Computer vision image processing for finding missing persons
This paper summarizes three computer vision algorithms used to triage and process imagery from small unmanned aerial systems used to search for 21 missing persons in the wake of the 2015 Texas Memorial Day floods. A single 20- minute SUAS flight over less than a mile of the Blanco River produced roughly 800 high-resolution images totaling 1.7GB. The images required manual examination by multiple trained observers as victims were likely to be covered in mud and debris. In order to triage the flood imagery, students through the National Science Foundation Computing for Disasters Research Experience for Undergraduates program at Texas A&M in conjunction with the University of Maryland rapidly created three algorithms. Two algorithms focused on identifying clothing or debris from homes that would have been swept away with victims. One identifies urban debris with straight lines or sharp corners and the other learned to detect bright colors using the Reed-Xiaoli spectral anomaly detection algorithm. A third algorithm used deep convolutional neural network learning to identify debris piles large enough to contain a victim. The algorithms show the value added by computer vision and image processing.
author list (cited authors)