测图像块的道路损伤类别和对应的位置信息。其中，道路损伤类别指根据道路损伤情况进行道路损伤类别分类后所得的类别名称。例如：道路损伤类别可以分为直线裂纹、新旧接缝直线裂纹、等间距直线裂纹、新旧接缝直线裂纹、龟裂纹、凹槽、人行横道模糊和白线模糊八种类型。位置信息指待检测图像块中出现道路损伤情况的具体位置。可选地，位置信息以通过一矩形区域表示。具体地，可以通过矩形区域上的四个点的坐标值表示。例如：位置信息表示为(x min，y min，x max，y max)；x min表示在矩形区域横方向上的最小值，y min表示在矩形区域竖方向上的最小值，x max表示在矩形区域横方向上的最大值，y max表示在矩形区域竖方向上的最大值。
在一具体实施例中，若将每一待检测图像块输入到预设的道路损伤检测模型中进行检测后，若待检测图像中没有存在道路损伤情况，则可以输出一个默认的信息，或者，对不存在道路损伤情况的待检测图像输出的道路损伤信息为空或者其他默认值。
S40：将每一待检测图像块的检测信息进行合并，得到待检测图像的道路损伤信息。
其中，道路损伤信息指将每一待检测图像块的检测信息进行合并后所得到的信息。在本实施例中，将每一待检测图像块的检测信息进行合并包括对相同道路损伤类别的待检测图像块进行合并，和将位置信息存在关联的待检测图像块进行合并的过程。
具体地，将相同道路损伤类别的待检测图像块进行合并可以先采用正则匹配法或字符串匹配法，将每一待检测图像块的道路损伤类别进行匹配，然后将匹配成功的待检测图像块确定为相同道路损伤类别的待检测图像块，将匹配失败的待检测图像块确定为不同道路损伤类别的待检测图像块。
进一步地，将位置信息存在关联的待检测图像块进行合并。可选地，确定每一待检测图像块的位置信息是否存在关联可通过预先设定一个位置信息关联标准来判断。具体地，位置信息关联标准可以为先确定由每一待检测图像块的位置信息所组成的区域块，然后，再将区域块存在重合部分的待检测图像块进行合并，或者将区域块之间的直线距离小于阈值距离的待检测图像块进行合并。其中，阈值距离指判断待检测图像块之间是否需要合并的距离。阈值距离可以为1cm,3cm或5cm等。用户可根据实际情况自定义设置。可选地，可采用openCV方法检测每一待检测图像块的位置信息是否存在关联。
在本实施例中，通过获取待检测图像，判断待检测图像的图像分辨率是否大于分辨率阈值；若待检测图像的图像分辨率大于分辨率阈值，则对待检测图像进行分割处理，得到N个待检测图像块；然后将每一待检测图像块输入到预设的道路损伤检测模型中进行检测，得到每一待检测图像块的检测信息；最后将每一待检测图像块的检测信息进行合并，得到待检测图像的道路损伤信息；不但解决了道路损伤检测结果不准确的问题，还提高了道路损伤检测的效率。
在一实施例中，如图3所示，待检测图像的图像分辨率包括待检测图像的长度值和宽度值，分辨率阈值包括长度阈值和宽度阈值，若待检测图像的图像分辨率大于分辨率阈值，则对待检测图像进行分割处理，得到N个待检测图像块，具体包括如下步骤：
S201：若待检测图像的长度值大于长度阈值且宽度值小于宽度阈值，或者，若待检测图像的长度值小于长度阈值且宽度值大于宽度阈值，则确定第一参考矩形的第一长度值和第一宽度值。
其中，第一参考矩形是指预先设定的一个参考区域。在本实施例中，第一参考矩形的第一长度值和第一宽度值是根据待检测图像的长度值与长度阈值，以及宽度值与宽度阈值的对比关系所确定的。具体地，若待检测图像的长度值大于长度阈值、且宽度值小于宽度阈值，则确定第一参考矩形的第一长度值等于长度阈值，第一宽度值等于检测损伤图像的宽度值；若待检测图像的宽度值大于宽度阈值、且待检测图像的长度值小于长度阈值，则确定第一参考矩形的第一宽度值等于宽度阈值，第一长度值等于检测损伤图像的长度值。
S202：从待检测图像中确定三个第一顶点，并根据三个第一顶点和第一参考矩形的第一长度值和第一宽度值确定三个第一目标矩形。
其中，第一顶点指在待检测图像上所确定的点。该待检测图像为长度值大于长度阈值且宽度值小于宽度阈值，或者长度值小于长度阈值且宽度值大于宽度阈值的图像。具体地，若待检测图像的长度值大于长度阈值、且待检测图像的宽度值小于宽度阈值，或者，待检测图像的宽度值大于宽度阈值、且待检测图像的长度值小于长度阈值，则从待检测图像中的确定三个第一顶点。
示例性地，建立一坐标系，以待检测图像的左下角为原点，将分辨率阈值表示为(M,N)，待检测图像的的长度值和宽度值表示为(x，y)，若待检测图像的长度值大于长度阈值、且待检测图像的宽度值小于宽度阈值，则将从待检测图像上确定的三个第一顶点设定为顶点A 1、顶点A 2和顶点A 3。具体地，从待检测图像上所确定的顶点A 1坐标为(M,y),从待检测图像上所确定的顶点A 2坐标为(x-M,y),从待检测图像上所确定的顶点A 3坐标为((x-M)/2,y)。进一步地，将根据上述步骤确定的顶点A 1与第一长度值和第一宽度值结合、且往待检测图像的左方向确定第一目标矩形；将根据上述步骤确定的顶点A 2与第一长度值和第一宽度值结合、且往待检测图像的右方向确定第一目标矩形；将根据上述步骤确定的顶点A 3与第一长度值和第一宽度值结合且往待检测图像的右方向确定第一目标矩形，从而在待检测图像上确定三个第一目标矩形。可以理解地，该所确定的三个第一目标矩形的长度值与长度阈值相同，宽度值小于宽度阈值
同样地，若待检测图像的宽度值大于宽度阈值、且待检测图像的长度值小于长度阈值，则将从待检测图像上确定的三个第一顶点设定为顶点B 1、顶点B 2和顶点B 3；从待检测图像上所确定的顶点B 1坐标为(x,y-N),从待检测图像上所确定的顶点B 2坐标为(x,N),从待检测图像上所确定的顶点B 3坐标为((x,(y-M)/2)。进一步地，将根据上述步骤确定的顶点B 1与第一长度值和第一宽度值结合、且往 待检测图像的上方向确定第一目标矩形；将根据上述步骤确定的顶点B 2与第一长度值和第一宽度值结合、且往待检测图像的下方向确定第一目标矩形；将根据上述步骤确定的顶点B 3与第一长度值和第一宽度值结合且往待检测图像的上方向确定第一目标矩形，从而在待检测图像上确定三个第一目标矩形。可以理解地，该所确定的三个第一目标矩形的宽度值与宽度阈值相同，长度值小于长度阈值。
S203：根据三个第一目标矩形，对待检测图像进行分割，得到三个第一矩形图像块。
具体地，在根据步骤S202确定了三个第一目标矩形后，则可根据三个第一目标矩形对待检测图像进行分割，得到三个第一矩形图像块。可选地，可采用图像二值化处理方法或采用图像处理工具实现对待检测图像的分割。
