Location
Hilton Hawaiian Village, Honolulu, Hawaii
Event Website
https://hicss.hawaii.edu/
Start Date
3-1-2024 12:00 AM
End Date
6-1-2024 12:00 AM
Description
Information about damaged buildings is crucial in disaster response owing to the risk they pose, including property damage and loss of human lives. However, it is difficult to capture this information rapidly during disaster. This study developed an automatic model to detect buildings damaged by earthquakes from aerial videos. It is composed of multiple object tracking model of buildings, classification model of damage, and decision tree model to output final estimation by each track. This system considers; (1) detection of damaged and collapsed buildings such as pancake collapse of wooden buildings and significant roof damage, (2) input of time-series information to determine the extent of building damage, (3) less annotation labor to train datasets, and (4) effective usage of decision tree nodes for disaster response. The obtained results indicated that the average recall of three classes was 47.9%, average precision was 48.4%, and average F-measure was 45.7%.
Recommended Citation
Fujita, Shono and Hatayama, Michinori, "Damaged Building Detection using Multiple Object Tracking and Decision Tree from Aerial Videos for Disaster Response" (2024). Hawaii International Conference on System Sciences 2024 (HICSS-57). 6.
https://aisel.aisnet.org/hicss-57/dg/disaster_resilience/6
Damaged Building Detection using Multiple Object Tracking and Decision Tree from Aerial Videos for Disaster Response
Hilton Hawaiian Village, Honolulu, Hawaii
Information about damaged buildings is crucial in disaster response owing to the risk they pose, including property damage and loss of human lives. However, it is difficult to capture this information rapidly during disaster. This study developed an automatic model to detect buildings damaged by earthquakes from aerial videos. It is composed of multiple object tracking model of buildings, classification model of damage, and decision tree model to output final estimation by each track. This system considers; (1) detection of damaged and collapsed buildings such as pancake collapse of wooden buildings and significant roof damage, (2) input of time-series information to determine the extent of building damage, (3) less annotation labor to train datasets, and (4) effective usage of decision tree nodes for disaster response. The obtained results indicated that the average recall of three classes was 47.9%, average precision was 48.4%, and average F-measure was 45.7%.
https://aisel.aisnet.org/hicss-57/dg/disaster_resilience/6