Labelling Point Clouds for Damage Assesment of Residential Areas in the Aftermath of Nashville Tornadoes

Pooja Rajeev Author

07/28/2020 Added

10 Plays

Description

A tornado event occurred in Nashville which destroyed several areas. We have been working on some of the datasets collected from this area affected due to tornado and determine the amount of damage distribution and study about some of the structures which sustained during the tornadic winds.

Searchable Transcript

Search:

[00:00:00.859]Hi everyone, my name is Pooja Rajeev
[00:00:02.949]and today I am presenting
[00:00:04.193]my poster which will discuss on
[00:00:05.709]labelling point clouds for damage
[00:00:07.509]assessment of residential areas
[00:00:09.159]in the aftermath of Nashville tornadoes.
[00:00:14.943]We all know that a tornado can
[00:00:16.633]cause several damage to different
[00:00:18.203]areas. There was a tornado outbreak
[00:00:20.233]in Nashville on March 3, 2020.
[00:00:22.182]The maximum estimated wind speed of
[00:00:24.272]this tornado was 175 miles per hour.
[00:00:26.872]And our exact site location is
[00:00:28.772]east of Nashville, near the
[00:00:30.334]community of Mt Juliet, Tennessee.
[00:00:32.514]And here Figure 1 shows a preliminary
[00:00:34.754]tornado path. So in this figure,
[00:00:38.244]you could see a green line which
[00:00:40.184]represents the tornado path and
[00:00:42.144]the red sign shows our exact site
[00:00:44.114]location. This research will also
[00:00:47.084]be helpful to determine the capacity
[00:00:49.254]of buildings at a certain wind speed
[00:00:51.304]and also will be further used to
[00:00:53.254]guide emergency responses and other
[00:00:55.514]recovery operations. Our main
[00:01:00.584]objective of this study is to
[00:01:02.214]determine the amount of damage
[00:01:03.944]destruction caused to this area
[00:01:05.754]due to tornadoes. Furthermore, it
[00:01:07.964]includes studying which structures
[00:01:09.834]had sustained damage due to tornadic
[00:01:11.764]winds, in order to identify damage
[00:01:13.814]patterns and design shortcomings.
[00:01:15.964]Also, we need to identify what
[00:01:19.064]type of structures remain undamaged
[00:01:21.264]at a certain wind speed. Here,
[00:01:23.134]Figure 2 shows a totally destroyed
[00:01:24.964]area aftermath tornado.
[00:01:27.829]The methods we used for data
[00:01:29.429]collection was through a UAS device
[00:01:31.535]with an onboard camera, which was
[00:01:33.409]previously collected. This device is
[00:01:35.519]shown in Figure 3. This image shows
[00:01:39.325]the device which we used to collect
[00:01:41.555]all of our data and this is called
[00:01:43.305]DJI Mavic Pro2. After collecting
[00:01:46.525]all the images using the UAS device,
[00:01:48.945]this data was later processed using
[00:01:51.005]Pix4D software, which then created
[00:01:53.145]a 3D point cloud and then we used
[00:01:55.345]CloudCompare software for post-
[00:01:57.455]processing the 3D point clouds.
[00:01:59.435]And figure 4 represents CloudCompare
[00:02:01.445]software. So in this figure, you
[00:02:04.021]could see the screenshot of
[00:02:05.551]CloudCompare software, where this
[00:02:07.841]is a 3D point cloud which was
[00:02:09.681]processed using the Pix4D software
[00:02:11.861]and we use this dataset to tag
[00:02:13.793]and classify the objects into
[00:02:15.503]14 different categories such as
[00:02:17.303]the structures and general objects.
[00:02:20.313]So these are two of the datasets
[00:02:21.883]which we have been working on.
[00:02:23.493]And theses are represented by
[00:02:24.913]Figures 5 and 6. Here, Figure 5
[00:02:28.623]which is the dataset of Mt Juliet
[00:02:30.263]Housing(Triple Crown) and Figure 6
[00:02:32.493]shows the West Lebanon Housing
[00:02:34.133](Stone Bridge). And from all the
[00:02:35.673]data classification and tagging,
[00:02:37.263]we could say that the West Lebanon
[00:02:39.293]Housing had sustained more damage