S204：对三个第一矩形图像块进行缩放处理，得到三个待检测图像块。
具体地，对三个第一目标矩形进行缩放处理指将得到的三个第一矩形图像块缩放成图像分辨率与分辨率阈值相同的图像的过程。具体地，可采用图像缩放算法实现对三个第一矩形图像块的缩放处理；或者采用图像缩放工具实现对三个第一矩形图像块的缩放处理，得到三个待检测图像块。可选地，图像缩放算法可以为双线性内插值算法或三线性卷积插值算法。图像缩放工具可以为photoshop、iResizer或FastStone Photo Resizer。
本实施例中，若待检测图像的长度值大于长度阈值，且宽度值小于宽度阈值，或待检测图像的长度值小于长度阈值，且宽度值大于宽度阈值，则确定第一参考矩形的第一长度值和第一宽度值；从待检测图像中确定三个第一顶点，并根据每一第一顶点和第一参考矩形的第一长度值和第一宽度值确定三个第一目标矩形；根据三个第一目标矩形，对所述待检测图像进行分割，得到三个第一矩形图像块，对三个第一矩形图像块进行缩放处理，得到三个待检测图像块；先对待检测图像进行分割处理，然后再对切割后的图像进行缩放，在保证能完整获取待检测图像所包含的所有道路损伤特征的同时，还提高了获取的待检测图像块的清晰度。
在一实施例中，如图4所示，待检测图像的图像分辨率包括待检测图像的长度值和宽度值，分辨率阈值包括长度阈值和宽度阈值，若待检测图像的图像分辨率大于分辨率阈值，则对待检测图像进行分割处理，得到N个待检测图像块，具体还包括如下步骤:
S205：若待检测图像的长度值大于长度阈值，且宽度值大于宽度阈值，则确定第二参考矩形的第二长度值和第二宽度值。
其中，待检测图像的长度值大于长度阈值，且宽度值大于宽度阈值包括：待检测图像的长度值大于长度阈值但小于长度阈值的两倍、且宽度值大于宽度阈值但小于宽度阈值的两倍；长度值大于长度阈值但小于长度阈值的两倍、且宽度值大于宽度阈值的两倍；宽度值大于宽度阈值但小于宽度阈值的两倍、且长度值大于长度阈值的两倍；以及长度值和宽度值都大于长度阈值和宽度阈值的两倍以上四种情况。
其中，第二参考矩形指预先设定的一个参考区域。在本实施例中，第二参考矩形的第二长度值和第二宽度值是根据待检测图像的长度值与长度阈值，以及宽度值与宽度阈值的对比关系所确定的。具体地，若待检测图像的长度值大于阈值长度值但小于长度阈值的两倍、且宽度值大于宽度阈值但小于宽度阈值的两倍，则确定第二矩形的第二长度值等于长度阈值，和第二宽度等于宽度阈值。若待检测图像的长度值和宽度值都大于宽度阈值的两倍，则确定第二矩形的第二宽度值等于待检测图像的宽度值的二分之一，第二长度值等于检测损伤图像的长度值的二分之一。若待检测图像的长度值大于长度阈值但小于长度阈值的两倍、且宽度值大于宽度阈值的两倍，则确定第二矩形的第二长度值等于长度阈值、且第二宽度值等于待检测图像的宽度值的二分之一。若待检测图像的宽度值大于宽度阈值但小于宽度阈值的两倍、且长度值大于长度阈值的两倍，则确定第二矩形的第二宽度值等于宽度阈值、且第二长度值等于待检测图像的长度值的二分之一。
S206：从待检测图像中确定五个第二顶点，并根据五个第二顶点和第二参考矩形的第二长度值和第二宽度值确定五个第二目标矩形。
其中，第二顶点指在待检测图像上所确定的点。该待检测图像为长度值大于长度阈值且宽度值大于宽度阈值的图像。具体地，对长度值大于长度阈值且宽度值大于宽度阈值的待检测图像从待检测图像上确定五个第一顶点。
示例性地，建立一坐标系，以待检测图像的左下角为原点；将分辨率阈值表示为(A,B)，待检测图像的长度值和宽度值表示为(c，d)；若待检测图像的长度值和宽度值分别大于对应的长度阈值和宽度阈值、但小于对应的长度阈值和宽度阈值的两倍，则将从待检测图像上确定的五个第二顶点设定为顶点C 1、顶点C 2、顶点C 3、顶点C 4和顶点C 5。从待检测图像上所确定的顶点C 1坐标为(A，d)，从待检测图像上所确定的顶点C 2坐标为(c-A，d)，从待检测图像上所确定的顶点C 3坐标为(A，0)，从待检测图像上所确定的顶点C 4坐标为(c-A，0)，从待检测图像上所确定的顶点C 5坐标为((c–A)/2，d+B/2)。若待检测图像的长度值和宽度值分别大于长度阈值和宽度阈值的两倍以上，则将从待检测图像上确定的五个第二顶点设定为顶点D 1、顶点D 2、顶点D 3、顶点D 4和顶点D 5,从待检测图像上所确定的顶点D 1坐标为(c/2，d)，从待检测图像上所确定的顶点D 2坐标为(c/2，d)，从待检测图像上所确定的顶点D 3坐标为(c/2，0)，从待检测图像上所确定的顶点D 4坐标为(c/2，0)，从待检测图像上所确定的顶点D 5坐标为((c–A)/2，d+B/2)。同样地，对其它两个情况的待检测图像确定的第二顶点坐标，与上述根据待检测图像的长度值和宽度值与分辨率阈值的关系确定的第二顶点坐标的方法相同，此处不做冗余赘述。
进一步地，将根据上述步骤确定的顶点C 1/D 1与第二长度值和第二宽度值结合、且往待检测图像的左上方向确定第二目标矩形；将根据上述步骤确定的顶点C 2/D 2与第二长度值和第二宽度值结合、且往待检测图像的右上方向确定第二目标矩形；将根据上述步骤确定的顶点C 3/D 3与第二长度值和第二宽度值结合、且往待检测图像的左下方向确定第二目标矩形；将根据上述步骤确定的顶点C 4/D 4与第二长度值和第二宽度值结合、且往待检测图像的右下方向确定第二目标矩形；将根据上述步骤确定的顶点C 5/D 5与第一长度值和第一宽度值结合且往待检测图像的中间方向确定第一目标矩形，从而在待检测图像上确定五个第二目标矩形。可以理解地，若待检测图像的长度值大于长度阈值但小于长度阈值的两倍、且宽度值大于宽度阈值但小于宽度阈值的两倍，则确定的第二目标矩形的长度值等于长度阈值、宽度值等于宽度阈值。若待检测图像的长度值和宽度值都大于长度阈值和宽度阈值的两倍，则确定的第二目标矩形的长度值于待检测图像的长度值的二分之一、宽度值等于待检测图像的宽度值的二分之一。
S207：根据五个第二目标矩形，对待检测图像进行分割，得到五个第二矩形图像块。
具体地，在根据步骤S206确定了五个第二目标矩形后，则可根据五个第二目标矩形对待检测图像进行分割，得到五个第二矩形图像块。可选地，可采用图像二值化处理方法或采用图像处理工具实现对待检测图像的分割。
S208：对五个第二矩形图像块进行缩放处理，得到五个待检测图像块。
对根据步骤S205确定的五个第二矩形图像块进行缩放处理，即可得到N个待检测图像块。可以理解地，在本步骤中，N个待检测图像块中的N为5。具体地，对五个第二矩形图像块的缩放处理过程和方法与步骤S203对三个第一目标矩形进行缩放处理的具体过程和方法相同，此处不做冗余赘述。
