CN114359195A - Glass curtain wall crack detection method - Google Patents
Glass curtain wall crack detection method Download PDFInfo
- Publication number
- CN114359195A CN114359195A CN202111612351.4A CN202111612351A CN114359195A CN 114359195 A CN114359195 A CN 114359195A CN 202111612351 A CN202111612351 A CN 202111612351A CN 114359195 A CN114359195 A CN 114359195A
- Authority
- CN
- China
- Prior art keywords
- glass curtain
- image
- curtain wall
- ulbp
- sub
- 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.)
- Pending
Links
Images
Landscapes
- Investigating Materials By The Use Of Optical Means Adapted For Particular Applications (AREA)
- Image Analysis (AREA)
Abstract
The invention relates to a glass curtain wall crack detection method, which comprises the following steps: step (1): collecting a glass curtain wall image; step (2): converting the glass curtain wall image into a gray level image; and (3): calculating a first feature of the grayscale image based on the modified ULBP descriptor; and (4): reducing the dimension of the first feature of the gray level image to obtain a macro feature; and (5): and inputting the macro features into a classifier, and detecting whether the glass curtain wall image has cracks or not through the classifier. The invention can effectively detect the cracks of the glass curtain wall and has higher detection precision.
Description
Technical Field
The invention relates to the technical field of glass crack detection, in particular to a glass curtain wall crack detection method.
Background
Detection of structural damage to a building is critical to the safety and maintenance of the building. Cracks in the glass facade of a high-rise building are one of the important factors influencing the safety of buildings, residents and passers-by. Thus, early detection thereof can save a great deal of time and maintenance cost and eliminate potential secondary hazards.
High-rise buildings such as skyscrapers are common in modern cities, so that the method has great significance for automatically detecting cracks of glass curtain walls of skyscrapers. The glass curtain wall of the building is mainly made of tempered glass and other decorative materials. Glass curtain walls need to be monitored and maintained regularly and permanently to ensure the safety of the building and to ensure the safety of the streets and pedestrians nearby under the building. However, manually inspecting glass curtain walls of such buildings is very time consuming, expensive, and may endanger the lives of inspectors. Some glass crack detection systems applied to indoor environment exist in the early days, but due to the change of conditions such as weather, illumination and the like and the influence of dirt and light reflection on glass, the glass crack detection system applied to indoor environment is not successfully applied to the detection of the external glass curtain wall of the building. In recent years, as the unmanned aerial vehicle technology becomes mature, the machine learning technology has made remarkable progress in the aspect of computer vision application. Image data may be acquired quickly using Unmanned Aerial Vehicle (UAV) equipment in environments unsuitable for human detection. At present, a few semi-automatic curtain wall inspection systems based on unmanned aerial vehicles exist in the market, but the unmanned aerial vehicles are only used as image acquisition equipment in the systems, and the specific image detection still depends on experts of an operation console to identify cracks. Similarly, the crack detection may be erroneously determined due to different influences of image size, illumination conditions, operator experience, and the like in a specific recognition process.
Disclosure of Invention
The invention aims to provide a glass curtain wall crack detection method, which can effectively detect cracks of a glass curtain wall.
The technical scheme adopted by the invention for solving the technical problems is as follows: the method for detecting the cracks of the glass curtain wall comprises the following steps:
step (1): collecting a glass curtain wall image;
step (2): converting the glass curtain wall image into a gray level image;
and (3): calculating a first feature of the grayscale image based on the modified ULBP descriptor;
and (4): reducing the dimension of the first feature of the gray level image to obtain a macro feature;
and (5): and inputting the macro features into a classifier, and detecting whether the glass curtain wall image has cracks or not through the classifier.
The step (3) comprises the following steps:
step (31): dividing the grayscale image into nxn sub-tiles;
step (32): counting a histogram of each sub-tile by using the improved ULBP descriptor;
step (33): and taking the histogram of each sub-tile as a first characteristic of the gray image.
The step (31) is specifically as follows: the grayscale image is divided into 3 x 3 sub-tiles.
