This Rapid Response Research (RAPID) project from Rochester Institute of Technology (RIT) seeks to develop and evaluate an automated target detection tool for disaster response. In the aftermath of the January 12, 2010 earthquake in Haiti and in accordance with the mission of the Information Products Laboratory for Emergency Response (IPLER) (Partnerships for Innovation (PFI) Award 0917839) to achieve innovation in disaster management, the objective of the development of this tool is to provide a mechanism that will assist emergency response personnel in quickly identifying desired targets in a large collection of imagery. The intent is for the tool to run on commonly available systems (supporting both desktops as well as laptops that might be used in the field), needing no more than recently collect imager to process. A user will examine one typical image and identify for the tool a particular target based on simple matches of color and tolerance properties. Using standard and well-established supervised image classification techniques, this tool will then proceed to locate similar targets in all other images across a large area, as directed by the user. The work will address the important question of how to provide useful information to onsite responders to disasters and emergencies, when the specific questions and the type of information that is useful are dependent upon local events and circumstances.
A prototype of the algorithm has already shown that it can locate encampments of internally displaced persons (IDPs) in the Haiti disaster environment using imagery from the RIT Wildfire Airborne Sensing Program (WASP),an airborne imaging system, utilized during seven days of flyovers (January 21-27, 2010) of the earthquake ravaged areas for Haiti funded by the World Bank. In the context of the 1TB of data collected during the Haiti flyovers, the project will be able construct an end-user tool and evaluate its usefulness, and, by proxy, the usefulness of similar analytical tools in an operational environment. In the near term, this automated tool can be provided to responders on the ground so that food, water, and medicine can be brought to the internally displaced persons (IDPs), thus contributing to meeting the urgent needs onsite in Haiti.
The basis for this project was consideration of the results of a major data collection that was conducted by RIT over Haiti following the January 12, 2010 earthquake. The World Bank had contracted with the RIT Information Products Laboratory for Emergency Response (IPLER) partners ImageCat Inc. and RIT's Laboratory for Imaging Algorithms and Systems to collect high quality aerial multi-spectral imagery and LIDAR information over the affected area to aid in the relief effort. The data collection produced over 1.1 TB of raw imagery data from 15,191 images x 4 cameras = 60,764 total images, and 131 GB of raw LIDAR scan data from over 2.9 Billion LIDAR returns. The major activities of this project were to review the collection, analysis and dissemination process and to consider how such high-value data could be made available to decision-makers and responders in the field. These considerations led to the development of a set of base assumptions and recommendations for the requirements of a system which would have the capability to meet the information needs of responders with the kind of interaction and time-frame that would be important for effective response. A system that has the capability to support a heterogeneous group of responders at a disaster site as well as a network of supervisors and planners from diverse agencies at remote locations is faced with challenging requirements. From the responders perspective, any field system must be easy to use, have modest computing demands, an intuitive and natural user interface that can be used by someone with minimal training, and with timely and useful data available. When data are collected remotely, such as by satellite, aircraft, or by users who are spread out over the landscape, there is a need to collect and organize diverse sources and then provide users with high-quality products. Low bandwidth and intermittent communications are to be expected in disaster arenas. However, the raw data are likely to be gathered in large quantities, with hundreds of gigabytes or even terabytes not unusual. This needs to be gathered organized, analyzed in a manner that is appropriate for different responder missions, and then the essential information products delivered in a timely and useful manner. It is recommended that consideration be given to further research targeted at the issues of widely scattered and intermittent data collection, the need for sophisticated and integrated analysis, and the need to communicate many kinds of information products to a heterogeneous collection of responders who possess only basic communication and display tools. Successful development of a framework for such systems could lead to high payoff in terms of faster and more effective response in confusing disaster situations.