在本实施例中，若待检测图像的长度值大于长度阈值，且宽度值大于宽度阈值，则确定第二参考矩形的第二长度值和第二宽度值；从待检测图像中确定五个第二顶点，并根据五个第二顶点和第二参考矩形的第二长度值和第二宽度值确定五个第二目标矩形，根据五个所述第二目标矩形，对待检测图像进行分割，得到五个第二矩形图像块；对五个第二矩形图像块进行缩放处理，得到N个待检测图像块；先对待检测图像进行分割处理，然后再对切割后的图像进行缩放，在保证能完整获取待检测图像所包含的所有道路损伤特征的同时，还提高了获取的待检测图像块的清晰度。
在一实施例中，如图5所示，检测信息包括道路损伤类别和位置信息，将每一待检测图像块的检测信息进行合并，得到待检测图像的道路损伤信息，包括如下步骤：
S401：根据道路损伤类别对检测信息进行分类，得到M个检测信息分类集，M为正整数。
其中，检测信息分类集指根据道路损伤类别对检测信息进行分类后所得的分类集。具体地，根据道路损伤类别对检测信息进行分类具体包括：采用正则匹配法，分别将每一检测信息中的道路损伤类别进行一一匹配，若匹配成功，则表示对应的检测信息中的道路损伤类别相同，若匹配失败，则表示对应的检测信息中的道路损伤类别不同；然后让匹配成功的相同道路损伤类别所对应的检测信息归为同一类，将匹配失败的不同道路损伤类别所对应的检测信息单独归类，最后得到M个检测信息分类集。可以理解 地，若获取的检测信息中的道路损伤类别都相同，则进行分类后得到的检测信息分类集为一个，若获取的检测信息中存在不同的道路损伤类别，则进行分类后得到的检测信息分类集为至少两个。
S402：在每一检测信息分类集中，基于位置信息对所述检测信息进行合并，得到检测信息合并集。
具体地，在每一检测信息分类集中，基于位置信息对检测信息进行合并是指将位置信息存在关联的待检测图像块进行合并的过程。在本实施例中，由于待检测图像块的位置信息是由一具体坐标值表示，因此，确定每一待检测图像块的位置信息是否存在关联可通过预先设定一个位置信息关联标准来判断。具体地，位置信息关联标准可以为先确定由每一待检测图像块的位置信息所组成的区域块，然后，再将区域块存在重合部分的待检测图像块进行合并，和将区域块之间的直线距离小于阈值距离的待检测图像块都进行合并。
示例性地，若有一道路损伤类别为DO的检测信息分类集中包含有两不同的位置信息分别为A 1(45,77,65,90)和A 2(55,90,80,100)，经检测可得位置A 1(45,77,65,90)和A 2(55,90,80,100)组成的区域块相互重叠，则将A 1和A 2所对应的检测信息进行合并，得到一检测信息合并集。
S403：将每一检测信息合并集组成待检测图像的道路损伤信息。
其中，目标检测集指由每一检测信息合并集组合形成的检测信息。具体地，将根据步骤S402得到的所有检测信息合并集进行合并，即可得到待检测图像块的道路损伤信息。
优选地，还可先根据基于位置信息对检测信息进行合并；然后在每一检测信息位置集中，根据道路损伤类别对每一检测信息位置集进行分类，将相同道路损伤类别的检测信息归为一类，将不相同道路损伤类别的检测信息单独归类，得到检测信息合并集；最后将每一检测信息合并集组成待检测图像块的道路损伤信息。
在本实施例中，根据道路损伤类别对检测信息进行分类，得到M个检测信息分类集；在每一所述检测信息分类集中，对所述位置信息存在相邻或重叠的检测信息进行合并，得到检测信息合并集；将每一检测信息合并集组成所述待检测图像的道路损伤信息，进一步保证了道路损伤检测结果的完整性。
在一实施例中，如图6所示，在将每一待检测图像块输入到预设的道路损伤检测模型中进行检测，得到每一待检测图像块的检测信息之前，道路损伤检测方法还包括：
S41：取L个原始样本图像，其中，每一原始样本图像都已进行了样本标注，L为正整数。
其中，原始样本图像指包含道路损伤特征的样本图像，L为正整数。每一原始样本图像都已进行了样本标注指预先对每一原始样本图像都进行了道路损伤类别和位置信息的标注。优选地，为了更好的对道路的损伤情况进行分析和判断，在本实施例中，将道路损伤类别分为八个等级，分别为：车辆行驶部分直线裂纹、新旧接缝直线裂纹、等间距直线裂纹、新旧接缝直线裂纹、龟裂纹、凹槽、人行横道模糊和白线模糊；和将位置信息通过过一矩形区域表示。具体地，可以通过矩形区域上的四个点的坐标表示。例如：位置信息表示为(x min，y min，x max，y max)。
优选地，为了方便区别不同道路损伤类别所对应的损伤等级，还可以给不同的道路损伤类别赋予对应的等级标识。其中，等级标识指为了区分不同道路损伤类别所对应的不同损伤等级的一种标识符，可以由阿拉伯数字、大写字母或小写字母中的任意一种组合表示。例如：车辆行驶部分直线裂纹的等级标识符为DO、新旧接缝直线裂纹的等级标识符为D1、等间距直线裂纹的等级标识符为D2、新旧接缝直线裂纹的等级标识符为D3、龟裂纹的等级标识符为D4、凹槽等级的标识符为D5、人行横道模糊的等级标识符为D6和白线模糊的等级标识符为D7。
S42：对每一原始样本图像进行样本预处理，得到目标样本图像。
其中，目标样本图像指对原始样本图像进行样本预处理后所得的图像。具体地，对原始样本图像进行样本预处理包括对原始样本图像进行图像裁剪和图像缩放。优选地，可采用opencv的图像处理算法实现对原始样本图像的图像裁剪和图像缩放，得到目标样本图像。其中，opencv是一个基于BSD许可发行的跨平台计算机视觉库，由一系列C函数和少量C++类构成；opencv同时提供了Python、Ruby、MATLAB等语言的接口，可实现图像处理和计算机视觉方面的很多通用算法。在本实施例中，由于道路损伤检测模型中全连接层的输入向量的维数是固定大小的，因此为了避免出现网络的动态变化，无法实现参数训练的目的，对每一原始样本图像进行样本预处理后得到的目标样本图像的尺寸大小也是固定大 小的。
S43：将每一目标样本图像输入到卷积神经网络中进行训练，得到道路损伤检测模型。
其中，道路损伤检测模型指通过大量的目标样本图像进行训练后所生成的，可对待检测图像进行检测的模型。具体地，将已进行道路损伤类别和位置信息标注的每一目标样本图像输入到卷积神经网络中进行end-to-end训练，当训练到一定迭代次数时，训练停止，得到最终的道路损伤检测模型。优选地，为了更好的支持卷积神经网络中使用的各种算法，该道路损伤检测模型可以使用TensorFlow计算框架进行训练。