The improved ULBP descriptor in the step (3) is specifically as follows: the 58 ULBP codes of the ULBP descriptor are removed from 0 and 255, resulting in the ULBP descriptor with 56 ULBP codes.
The step (4) comprises the following steps:
step (41): extracting a second feature of each sub-image block histogram in the gray level image;
step (42): and connecting the second features of the histograms of the sub-image blocks in series to obtain the macro features of the gray level image.
The second feature of each sub-picture block histogram in said step (41) comprises: mean, standard deviation, skewness, kurtosis, energy, and entropy;
wherein n represents the total number of modified ULBP codes, YiIndicating the frequency of the modified ULBP coding of item i.
The macro feature in the step (4) is 54-dimensional.
The classifier in the step (5) is a support vector machine model based on a 10-fold cross validation sampling method.
Advantageous effects
Due to the adoption of the technical scheme, compared with the prior art, the invention has the following advantages and positive effects: the invention is improved based on an equivalent binary pattern (ULBP), two codes, namely 0 and 255 are removed, the two codes represent bright spots and dark spots without texture change, and no additional discrimination capability is provided, so that the improved LBP descriptor has better identification performance and faster calculation speed; the invention divides the gray level image into 3 x 3 equal blocks, aiming at improving the accuracy of local feature expression and further improving the identification performance; the invention extracts the histogram of each sub-image block, and then extracts the characteristics (namely dimension reduction) of the histogram of each sub-image block to form macro characteristics, thereby ensuring the accelerated calculation and effectively ensuring the identification efficiency; the method can effectively identify whether the glass curtain wall has cracks or not.
Drawings
FIG. 1 is a process flow diagram of an embodiment of the present invention;
FIG. 2 is a schematic illustration of an LBP descriptor principle of an embodiment of the present invention;
FIG. 3 is a schematic representation of a cracked glass panel and its ULBP (56-dimensional) transition in accordance with an embodiment of the present invention;
FIG. 4 is a schematic representation of an uncracked glass panel and its ULBP (56-dimensional) transition in accordance with an embodiment of the present invention;
FIG. 5 is a block diagram of a grayscale image according to an embodiment of the present invention;
fig. 6 is a schematic diagram of another gray scale image block according to an embodiment of the present invention.
Detailed Description
The invention will be further illustrated with reference to the following specific examples. It should be understood that these examples are for illustrative purposes only and are not intended to limit the scope of the present invention. Further, it should be understood that various changes or modifications of the present invention may be made by those skilled in the art after reading the teaching of the present invention, and such equivalents may fall within the scope of the present invention as defined in the appended claims.
The embodiment of the invention relates to a crack detection method for a glass curtain wall, which comprises the steps of firstly deploying an unmanned aerial vehicle with a camera, acquiring a video image of the glass curtain wall through the unmanned aerial vehicle, and extracting features by adopting an improved equivalent local binary pattern (ULBP) texture analysis algorithm to identify cracks, wherein the method is different from a traditional ULBP descriptor, and firstly converting an input image into a gray image and equally dividing the gray image into 3 multiplied by 3 sub-image blocks; further, in the present embodiment, feature extraction is performed on each sub-block by using an improved ULBP, and the extracted features are converted into a histogram (equivalent to a first feature); finally, the present embodiment reads the second features from the histogram obtained from each sub-block, concatenates them into macro-features, and inputs them into the classifier for classification. The present embodiment trains a classifier to identify glass cracks using a Support Vector Machine (SVM) method. Test results show that the crack image can be effectively detected by the method, and the accuracy is as high as 95%.
The invention is further illustrated by the following specific embodiments:
referring to fig. 1, the unmanned aerial vehicle is configured to acquire a video of a glass curtain wall to be inspected, convert a single video frame into a grayscale image, and perform blocking, improved ULBP calculation, histogram construction, feature extraction, and series connection in sequence, and finally classify to predict whether the glass curtain wall has cracks, which is described in detail below:
1. unmanned aerial vehicle for data acquisition
This embodiment adopts commercial big jiang unmanned aerial vehicle, and the model is: DJI phantom pro, the unmanned aerial vehicle is used for video acquisition of the glass curtain wall, and each frame of the video is regarded as an image to be used as the input of the detection method.
