Wildlife crime has reached a crisis point and we want to bring a solution to this problem. Unmanned aerial vehicle (UAV) provide a cost-effective option to monitor wildlife and poaching. We want to develop an system that helps detecting poachers in aerial images and save animals from extincton.
What it does
The Syndicate algorithm detects in aerial images living objects and distinguishes between animals and humans. To do so we applied a transfer learning approach with a pre trained yolo network. We decided to use yolo instead of rcnn due to the faster inference time. Hence allowing real time classification of poachers. The trained model is accessible by our flask backend. Which allows real time inference for a given frame. Additionally we mocked a frontend UI which shows possible features our project would be able to provide.
How I built it
Our project consists of 3 main components:
- frontend: mocked angular ui for users Demo
- backend: flask server. Provides rest-api access to inference our model
- python scripts: scripts for converting the dataset into a yolo compatible format as well as train the real model.
Challenges I ran into
our biggest issue during the hackathon was the limited time. Since data preparation and model training took a vast amount of our time, we weren't able to finish everything and focused only on specific parts of our project.
What's next for Syndicate
We want to reduce wildlife crime and continue developing our algorithm. Our goal is to ensure that it is used in real life. As the next step we want to connect our different components into a functioning system. Additionally we would like to optimized our trained model with a post processing sequence modelling task. As a further step we would like to test whether our system runs on UAV.