优选地，在将每一目标样本图像输入到卷积神经网络中进行训练，得到道路损伤检测模型后，还可以利用bazel工具将使用tensorflow框架训练的模型编译成.so文件和jar包，然后进行android配置，实现将道路损伤检测模型从PC端移植到安卓端。其中，Bazel工具是一个开源的构建和测试工具，它使用一种人易于理解的高级构建语言、且支持多种开发语言的项目，能够基于多个平台来构建。
在本实施例中，首先取L个原始样本图像，其中，每一原始样本图像都已进行了样本标注，L为正整数；然后对每一原始样本图像进行样本预处理，得到目标样本图像；最后将每一目标样本图像输入到卷积神经网络中进行训练，得到道路损伤检测模型；保证了得到的道路损伤检测模型的准确性。
在一实施例中，如图7所示，对每一原始样本图像进行样本预处理，得到目标样本图像，具体包括如下步骤：
S421:获取原始样本图像，对每一原始样本图像进行检测，得到每一原始样本图像的检测信息。
由于获取的原始样本图像中包含道路损伤的图像部分可能占总原始样本图像的比例较小，即一张原始样本图像中可能包含很大部分无用的背景图像。因此，为了提高目标样本图像的准确性和模型训练的效率，需预先对获取的原始样本图像进行检测，得到每一原始样本图像的检测信息，以区分每一原始样本图像中的道路损伤图像部分和背景图像部分。其中，检测信息是指将每一原始样本图像中的道路损伤图像部分与背景图像部分进行划分的标记信息。可选地，可采用边缘检测算法实现每一原始样本图像进行检测，得到每一原始样本图像的检测信息。边缘检测算法可以为opencv的Canny边缘检测算法、Sobel算法或Laplacian算法中的任意一种。
S422：根据检测信息，对每一原始样本图像进行裁剪处理，得到训练样本图像。
其中，训练样本图像指经过图像裁剪处理后得到的样本图像。对原始样本图像进行裁剪处理指根据每一原始样本图像中的检测信息，从原始样本图像中裁剪掉背景图像部分的过程。具体地，采用图像裁剪工具实现对每一原始样本图像的裁剪处理。可选地，图像裁剪工具可以为jQuery Jcrop图像裁剪工具或FOTOE图像裁剪工具等。优选地，还可以采用opencv的图像分割算法自动实现对每一原始样本图像的裁剪处理，得到训练样本图像。
S423：获取预设图像比例，根据预设图像比例对每一训练样本图像进行缩放处理，得到目标样本图像。
其中，预设图像比例指预先设定的标准图像大小。可选地，预设图像比例可以为600*600、750*750或800*800等。在本实施例中，预设图像比例需与步骤S10的分辨率阈值大小相同。同样地，在本实施例中，将预设图像比例设定为600*600。目标样本图像指进行缩放处理后、可直接进行训练的样本图像。具体地，根据预设图像比例对每一训练样本图像进行缩放处理指将每一训练样本图像缩放成预设图像比例大小的目标样本图像的过程。可选地，可采用图像缩放处理算法实现对每一训练样本图像进行缩放处理，得到目标样本图像。图像缩放处理算法可以为邻近插值算法、双线性插值算法或三次卷积法等。优选地，还可以采用opencv的图像缩放API工具自动实现对裁剪后的训练样本图像进行缩放处理，得到目标样本图像。
在本实施例中，通过获取原始样本图像，对每一原始样本图像进行检测，得到每一原始样本图像的检测信息；然后根据检测信息，对每一原始样本图像进行裁剪处理，得到训练样本图像；最后获取预设图像比例，根据预设图像比例对每一训练样本图像进行缩放处理，得到目标样本图像，进一步保证了获取的目标样本图像的准确性和有效性。
在一实施例中，如图8所示，将每一目标样本图像输入到卷积神经网络中进行训练，得到道路损伤 检测模型，具体包括如下步骤：
S441：初始化卷积神经网络模型的模型参数。
其中，初始化卷积神经网络指预先初始化卷积神经网络模型的模型参数(即卷积核和偏置)。卷积核是指卷积神经网络的权值，当输入训练数据时，会乘上一个权值即卷积核，然后得到神经元的输出，它反映了训练数据的重要程度。偏置是用于更改权重乘输入的范围的线性分量。
S442：采用卷积神经网络对每一目标样本图像进行特征提取，得到道路损伤特征。
其中，道路损伤特征是采用卷积神经网络对训练集中的目标样本图像进行特征提取所得到的损伤特征。具体地，采用卷积神经网络对目标样本图像进行特征提取。卷积运算的计算公式包括
Figure PCTCN2019102670-appb-000001
*代表卷积运算；x j代表第j个输入特征图；y j代表第j个输出特征图；w ij是第i个输入特征图与第j个输出特征图之间的卷积核(权值)；b j代表第j个输出特征图的偏置项。采用最大池化下采样对卷积后的特征图进行下采样操作以实现对特征图的降维，其计算公式为
Figure PCTCN2019102670-appb-000002
其中，y j表示下采样过程中的第i个输出谱(即下采样后的特征图)，下采样过程中的每一个神经元是从第i个输入谱(卷积后的特征图)中采用S*S的下采样框局部采样得到的，即对输入谱进行S倍下采样，其中S的具体数值可以根据实际采样需要进行设定；m与n分别表示下采样框移动的步长。
S443：将道路损伤特征输入到卷积神经网络模型中进行训练，得到道路损伤检测模型。
具体地，基于卷积神经网络的特征，将根据步骤S442得到的道路损伤特征输入到卷积神经网络模型中进行训练，即可得到道路损伤检测模型。
在本实施例中，通过初始化卷积神经网络模型的模型参数；然后采用卷积神经网络对每一目标样本图像进行特征提取，得到道路损伤特征；最后将道路损伤特征输入到卷积神经网络模型中进行训练，得到道路损伤检测模型；进一步提高了道路损伤检测模型的准确性。
应理解，上述实施例中各步骤的序号的大小并不意味着执行顺序的先后，各过程的执行顺序应以其功能和内在逻辑确定，而不应对本申请实施例的实施过程构成任何限定。
在一实施例中，提供一种道路损伤检测装置，该道路损伤检测装置与上述实施例中道路损伤检测方法一一对应。如图9所示，该道路损伤检测装置包括分辨率判断模块10、分割处理模块20、检测模块30和合并模块40。各功能模块详细说明如下：
分辨率判断模块10，用于获取待检测图像，判断待检测图像的图像分辨率是否大于分辨率阈值；
分割处理模块20，用于在待检测图像的图像分辨率大于分辨率阈值时，对待检测图像进行分割处理，得到N个待检测图像块，N为正整数；
检测模块30，用于将每一待检测图像块输入到预设的道路损伤检测模型中进行检测，得到每一待检测图像块的检测信息；
合并模块40，用于将每一待检测图像块的检测信息进行合并，得到待检测图像的道路损伤信息。
