Flow and Find Pitch - UCL First Response in a Box Design Hackathon 2023
0Flow and Find Pitch - UCL First Response in a Box Design Hackathon 2023
We propose two online AI models and combine them in the Mobilde application:
Online AI Model - Problem 1: Population Displacement
Usefulness: The online AI model for predicting optimal shelter locations post-disaster is valuable for immediate decision-making during crises. It minimizes evacuation time and reduces the risk associated with manual assessments. Predicting post-disaster population displacement enables the creation of spatial distribution maps to identify areas with high displacement, allowing for efficient resource allocation.
Data Needed:
IOM's Displacement Tracking Matrices (DTM): This data source provides information on population displacement patterns and is crucial for predicting where people are relocating.
Historical Data: Past disaster data offers insights into how populations have reacted to similar events in the past and can help improve predictions.
Geospatial Data: Obtained from satellite imagery and GIS sources, geospatial data provides information on terrain, land use, and infrastructure, which is vital for identifying suitable shelter locations.
Call Detail Records (CDR): Communication patterns data help understand how and where people are communicating during a disaster, aiding in tracking population movements.
Demographic Data: Information such as population density and age distribution helps in understanding the demographics of the affected population.
Disaster-Specific Data: Data on flood zones, earthquake-prone areas, and other relevant disaster-specific factors contribute to the model's accuracy.
Data Augmentation: AI techniques generate synthetic samples to enhance the model's robustness and prevent overfitting.
Offline AI Model - Problem 2: Finding Relatives After Disasters
Usefulness: The offline AI model for identifying relatives in disaster scenarios is crucial for reuniting loved ones quickly. It offers a human-centric approach, where individuals can find their family members without disclosing personal information, making it valuable in maintaining privacy during sensitive times. This model is especially useful when internet connectivity is limited.
Data Needed:
Facial Recognition Data: To match individuals' faces, facial recognition data is essential. This can come from databases of disaster victims and possibly passport systems if available.
Demographic Information: Although not explicitly mentioned, demographic information such as name, date of birth, or age may be used when available to assist in matching.
Facial Algorithm Data: The model may require specific facial algorithms trained on the faces of displaced individuals, which can improve the success rate of recognition.
Manual Data Entry: In cases where cooperation for access to passport systems is lacking or internet connectivity is limited, manual data entry is an alternative data source. This would involve creating a database of individuals checked into aid centers, which can be searched using facial recognition.
These AI models address critical challenges in disaster response by optimizing shelter allocation and improving the efficiency of locating loved ones, both of which are essential in the immediate aftermath of disasters.