SecureEscape Pitch - UCL First Response in a Box Design Hackathon 2023
0SecureEscape Pitch - UCL First Response in a Box Design Hackathon 2023
The model we are using will work with publicly available datasets such as geospatial data and satellite imagery. Alongside these, users will upload pictures of wildfires and their location, and we will mine social media to spot wildfires as quickly as possible (for instance by using the map features available in Snapchat and Instagram. The model will then use an offline k.Lab model for fire detection, as seen in Marquez Torres et al.'s 2023 paper on wildfire propagation in Italy. It uses parameters from watsonx.ai's geoclassificaiton data, which will take into account the area affected, state of nature in the area and propagation of the fire.
The downloaded app will have a trained VertexAI Online Model which will be trained on images of a wildfire, both from the perspective of a pedestrian and a vehicle. This will be trained in a supervised environment, which allows the user to then add their own inputs to tag dangerous situations. There will be colour histograms to classify the fire's size, and optical flow for direction.
The final output of the app will be to give a itinerary to a safe space