优选地，分割处理模块20包括：
第一确定单元，用于在待检测图像的长度值大于长度阈值且宽度值小于宽度阈值，或者，待检测图像的长度值小于长度阈值且宽度值大于宽度阈值时，确定第一参考矩形的第一长度值和第一宽度值；
第二确定单元，用于从待检测图像中确定三个第一顶点，并根据三个第一顶点和第一参考矩形的第一长度值和第一宽度值确定三个第一目标矩形；
第一分割单元，用于根据三个第一目标矩形，对待检测图像进行分割，得到三个第一矩形图像块；
第一缩放处理单元，用于对三个第一矩形图像块进行缩放处理，得到三个待检测图像块。
优选地，分割处理模块20还包括：
第三确定单元，用于在待检测图像的长度值大于长度阈值，且宽度值大于宽度阈值时，确定第二参 考矩形的第二长度值和第二宽度值；
第四确定单元，用于从待检测图像中确定五个第二顶点，并根据五个第二顶点和第二参考矩形的第二长度值和第二宽度值确定五个第二目标矩形；
第二分割单元，用于根据五个第二目标矩形，对待检测图像进行分割，得到五个第二矩形图像块；
第二缩放处理单元，用于对五个第二矩形图像块进行缩放处理，得到五个待检测图像块。
优选地，合并模块40包括：
分类单元，用于根据道路损伤类别对检测信息进行分类，得到M个检测信息分类集,M为正整数；
合并单元，用于在每一检测信息分类集中，基于位置信息对检测信息进行合并，得到检测信息合并集；
组成单元，用于将每一检测信息合并集组成待检测图像的道路损伤信息。
优选地，道路损伤检测装置还包括：
原始样本图像获取模块，用于获取L个原始样本图像，其中，每一原始样本图像都已进行了样本标注，L为正整数；
预处理模块，用于对每一原始样本图像进行样本预处理，得到目标样本图像；
训练模块，用于将每一目标样本图像输入到卷积神经网络中进行训练，得到道路损伤检测模型。
优选地，预处理模块包括：
检测单元，用于获取原始样本图像，对每一原始样本图像进行检测，得到每一原始样本图像的检测信息；
裁剪处理单元，用于根据检测信息，对每一原始样本图像进行裁剪处理，得到训练样本图像；
第三缩放处理单元，用于获取预设图像比例，根据预设图像比例对每一训练样本图像进行缩放处理，得到目标样本图像。
优选地，训练模块包括：
初始化单元，用于初始化卷积神经网络模型的模型参数；
特征提取单元，用于采用卷积神经网络对每一目标样本图像进行特征提取，得到道路损伤特征；
训练单元，用于将道路损伤特征输入到卷积神经网络模型中进行训练，得到道路损伤检测模型。
关于道路损伤检测装置的具体限定可以参见上文中对于道路损伤检测方法的限定，在此不再赘述。上述道路损伤检测装置中的各个模块可全部或部分通过软件、硬件及其组合来实现。上述各模块可以硬件形式内嵌于或独立于计算机设备中的处理器中，也可以以软件形式存储于计算机设备中的存储器中，以便于处理器调用执行以上各个模块对应的操作。
在一个实施例中，提供了一种计算机设备，该计算机设备可以是服务器，其内部结构图可以如图10所示。该计算机设备包括通过系统总线连接的处理器、存储器、网络接口和数据库。其中，该计算机设备的处理器用于提供计算和控制能力。该计算机设备的存储器包括非易失性存储介质、内存储器。该非易失性存储介质存储有操作系统、计算机可读指令和数据库。该内存储器为非易失性存储介质中的操作系统和计算机可读指令的运行提供环境。该计算机设备的数据库用于存储上述实施例中的道路损伤检测方法中使用到的数据。该计算机设备的网络接口用于与外部的终端通过网络连接通信。该计算机可读指令被处理器执行时以实现一种道路损伤检测方法。
在一个实施例中，提供了一种计算机设备，包括存储器、处理器及存储在存储器上并可在处理器上运行的计算机可读指令，处理器执行计算机可读指令时实现以下步骤：
获取待检测图像，判断所述待检测图像的图像分辨率是否大于分辨率阈值；
若所述待检测图像的图像分辨率大于所述分辨率阈值，则对所述待检测图像进行分割处理，得到N个待检测图像块，N为正整数；
将每一所述待检测图像块输入到预设的道路损伤检测模型中进行检测，得到每一所述待检测图像块的检测信息；
将每一所述待检测图像块的所述检测信息进行合并，得到所述待检测图像的道路损伤信息。
在一个实施例中，一个或多个存储有计算机可读指令的非易失性可读存储介质，所述计算机可读指 令被一个或多个处理器执行时，使得所述一个或多个处理器执行如下步骤：
获取待检测图像，判断所述待检测图像的图像分辨率是否大于分辨率阈值；
若所述待检测图像的图像分辨率大于所述分辨率阈值，则对所述待检测图像进行分割处理，得到N个待检测图像块，N为正整数；
将每一所述待检测图像块输入到预设的道路损伤检测模型中进行检测，得到每一所述待检测图像块的检测信息；
将每一所述待检测图像块的所述检测信息进行合并，得到所述待检测图像的道路损伤信息。
本领域普通技术人员可以理解实现上述实施例方法中的全部或部分流程，是可以通过计算机可读指令来指令相关的硬件来完成，所述的计算机可读指令可存储于一非易失性计算机可读取存储介质中，该计算机可读指令在执行时，可包括如上述各方法的实施例的流程。其中，本申请所提供的各实施例中所使用的对存储器、存储、数据库或其它介质的任何引用，均可包括非易失性和/或易失性存储器。非易失性存储器可包括只读存储器(ROM)、可编程ROM(PROM)、电可编程ROM(EPROM)、电可擦除可编程ROM(EEPROM)或闪存。易失性存储器可包括随机存取存储器(RAM)或者外部高速缓冲存储器。作为说明而非局限，RAM以多种形式可得，诸如静态RAM(SRAM)、动态RAM(DRAM)、同步DRAM(SDRAM)、双数据率SDRAM(DDRSDRAM)、增强型SDRAM(ESDRAM)、同步链路(Synchlink)DRAM(SLDRAM)、存储器总线(Rambus)直接RAM(RDRAM)、直接存储器总线动态RAM(DRDRAM)、以及存储器总线动态RAM(RDRAM)等。
所属领域的技术人员可以清楚地了解到，为了描述的方便和简洁，仅以上述各功能单元、模块的划分进行举例说明，实际应用中，可以根据需要而将上述功能分配由不同的功能单元、模块完成，即将所述装置的内部结构划分成不同的功能单元或模块，以完成以上描述的全部或者部分功能。
以上所述实施例仅用以说明本申请的技术方案，而非对其限制；尽管参照前述实施例对本申请进行了详细的说明，本领域的普通技术人员应当理解：其依然可以对前述各实施例所记载的技术方案进行修改，或者对其中部分技术特征进行等同替换；而这些修改或者替换，并不使相应技术方案的本质脱离本申请各实施例技术方案的精神和范围，均应包含在本申请的保护范围之内。