2. Data pre-processing
The method proposed by this embodiment uses a modified ULBP descriptor, and the algorithm is adapted to analyze the gray value of a pixel, so that a color image (i.e. each frame of video) needs to be converted into a gray image before the process starts.
3. Extracting special LBP features for crack analysis
Local Binary Pattern (LBP) is a visual texture descriptor that is successfully applied in many computer vision tasks, and thus the use of LBP descriptors in automated crack identification methods is expected to help detect cracks. The LBP descriptor has advantages in its ability to distinguish texture and fast computation, and has gray scale invariance and rotation invariance. The gray-scale invariant characteristic of the LBP descriptor is particularly important for crack identification, because the characteristic can effectively resist the interference of the image under different illumination conditions; and the rotation invariant characteristic of the LBP descriptor can be well suitable for images shot by the unmanned aerial vehicle from different angles.
The LBP descriptor can generate a total of 256 different codes according to the variation of the gray value of each neighborhood pixel. The specific working principle of the LBP descriptor is as follows: by comparing the center pixel value with the neighborhood pixel values one by one (see grey value comparison of pixel values in fig. 2): a binary value of 1 is given if the grey value of the neighborhood pixel is greater than or equal to the grey value of the central pixel, otherwise a binary value of 0 is given. Repeating this comparison process for eight neighborhood pixels of each target pixel produces an 8-bit binary code, which is then converted to a decimal representing the LBP descriptor of the current pixel (the center pixel in fig. 2 corresponds to a decimal 157). In the conventional LBP descriptor, a histogram with 256 groups can be extracted for each test image and used as a feature vector input to the classifier after normalization.
The LBP descriptor can reflect different texture modes and is widely used in a plurality of texture analysis applications, but the characteristics of the LBP descriptor are redundant, so that the calculation speed is slow, and dimension reduction is needed. This embodiment uses a special type of LBP descriptor, called the equivalent local binary pattern (ULBP), which has almost the same recognition capability as the LBP descriptor, but with smaller feature dimensions. The ULBP code is a special subset of LBP codes, which when read in a round-robin fashion (e.g., 11110000 or 11000011) consist of consecutive 0's or 1's, for a total of 58 ULBP codes. According to experimental analysis, when reading the glass curtain wall image, the ULBP descriptor can express about 82% -90% of information, and in addition, the embodiment also deletes two types of ULBP codes, namely 0 and 255 (namely 00000000 and 111111111), wherein the two codes represent bright spots and dark spots without texture change, and do not provide extra discrimination capability, so that the ULBP descriptor modified by the embodiment can be deleted from the feature vector to obtain better efficiency, namely the ULBP descriptor has 56-dimensional features.
Fig. 3 (a) shows a cracked glass panel, and fig. 3 (b) shows a 56-dimensional ULBP converted image of the cracked glass panel; fig. 4 (a) shows an uncracked glass panel, and fig. 4 (b) shows a 56-dimensional ULBP converted image of the uncracked glass panel. As can be seen from fig. 3 and 4, there is a clear difference between the cracked and non-cracked glass panels, where the cracked panel showed a large variation, while the non-cracked panel showed more uniformity. It can be seen that the improved ULBP feature proposed by the present embodiment is very effective for identifying cracks from glass images.
4. Crack identification from glass images
As a common practice, most of the current image analysis work related to ULBP features does not directly use ULBP coding, but constructs a histogram of ULBP coding as a new feature to achieve the purpose of unifying feature vector lengths. In the method proposed in this embodiment, the grayscale image needs to be further divided into 3 × 3 equal sub-tiles (as shown in fig. 5 and fig. 6), in order to improve the accuracy of local feature expression and thus improve the recognition performance.