[00:02:41.343]compared to Triple Crown
[00:02:42.583]neighborhood. From these two datasets,
[00:02:46.173]each dataset has up to 14 different
[00:02:48.213]categories where we divide them
[00:02:50.113]mainly into two different groups:
[00:02:51.793]structures and general objects.
[00:02:53.773]These are some of the instance
[00:02:55.353]examples of the structures, and
[00:02:57.713]there are 5 classes of these
[00:02:59.083]structures, where we have the
[00:03:00.473]damaged structure which has more
[00:03:02.043]than 15% of roof structure failure,
[00:03:04.516]severe structure which has more
[00:03:06.406]than 50% of roof cover, moderate
[00:03:08.766]structure have between 15% and
[00:03:11.116]50% of roof cover, minor damage
[00:03:13.837]has less than 15% of roof cover
[00:03:16.476]and then there is the no damage
[00:03:18.536]structure where there is no exterior
[00:03:20.266]damage. These are some of the
[00:03:23.956]instance examples of general objects,
[00:03:25.996]where we have debris, roadway
[00:03:28.476]vehicles, water bodies, poles
[00:03:30.464]terrain, fallen trees and trees.
[00:03:33.914]After tagging and classifying the
[00:03:35.494]whole data, we got the results
[00:03:37.104]for Mt Juliet Housing(Triple Crown)
[00:03:39.114]dataset and this is represented
[00:03:40.722]by Figure 7. In this picture, you
[00:03:43.882]could see the results of the tagged
[00:03:46.052]dataset of Mt Juliet Housing
[00:03:47.522](Triple Crown) neighborhood
[00:03:48.982]and this is how it looks like after
[00:03:50.842]processing all the different
[00:03:52.982]objects into different categories
[00:03:54.582]and here we have all the structures,
[00:03:56.892]where red represents destroyed,
[00:03:59.262]orange for severe, yellow for
[00:04:01.232]moderate, green for minor and blue
[00:04:03.762]for undamaged. And in case of the
[00:04:06.902]general objects, here we have the
[00:04:09.473]black, it represents the debris
[00:04:12.053]and these are all the roadway
[00:04:14.073]and green represents the trees
[00:04:15.893]and brown is for fallen trees
[00:04:18.523]and the blue all around here
[00:04:20.143]represents the terrain. Also, from
[00:04:22.733]this dataset, from the results of
[00:04:24.333]this dataset classification, we
[00:04:26.463]could clearly say that most of the
[00:04:28.563]Northern part had sustained the
[00:04:30.914]damage compared to Southern part
[00:04:32.563]of this dataset. From this study,
[00:04:35.613]we could conclude by saying that
[00:04:37.390]most of the damage occurred at
[00:04:38.860]the Northern part due to roof
[00:04:40.610]structure failure. So in future,
[00:04:42.350]if new buildings are constructed
[00:04:43.920]with higher performance, in order
[00:04:45.520]to survive wind storm, that could
[00:04:47.100]produce less damage and less debris
[00:04:48.970]during future natural hazard events.
[00:04:53.099]For future work, this whole
[00:04:54.779]tagged data would be carried out
[00:04:56.849]through 3D Convolutional Neutral
[00:04:58.699]Network(3DCNN) to determine the
[00:05:00.879]performance of structures and
[00:05:02.439]determine the amount of damage
[00:05:04.119]distribution in the area caused
[00:05:05.899]by tornado. I would like to thank
[00:05:10.508]Undergraduate Creative Activities
[00:05:12.308]and Research Experience(UCARE)
[00:05:14.268]program and data collected by
[00:05:15.888]StEER team for funding my work
[00:05:19.708]and for supporting with data
[00:05:21.078]collection, respectively. This
[00:05:22.518]work was completed utilizing
[00:05:24.898]the Holland Computing Center of
[00:05:26.908]University of Nebraska, which receives
[00:05:28.738]support from Nebraska Research Initiative.
[00:05:31.938]Thanks for listening!

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

Labelling Point Clouds for Damage Assesment of Residential Areas in the Aftermath of Nashville Tornadoes

Description

Searchable Transcript

Comments icon comment

Related Channels

Comments