Claims (20)

  1. 一种道路损伤检测方法，其特征在于，包括：
    获取待检测图像，判断所述待检测图像的图像分辨率是否大于分辨率阈值；
    若所述待检测图像的图像分辨率大于所述分辨率阈值，则对所述待检测图像进行分割处理，得到N个待检测图像块，N为正整数；
    将每一所述待检测图像块输入到预设的道路损伤检测模型中进行检测，得到每一所述待检测图像块的检测信息；
    将每一所述待检测图像块的所述检测信息进行合并，得到所述待检测图像的道路损伤信息。
  2. 如权利要求1所述的道路损伤检测方法，其特征在于，所述待检测图像的图像分辨率包括待检测图像的长度值和宽度值，所述分辨率阈值包括长度阈值和宽度阈值；
    若所述待检测图像的图像分辨率大于分辨率阈值，则对待检测图像进行分割处理，得到N个待检测图像块，包括：
    若所述待检测图像的长度值大于长度阈值且宽度值小于宽度阈值，或者，若所述待检测图像的长度值小于长度阈值且宽度值大于宽度阈值，则确定第一参考矩形的第一长度值和第一宽度值；
    从所述待检测图像中确定三个第一顶点，并根据三个所述第一顶点和所述第一参考矩形的所述第一长度值和所述第一宽度值确定三个第一目标矩形；
    根据三个所述第一目标矩形，对所述待检测图像进行分割，得到三个第一矩形图像块；
    对所述三个第一矩形图像块进行缩放处理，得到三个待检测图像块。
  3. 如权利要求1所述的道路损伤检测方法，其特征在于，所述待检测图像的图像分辨率包括待检测图像的长度值和宽度值，所述分辨率阈值包括长度阈值和宽度阈值；
    若所述待检测图像的图像分辨率大于分辨率阈值，则对待检测图像进行分割处理，得到N个待检测图像块，包括：
    若所述待检测图像的长度值大于长度阈值，且宽度值大于宽度阈值，则确定第二参考矩形的第二长度值和第二宽度值；
    从所述待检测图像中确定五个第二顶点，并根据五个所述第二顶点和所述第二参考矩形的所述第二长度值和所述第二宽度值确定五个第二目标矩形；
    根据五个所述第二目标矩形，对所述待检测图像进行分割，得到五个第二矩形图像块；
    对所述五个第二矩形图像块进行缩放处理，得到五个待检测图像块。
  4. 如权利要求1所述的道路损伤检测方法，其特征在于，所述检测信息包括道路损伤类别和位置信息；
    所述将每一待检测图像块的检测信息进行合并，得到待检测图像的道路损伤信息，包括：
    根据所述道路损伤类别对所述检测信息进行分类，得到M个检测信息分类集,M为正整数；
    在每一所述检测信息分类集中，基于所述位置信息对所述检测信息进行合并，得到检测信息合并集；
    将每一检测信息合并集组成所述待检测图像的道路损伤信息。
  5. 如权利要求1所述的道路损伤检测方法，其特征在于，在所述将每一待检测图像块输入到预设的道路损伤检测模型中进行检测，得到每一待检测图像块的检测信息之前，所述道路损伤检测方法还包括：
    获取L个原始样本图像，其中，每一所述原始样本图像都已进行了样本标注，L为正整数；
    对每一所述原始样本图像进行样本预处理，得到目标样本图像；
    将每一所述目标样本图像输入到卷积神经网络中进行训练，得到道路损伤检测模型。
  6. 如权利要求5所述的道路损伤检测方法，其特征在于，所述对每一原始样本图像进行样本预处理，得到目标样本图像，包括：
    获取原始样本图像，对每一所述原始样本图像进行检测，得到每一所述原始样本图像的检测信息；
    根据所述检测信息，对每一所述原始样本图像进行裁剪处理，得到训练样本图像；
    获取预设图像比例，根据所述预设图像比例对每一所述训练样本图像进行缩放处理，得到目标样本 图像。
  7. 如权利要求5所述的道路损伤检测方法，其特征在于，所述将每一目标样本图像输入到卷积神经网络中进行训练，得到道路损伤检测模型，包括：
    初始化卷积神经网络模型的模型参数；
    采用卷积神经网络对每一所述目标样本图像进行特征提取，得到道路损伤特征；
    将所述道路损伤特征输入到卷积神经网络模型中进行训练，得到道路损伤检测模型。
  8. 一种道路损伤检测装置，其特征在于，包括：
    分辨率判断模块，用于获取待检测图像，判断所述待检测图像的图像分辨率是否大于分辨率阈值；
    分割处理模块，用于在所述待检测图像的图像分辨率大于所述分辨率阈值时，对所述待检测图像进行分割处理，得到N个待检测图像块，N为正整数；
    检测模块，用于将每一所述待检测图像块输入到预设的道路损伤检测模型中进行检测，得到每一所述待检测图像块的检测信息；
    合并模块，用于将每一所述待检测图像块的所述检测信息进行合并，得到所述待检测图像的道路损伤信息。
  9. 如权利要求8所述的道路损伤检测装置，其特征在于，所述分割处理模块，包括；
    第一确定单元，用于在待检测图像的长度值大于长度阈值且宽度值小于宽度阈值，或者，待检测图像的长度值小于长度阈值且宽度值大于宽度阈值时，确定第一参考矩形的第一长度值和第一宽度值；
    第二确定单元，用于从待检测图像中确定三个第一顶点，并根据三个第一顶点和第一参考矩形的第一长度值和第一宽度值确定三个第一目标矩形；
    第一分割单元，用于根据三个第一目标矩形，对待检测图像进行分割，得到三个第一矩形图像块；
    第一缩放处理单元，用于对三个第一矩形图像块进行缩放处理，得到三个待检测图像块。
  10. 如权利要求8所述的道路损伤检测装置，其特征在于，所述分割处理模块，还包括；
    第三确定单元，用于在待检测图像的长度值大于长度阈值，且宽度值大于宽度阈值时，确定第二参考矩形的第二长度值和第二宽度值；
    第四确定单元，用于从待检测图像中确定五个第二顶点，并根据五个第二顶点和第二参考矩形的第二长度值和第二宽度值确定五个第二目标矩形；
    第二分割单元，用于根据五个第二目标矩形，对待检测图像进行分割，得到五个第二矩形图像块；
    第二缩放处理单元，用于对五个第二矩形图像块进行缩放处理，得到五个待检测图像块。
  11. 如权利要求8所述的道路损伤检测装置，其特征在于，所述合并模块，包括：
    分类单元，用于根据道路损伤类别对检测信息进行分类，得到M个检测信息分类集,M为正整数；
    合并单元，用于在每一检测信息分类集中，基于位置信息对检测信息进行合并，得到检测信息合并集；
    组成单元，用于将每一检测信息合并集组成待检测图像的道路损伤信息。
  12. 如权利要求8所述的道路损伤检测装置，其特征在于，所述道路损伤检测装置还包括：
    原始样本图像获取模块，用于获取L个原始样本图像，其中，每一原始样本图像都已进行了样本标注，L为正整数；
    预处理模块，用于对每一原始样本图像进行样本预处理，得到目标样本图像；
    训练模块，用于将每一目标样本图像输入到卷积神经网络中进行训练，得到道路损伤检测模型。
  13. 如权利要求12所述的道路损伤检测装置，其特征在于，所述预处理模块，包括：
    检测单元，用于获取原始样本图像，对每一原始样本图像进行检测，得到每一原始样本图像的检测信息；
    裁剪处理单元，用于根据检测信息，对每一原始样本图像进行裁剪处理，得到训练样本图像；
    第三缩放处理单元，用于获取预设图像比例，根据预设图像比例对每一训练样本图像进行缩放处理，得到目标样本图像。
  14. 如权利要求12所述的道路损伤检测装置，其特征在于，所述训练模块包括：
    初始化单元，用于初始化卷积神经网络模型的模型参数；
    特征提取单元，用于采用卷积神经网络对每一目标样本图像进行特征提取，得到道路损伤特征；
    训练单元，用于将道路损伤特征输入到卷积神经网络模型中进行训练，得到道路损伤检测模型。
  15. 一种计算机设备，包括存储器、处理器以及存储在所述存储器中并可在所述处理器上运行的计算机可读指令，其特征在于，所述处理器执行所述计算机可读指令时实现如下步骤：
    获取待检测图像，判断所述待检测图像的图像分辨率是否大于分辨率阈值；
    若所述待检测图像的图像分辨率大于所述分辨率阈值，则对所述待检测图像进行分割处理，得到N个待检测图像块，N为正整数；
    将每一所述待检测图像块输入到预设的道路损伤检测模型中进行检测，得到每一所述待检测图像块的检测信息；
    将每一所述待检测图像块的所述检测信息进行合并，得到所述待检测图像的道路损伤信息。
  16. 如权利要求15所述的计算机设备，其特征在于，所述待检测图像的图像分辨率包括待检测图像的长度值和宽度值，所述分辨率阈值包括长度阈值和宽度阈值；
    若所述待检测图像的图像分辨率大于分辨率阈值，则对待检测图像进行分割处理，得到N个待检测图像块，包括：
    若所述待检测图像的长度值大于长度阈值且宽度值小于宽度阈值，或者，若所述待检测图像的长度值小于长度阈值且宽度值大于宽度阈值，则确定第一参考矩形的第一长度值和第一宽度值；
    从所述待检测图像中确定三个第一顶点，并根据三个所述第一顶点和所述第一参考矩形的所述第一长度值和所述第一宽度值确定三个第一目标矩形；
    根据三个所述第一目标矩形，对所述待检测图像进行分割，得到三个第一矩形图像块；
    对所述三个第一矩形图像块进行缩放处理，得到三个待检测图像块。
  17. 如权利要求15所述的计算机设备，其特征在于，所述待检测图像的图像分辨率包括待检测图像的长度值和宽度值，所述分辨率阈值包括长度阈值和宽度阈值；
    若所述待检测图像的图像分辨率大于分辨率阈值，则对待检测图像进行分割处理，得到N个待检测图像块，包括：
    若所述待检测图像的长度值大于长度阈值，且宽度值大于宽度阈值，则确定第二参考矩形的第二长度值和第二宽度值；
    从所述待检测图像中确定五个第二顶点，并根据五个所述第二顶点和所述第二参考矩形的所述第二长度值和所述第二宽度值确定五个第二目标矩形；
    根据五个所述第二目标矩形，对所述待检测图像进行分割，得到五个第二矩形图像块；
    对所述五个第二矩形图像块进行缩放处理，得到五个待检测图像块。
  18. 如权利要求15所述的计算机设备，其特征在于，所述检测信息包括道路损伤类别和位置信息；
    所述将每一待检测图像块的检测信息进行合并，得到待检测图像的道路损伤信息，包括：
    根据所述道路损伤类别对所述检测信息进行分类，得到M个检测信息分类集,M为正整数；
    在每一所述检测信息分类集中，基于所述位置信息对所述检测信息进行合并，得到检测信息合并集；
    将每一检测信息合并集组成所述待检测图像的道路损伤信息。
  19. 如权利要求15所述的计算机设备，其特征在于，在所述将每一待检测图像块输入到预设的道路损伤检测模型中进行检测，得到每一待检测图像块的检测信息之前，所述处理器执行所述计算机可读指令时还实现如下步骤：
    获取L个原始样本图像，其中，每一所述原始样本图像都已进行了样本标注，L为正整数；
    对每一所述原始样本图像进行样本预处理，得到目标样本图像；
    将每一所述目标样本图像输入到卷积神经网络中进行训练，得到道路损伤检测模型。
  20. 一个或多个存储有计算机可读指令的非易失性可读存储介质，所述计算机可读指令被一个或多个处理器执行时，使得所述一个或多个处理器执行如下步骤:
    获取待检测图像，判断所述待检测图像的图像分辨率是否大于分辨率阈值；
    若所述待检测图像的图像分辨率大于所述分辨率阈值，则对所述待检测图像进行分割处理，得到N个待检测图像块，N为正整数；
    将每一所述待检测图像块输入到预设的道路损伤检测模型中进行检测，得到每一所述待检测图像块的检测信息；
    将每一所述待检测图像块的所述检测信息进行合并，得到所述待检测图像的道路损伤信息。
PCT/CN2019/102670 2019-06-14 2019-08-27 道路损伤检测方法、装置、计算机设备及存储介质 Ceased WO2020248371A1 (zh)