For each sub-tile in the extracted 3 × 3 block, a histogram (equivalent to the first feature) of its 56 ULBP encodings is computed. The group number of the histogram is 56, and the histogram respectively corresponds to 56 ULBP codes; the vertical axis is the frequency of occurrence of each corresponding code in the analyzed tile. Considering that the analysis of each frame of video (each image) will generate a feature vector with dimensions of 9 × 56 ═ 504 in total, which is easy to cause overfitting of the model; therefore, the embodiment further performs a second feature extraction on the histogram to realize that the original high-dimensional features are described with fewer parameters. Specifically, with YiRepresenting the frequency of the ith item of encoding; n represents the total number of codes, i.e. 56; the calculation methods of the total 6 features (i.e., the second features) extracted by the present embodiment can be expressed as follows:
(1) mean value:
(2) standard deviation:
(3) skewness:
(4) kurtosis:
(5) energy:
(6) entropy:
the six extracted features are sequentially connected from the upper left corner to the lower right corner of the 9 sub-tiles (from left to right, from the top row to the bottom row), and finally a 9 × 6-54-dimensional feature vector is generated. By means of the method and the device, the complexity (dimensionality) of the features can be reduced to the maximum extent on the premise that the feature information is kept as much as possible, and therefore the efficiency and robustness of model training are greatly improved.
Finally, the features proposed by the invention are extracted from different glass panel images acquired by the unmanned aerial vehicle and then used for training a classifier, and the classifier can divide the glass curtain wall images into two types, namely 'cracked' or 'uncracked'. Specifically, the classifier trained by the embodiment is a Support Vector Machine (SVM) model based on a 10-fold cross validation sampling method. The experimental result shows that the highest accuracy of the method provided by the embodiment on the test data set can reach 95%.
Therefore, the method is improved based on an equivalent binary pattern (ULBP), two codes of 0 and 255 are removed, the gray level image is divided into blocks with the same size of 3 multiplied by 3, the histogram of the blocks is extracted, the features (namely dimension reduction) of the histogram of each sub-block are extracted and connected in series to form the macro features, the identification efficiency can be effectively guaranteed while calculation is accelerated, and the crack identification performance of the glass curtain wall can be effectively improved through the improved points.
Claims (8)
1. A glass curtain wall crack detection method is characterized by comprising the following steps:
step (1): collecting a glass curtain wall image;
step (2): converting the glass curtain wall image into a gray level image;
and (3): calculating a first feature of the grayscale image based on the modified ULBP descriptor;
and (4): reducing the dimension of the first feature of the gray level image to obtain a macro feature;
and (5): and inputting the macro features into a classifier, and detecting whether the glass curtain wall image has cracks or not through the classifier.
2. The method for detecting cracks on a glass curtain wall as claimed in claim 1, wherein the step (3) comprises:
step (31): dividing the grayscale image into nxn sub-tiles;
step (32): counting a histogram of each sub-tile by using the improved ULBP descriptor;
step (33): and taking the histogram of each sub-tile as a first characteristic of the gray image.
3. The method for detecting cracks of a glass curtain wall as claimed in claim 2, wherein the step (31) is specifically as follows: the grayscale image is divided into 3 x 3 sub-tiles.
4. The method for detecting cracks of glass curtain walls according to claim 1, wherein the modified ULBP descriptor in the step (3) is specifically: the 58 ULBP codes of the ULBP descriptor are removed from 0 and 255, resulting in the ULBP descriptor with 56 ULBP codes.
5. The method for detecting cracks on a glass curtain wall as claimed in claim 2, wherein the step (4) comprises:
step (41): extracting a second feature of each sub-image block histogram in the gray level image;
step (42): and connecting the second features of the histograms of the sub-image blocks in series to obtain the macro features of the gray level image.
6. The glass curtain wall crack detection method of claim 5 wherein the second characteristic of each sub-block histogram of the step (41) comprises: mean, standard deviation, skewness, kurtosis, energy, and entropy;
wherein n represents the total number of modified ULBP codes, YiIndicating the frequency of the modified ULBP coding of item i.
