From Raw Data to Deployment: Engineering a Machine Learning Solution

Saving Nature partners with local NGOs and scientists to connect fragmented ecosystems. Through land purchases and the construction of nature corridors, the organization seeks to revitalize populations of endangered species that have lost access to their historical habitats and mating grounds to industrialization. To validate the effectiveness of purchased corridors, Saving Nature manually reviews hundreds of hours of trail camera footage of animal activity in the corridors. The primary objective of our project is to reduce the manual time burden it takes to review trail camera footage.

In our project we built a model can adequately filter videos that had significant movement to activate the camera, but did not contain an animal. This eliminates the need for manual review of these videos which will save the organization a significant amount of time. Furthermore, we have built a robust infrastructure that will make it easy to scale and improve the model as they accumulate more data moving forward. Part of this infrastructure is a web application that will help Saving Nature save time ingesting, and format metadata for their internal database. All of these improvements were made in the framework of an application that requires no technical expertise on Saving Natures behalf.