Applications Claiming Priority (2)

Application Number Priority Date Filing Date Title
CN201910517637.0A CN110390666B (zh) 2019-06-14 2019-06-14 道路损伤检测方法、装置、计算机设备及存储介质
CN201910517637.0 2019-06-14

Publications (1)

Publication Number Publication Date
WO2020248371A1 true WO2020248371A1 (zh) 2020-12-17

Family

ID=68285452

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/CN2019/102670 Ceased WO2020248371A1 (zh) 2019-06-14 2019-08-27 道路损伤检测方法、装置、计算机设备及存储介质

Country Status (2)

Country Link
CN (1) CN110390666B (zh)
WO (1) WO2020248371A1 (zh)

Cited By (10)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN112597960A (zh) * 2020-12-30 2021-04-02 珠海广通汽车有限公司 图像的处理方法、处理装置与计算机可读存储介质
US20210110200A1 (en) * 2019-10-11 2021-04-15 Cargo Spectre Method, system, and apparatus for damage assessment and classification
CN113869134A (zh) * 2021-09-03 2021-12-31 中航航空电子有限公司 图像中目标的检测方法、装置、计算机设备及存储介质
CN114170171A (zh) * 2021-11-30 2022-03-11 北京建筑大学 文物岩石表面裂缝检测方法、装置、电子设备及存储介质
CN114445627A (zh) * 2021-12-29 2022-05-06 国创移动能源创新中心(江苏)有限公司 运营充电桩的充电枪检测方法和装置
CN115565053A (zh) * 2022-10-13 2023-01-03 中国国家铁路集团有限公司 一种无砟轨道表观伤损样本数据库的构建方法
CN116935351A (zh) * 2022-03-31 2023-10-24 成都圭目机器人有限公司 道面痕迹类别确定方法、装置、设备及存储介质
CN117437221A (zh) * 2023-12-18 2024-01-23 浙江祥晋汽车零部件股份有限公司 一种基于图像检测的亮饰条检测方法及系统
CN119169015A (zh) * 2024-11-21 2024-12-20 浙江大华技术股份有限公司 一种破损伸缩缝的检测方法、装置及存储介质
CN120031733A (zh) * 2025-04-23 2025-05-23 江苏智城慧宁交通科技有限公司 公路巡查养护图像的对抗生成数据增强方法及系统

Families Citing this family (17)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN111178428B (zh) * 2019-12-27 2024-03-22 上海联影智能医疗科技有限公司 软骨损伤分类方法、装置、计算机设备和存储介质
CN111263085A (zh) * 2020-01-21 2020-06-09 中国航空无线电电子研究所 基于块式存储操作的航空显示任务旋转处理系统
CN111476799A (zh) * 2020-03-23 2020-07-31 福建星网物联信息系统有限公司 一种图像分析方法及存储介质
CN111402275A (zh) * 2020-04-27 2020-07-10 Oppo广东移动通信有限公司 镂空检测方法、系统、设备及存储介质
CN111860166B (zh) * 2020-06-18 2024-07-12 浙江大华技术股份有限公司 图像检测的方法、装置、计算机设备和存储介质
CN112098636B (zh) * 2020-09-15 2021-06-11 甘肃天马公路建筑工程有限公司 一种探测道路内部损伤程度的便于移动的装置
CN112488177B (zh) * 2020-11-26 2025-03-21 金蝶软件（中国）有限公司 一种图像匹配方法及相关设备
CN112446870B (zh) * 2020-12-02 2024-07-09 平安科技（深圳）有限公司 管道损伤检测方法、装置、设备及存储介质
CN112950465B (zh) * 2021-01-26 2025-09-19 北京小米移动软件有限公司 视频超分处理方法、视频超分处理装置及存储介质
CN112819814A (zh) * 2021-02-25 2021-05-18 汇鸿智能科技(辽宁)有限公司 金相组织的识别方法、装置、计算机设备及存储介质
CN112907557B (zh) * 2021-03-15 2025-08-08 腾讯科技（深圳）有限公司 道路检测方法、装置、计算设备及存储介质
CN113537016B (zh) * 2021-07-06 2023-01-06 南昌市微轲联信息技术有限公司 一种道路巡查中道路损伤自动检测预警的方法
CN113901926B (zh) * 2021-10-12 2025-03-18 智道网联科技(北京)有限公司 用于自动驾驶的目标检测方法、装置、设备及存储介质
CN115115611B (zh) * 2022-07-21 2023-04-07 明觉科技（北京）有限公司 车辆损伤识别方法、装置、电子设备和存储介质
CN115861218A (zh) * 2022-11-28 2023-03-28 天翼云科技有限公司 一种基于图像处理的道路损伤检测方法
WO2024260008A1 (zh) * 2023-06-21 2024-12-26 清华大学 道路损伤检测方法及装置、道路传感器布设位置的确定方法及装置、计算机设备和存储介质
CN116596931B (zh) * 2023-07-18 2023-11-17 宁德时代新能源科技股份有限公司 图像处理方法、装置、设备、存储介质和程序产品

Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN107424150A (zh) * 2017-07-27 2017-12-01 济南浪潮高新科技投资发展有限公司 一种基于卷积神经网络的道路破损检测方法及装置
CN109146849A (zh) * 2018-07-26 2019-01-04 昆明理工大学 一种基于卷积神经网络和图像识别的路面裂缝检测方法
KR20190030328A (ko) * 2017-09-14 2019-03-22 한국도로공사 도로의 손상을 탐지하는 단말 및 이를 적용한 도로 손상 탐지 시스템
CN109801282A (zh) * 2019-01-24 2019-05-24 湖北大学 路面状况检测方法、处理方法、装置及系统

Family Cites Families (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN105303189B (zh) * 2014-07-29 2019-08-20 阿里巴巴集团控股有限公司 一种用于检测预定区域中特定标识图像的方法及装置
CN105631809B (zh) * 2015-12-31 2018-11-06 北京理工大学 一种非均匀分辨率球面全景图生成方法
CN107958458B (zh) * 2016-10-17 2021-01-22 京东方科技集团股份有限公司 图像分割方法、图像分割系统及包括其的设备
US11704771B2 (en) * 2017-12-01 2023-07-18 Huawei Technologies Co., Ltd. Training super-resolution convolutional neural network model using a high-definition training image, a low-definition training image, and a mask image

Patent Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN107424150A (zh) * 2017-07-27 2017-12-01 济南浪潮高新科技投资发展有限公司 一种基于卷积神经网络的道路破损检测方法及装置
KR20190030328A (ko) * 2017-09-14 2019-03-22 한국도로공사 도로의 손상을 탐지하는 단말 및 이를 적용한 도로 손상 탐지 시스템
CN109146849A (zh) * 2018-07-26 2019-01-04 昆明理工大学 一种基于卷积神经网络和图像识别的路面裂缝检测方法
CN109801282A (zh) * 2019-01-24 2019-05-24 湖北大学 路面状况检测方法、处理方法、装置及系统

Cited By (12)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20210110200A1 (en) * 2019-10-11 2021-04-15 Cargo Spectre Method, system, and apparatus for damage assessment and classification
US11727522B2 (en) * 2019-10-11 2023-08-15 Cargo Spectre Method, system, and apparatus for damage assessment and classification
CN112597960A (zh) * 2020-12-30 2021-04-02 珠海广通汽车有限公司 图像的处理方法、处理装置与计算机可读存储介质
CN113869134A (zh) * 2021-09-03 2021-12-31 中航航空电子有限公司 图像中目标的检测方法、装置、计算机设备及存储介质
CN114170171A (zh) * 2021-11-30 2022-03-11 北京建筑大学 文物岩石表面裂缝检测方法、装置、电子设备及存储介质
CN114445627A (zh) * 2021-12-29 2022-05-06 国创移动能源创新中心(江苏)有限公司 运营充电桩的充电枪检测方法和装置
CN116935351A (zh) * 2022-03-31 2023-10-24 成都圭目机器人有限公司 道面痕迹类别确定方法、装置、设备及存储介质
CN115565053A (zh) * 2022-10-13 2023-01-03 中国国家铁路集团有限公司 一种无砟轨道表观伤损样本数据库的构建方法
CN117437221A (zh) * 2023-12-18 2024-01-23 浙江祥晋汽车零部件股份有限公司 一种基于图像检测的亮饰条检测方法及系统
CN117437221B (zh) * 2023-12-18 2024-04-26 浙江祥晋汽车零部件股份有限公司 一种基于图像检测的亮饰条检测方法及系统
CN119169015A (zh) * 2024-11-21 2024-12-20 浙江大华技术股份有限公司 一种破损伸缩缝的检测方法、装置及存储介质
CN120031733A (zh) * 2025-04-23 2025-05-23 江苏智城慧宁交通科技有限公司 公路巡查养护图像的对抗生成数据增强方法及系统

Also Published As

Publication number Publication date
CN110390666B (zh) 2023-06-27
CN110390666A (zh) 2019-10-29

Similar Documents

Publication Publication Date Title
WO2020248371A1 (zh) 道路损伤检测方法、装置、计算机设备及存储介质
US12073542B2 (en) Image processing method, image processing apparatus, and non-transitory storage medium
US11682197B2 (en) Computer vision systems and methods for geospatial property feature detection and extraction from digital images
CN109492643B (zh) 基于ocr的证件识别方法、装置、计算机设备及存储介质
CN109241904B (zh) 文字识别模型训练、文字识别方法、装置、设备及介质
US11003941B2 (en) Character identification method and device
CN110163842B (zh) 建筑裂缝检测方法、装置、计算机设备和存储介质
WO2021017261A1 (zh) 识别模型训练方法、图像识别方法、装置、设备及介质
WO2021212659A1 (zh) 视频数据处理方法、装置、计算机设备及存储介质
CN110659647A (zh) 印章图像识别方法及装置、智能发票识别设备和存储介质
WO2018233038A1 (zh) 基于深度学习的车牌识别方法、装置、设备及存储介质
WO2021151319A1 (zh) 卡片边框检测方法、装置、设备及可读存储介质
CN106683073B (zh) 一种车牌的检测方法及摄像机和服务器
WO2021012382A1 (zh) 配置聊天机器人的方法、装置、计算机设备和存储介质
WO2018233055A1 (zh) 保单信息录入的方法、装置、计算机设备及存储介质
CN111242126A (zh) 不规则文本校正方法、装置、计算机设备和存储介质
WO2023130648A1 (zh) 一种图像数据增强方法、装置、计算机设备和存储介质
WO2021047484A1 (zh) 文字识别方法和终端设备
CN112307989A (zh) 路面物体识别方法、装置、计算机设备和存储介质
CN111259878A (zh) 一种检测文本的方法和设备
WO2021098861A1 (zh) 识别文本的方法、装置、识别设备和存储介质
CN115272887A (zh) 基于无人机检测的海岸带垃圾识别方法、装置以及设备
WO2022134354A1 (zh) 车损检测模型训练、车损检测方法、装置、设备及介质
WO2024001051A1 (zh) 空间组学单细胞数据获取方法及装置和电子设备
CN114359352A (zh) 图像处理方法、装置、设备、存储介质及计算机程序产品

Legal Events

Date Code Title Description
121 Ep: the epo has been informed by wipo that ep was designated in this application

Ref document number: 19933142

Country of ref document: EP

Kind code of ref document: A1

NENP Non-entry into the national phase

Ref country code: DE

122 Ep: pct application non-entry in european phase

Ref document number: 19933142

Country of ref document: EP

Kind code of ref document: A1