Computer-Vision Based Health Monitoring of Aging Rural Bridge Infrastructures

Jalen A Garza Author

07/27/2021 Added

12 Plays

Description

Monitoring the Health of Aging bridges using human inspectors and engineers can be expensive and time-consuming. By using a UAV computer system we can capture images, scan for cracks, measure the cracks, and construct a labeled map of cracks in the bridge without any manual inspection methods. This allows for the inspection operation to be faster, more accurate, and cheaper in assessing the health and performance of a bridge.

Searchable Transcript

Search:

[00:00:01.070]Hello, my name is Jalen Garza, and I am a civil engineering
[00:00:04.010]undergraduate student at the university of Texas Rio Grande Valley.
[00:00:07.205]This summer I had the opportunity to conduct research at UNL
[00:00:10.056]and the title of my project is Computer vision base health monitoring
[00:00:13.336]of aging rural bridge infrastructures.
[00:00:18.200]Bridges are a vital piece of civil infrastructure and cracks can cause
[00:00:21.627]significant damage to the lifespan of a bridge.
[00:00:23.962]Bridge inspections that collect crack properties is critical to regularly
[00:00:27.402]maintaining the safety of the bridge
[00:00:29.131]A human inspector process involves trained engineers and inspectors to manually mark
[00:00:33.142]and chart out the bridge so they can assess the damage and health of bridge
[00:00:36.714]but it can be time consuming, expensive and subjective
[00:00:39.553]A UAV will be able to conduct this operation in a faster and safer way and
[00:00:43.102]will lead to reducing the time and costs
[00:00:45.012]In this study, we created and developed a computer vision-based system
[00:00:48.941]that can autonomously monitor aging bridge infrastructures using a UAV.
[00:00:55.103]The methodology will consist of 4 main stages that helps complete our objective
[00:00:59.049]Data Acquisition (collecting and cropping photos)
[00:01:02.062]Crack Identification (models to detect crack)
[00:01:04.928]Crack Assessment (taking measurements of individual cracks)
[00:01:07.872]and Finally creating a visualization Orthomosaic map
[00:01:13.940]Firstly, Data Acquisition is acquiring the images captured from a UAV.
[00:01:17.566]The UAV was flown over a bridge in Omaha, NE collecting around 200 images
[00:01:21.151]that are further cropped into 3,044 smaller image
[00:01:25.630]Two important factors in this step is the height and overlapping percentage.
[00:01:29.308]We needed to have a consistent height that allow for good resolution in a image
[00:01:33.030]while also covering a large enough area, so the height we used was 15 feet.
[00:01:38.690]The pattern the UAV followed is shown in the figure in bottom right
[00:01:41.922]this is done because the overlapping on each image must be in between 60 to 70%
[00:01:46.327]so when stitching map there won’t be any missing areas or stretching causing
[00:01:49.919]unscaled and false calculations
[00:01:55.043]Crack Identification is where the Artificial Neural Network is introduced
[00:01:59.065]Mask R-CNN is a model of deep learning that requires thousands of images
[00:02:02.492]for training data
[00:02:03.512]We acquired over 5,500 images of cracks and non-cracks that were used to help
[00:02:07.263]train the model and predict the cracks. When inputting an image, Mask R-CNN
[00:02:11.744]localizes the searched object for this research in particular
[00:02:14.721]we are trying to detect a crack. Next it creates a bounding box around
[00:02:18.499]the object and finally creates a segmented mask of the crack
[00:02:21.606]For better Assessment and visualization extract the fill and plotted
[00:02:24.793]on a blank white image
[00:02:29.043]The crack assessment is all done using MATLAB code
[00:02:31.430]First, we convert the segmented fill found using Mask R-CNN into a binary code
[00:02:35.161]converting the crack fill into a dataset of 1 and 0s
[00:02:38.728]1 being the fill and 0 the background
[00:02:41.435]We solved for length width and orientation The Length is the sum of all pixels
[00:02:45.144]in the binary skeleton if they are diagonal to each other its
[00:02:48.029]square root of two due to the Pythagorean theorem and if
[00:02:50.749]vertical or horizontal it’s 1. The width is done by finding the Euclidean
[00:02:56.022]distance which is the distance from the skeleton to the nearest edge
[00:03:00.849]multiplied by two, to get the total width at each individual pixel
[00:03:06.835]To find the average width you have to find the summation of each individual
[00:03:11.445]pixel width by the total amount of skeleton pixels
[00:03:15.580]The orientation tells us the type of crack it is. By creating a line
[00:03:18.909]between the endpoints of crack and trying to find the angle between that line
[00:03:22.674]and a horizontal line across the bridge will result in
[00:03:26.441]transverse longitudinal or shear cracks depending on bottom figure and its angle
[00:03:32.190]The orthomaosic map is our final goal of the research
[00:03:35.027]By stitching up the images with the labeled cracks and measurements
[00:03:38.210]we can construct a full-scale bridge that can be assessed and studied
[00:03:41.535]without having to manually record data. The bottom figure shows a sample
[00:03:45.909]of a previous map provided by Dr. Won that we used for our methods
[00:03:49.482]Unfortunately we didn’t have time to
[00:03:51.148]construct the map for our testing bridge
[00:03:53.099]but Dr. Won's map was more than enough (to test our methods)
[00:03:56.473]Some Results from this method is that it detected 266 cracks
[00:03:59.413]from top of bridge deck. The first chart shows the variation of cracks
[00:04:02.780]width and crack length. The average width of the
[00:04:08.399]our methods was about 0.26 inches wide and average length was 5.53 in long
[00:04:15.112]The second pie chart illustrates the crack types and their respective amounts
[00:04:18.742]transverse was the most common type of crack
[00:04:20.887]which is horizontally across bridge.
[00:04:24.678]To conclude, using a UAV assisted bridge inspection has huge potential in years
[00:04:28.987]to come with civil infrastructure and technology growing
[00:04:32.199]Using this method can help aid inspectors asses the health
[00:04:35.009]of the bridge by providing various crack properties with a more efficient,
[00:04:38.545]accurate and safer inspection methods
[00:04:41.453]Some future work in expanding this method
[00:04:43.479]is trying to find and detect other defects such as spall and corrosion
[00:04:47.703]also repeating the process every couple of years to see how much the cracks
[00:04:52.477]develop over time to indicate the performance of the bridge over time
[00:04:59.788]I’d like to thank the funding provided by National science foundation, I also
[00:05:03.908]thank university of Nebraska Lincoln for the opportunity for this summer
[00:05:07.387]research program, and my research advisors and mentors.
[00:05:10.877]Thank you

The screen size you are trying to search captions on is too small!

You can always jump over to MediaHub and check it out there.

Comments

0 Comments

Computer-Vision Based Health Monitoring of Aging Rural Bridge Infrastructures

Description

Searchable Transcript

Comments icon comment

Related Channels

Comments