7. The method for detecting cracks on glass curtain walls according to claim 1, wherein the macro feature in the step (4) is 54-dimensional.
8. The method for detecting cracks on a glass curtain wall as claimed in claim 1, wherein the classifier in the step (5) is a support vector machine model based on a 10-fold cross validation sampling method.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202111612351.4A CN114359195A (en) | 2021-12-27 | 2021-12-27 | Glass curtain wall crack detection method |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202111612351.4A CN114359195A (en) | 2021-12-27 | 2021-12-27 | Glass curtain wall crack detection method |
Publications (1)
Publication Number | Publication Date |
---|---|
CN114359195A true CN114359195A (en) | 2022-04-15 |
Family
ID=81101514
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN202111612351.4A Pending CN114359195A (en) | 2021-12-27 | 2021-12-27 | Glass curtain wall crack detection method |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN114359195A (en) |
Cited By (1)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN114881961A (en) * | 2022-04-29 | 2022-08-09 | 江苏仙岳材料科技有限公司 | Glass fiber plate crack detection method and device |
-
2021
- 2021-12-27 CN CN202111612351.4A patent/CN114359195A/en active Pending
Cited By (1)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN114881961A (en) * | 2022-04-29 | 2022-08-09 | 江苏仙岳材料科技有限公司 | Glass fiber plate crack detection method and device |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN103955660B (en) | Method for recognizing batch two-dimension code images | |
CN114998852A (en) | Intelligent detection method for road pavement diseases based on deep learning | |
CN111339883A (en) | Method for identifying and detecting abnormal behaviors in transformer substation based on artificial intelligence in complex scene | |
CN110335233B (en) | Highway guardrail plate defect detection system and method based on image processing technology | |
CN111539330B (en) | Transformer substation digital display instrument identification method based on double-SVM multi-classifier | |
CN108009574B (en) | Track fastener detection method | |
CN111292321A (en) | Method for identifying defect image of insulator of power transmission line | |
CN111091544A (en) | Method for detecting breakage fault of side integrated framework of railway wagon bogie | |
CN112085651B (en) | Automatic shock wave detection and tracking algorithm based on image self-adaptive threshold and feature extraction | |
CN115995056A (en) | Automatic bridge disease identification method based on deep learning | |
CN112329656A (en) | Feature extraction method for human action key frame in video stream | |
CN111259736B (en) | Real-time pedestrian detection method based on deep learning in complex environment | |
CN111767874A (en) | Pavement disease detection method based on deep learning | |
CN103413131A (en) | Tower crane recognition method based on spectral and geometric characteristics | |
CN115841488A (en) | Hole checking method of PCB (printed Circuit Board) based on computer vision | |
CN114359195A (en) | Glass curtain wall crack detection method | |
CN105469099B (en) | Pavement crack detection and identification method based on sparse representation classification | |
CN112116600A (en) | Photovoltaic panel counting method based on image processing | |
CN109657577B (en) | Animal detection method based on entropy and motion offset | |
CN116524410A (en) | Deep learning fusion scene target detection method based on Gaussian mixture model | |
CN111832337A (en) | License plate recognition method and device | |
CN116051539A (en) | Diagnosis method for heating fault of power transformation equipment | |
Guo et al. | Fault diagnosis of power equipment based on infrared image analysis | |
CN115690584A (en) | SSD (solid State drive) -based improved power distribution room foreign matter detection method | |
CN115205747A (en) | Automatic road detection system based on image recognition |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
PB01 | Publication | ||
PB01 | Publication | ||
SE01 | Entry into force of request for substantive examination | ||
SE01 | Entry into force of request for substantive examination | ||
CB02 | Change of applicant information |
Address after: 201114 room 101-8, building 3, No. 2388, Chenhang Road, Minhang District, Shanghai Applicant after: Shiwei Xinzhi medical technology (Shanghai) Co.,Ltd. Address before: 1628 suzhao Road, Minhang District, Shanghai 201112 Applicant before: Shiwei Xinzhi medical technology (Shanghai) Co.,Ltd. |
|
CB02 | Change of applicant information |