PoachNot

Login Page for Monitoring Dashboard
Options
Poacher Detected
Poacher Not Detected

Saving animals, one step at a time

Inspiration

Over the course of our lifetimes, we've seen many species go from being at-risk to being endangered to becoming extinct because of poaching. Rangers simply cannot patrol the large reserves under their jurisdiction to prevent poachers from hunting these endangered animals.

What it does

PoachNot uses microphones set up at multiple points in a wildlife reserve to monitor the area for signs of poaching activity. After hearing gunshot sounds, it uses the microphones to pinpoint the location of the poacher and alerts the authorities so that the poacher can be caught.

How we built it

We built a dashboard on React which communicates with a Python backend using FastAPI. Our frontend polls the backend to detect whether there are gunshots nearby. We trained a model on PyTorch using data labelled on Scale's Rapid platform which we use to distinguish between gunshots and other sounds. We constructed our features from each audio file by constructing a spectrogram of the sound signal. We also trained a model on Mage which ascribes a probability of a poacher being present in a given location given historic data, in order to ascertain we have not reached a false-positive.

Challenges we ran into

We had an authentication platform that could allow correct access to the right people to use the platform. We ended up using Docker and Redis as our database holders to keep track of all this information. We faced some difficulties setting up our Docker environment and some of our python modules would not be recognized despite having been included in the requirements.txt and installed in the environment manually.

Accomplishments that we're proud of

We're proud of the accuracy we achieved with the limited amount of data we had in distinguishing between gunshots and other sounds. We achieved an accuracy of over 95%.

What we learned

On the technical side, we learned about the plethora of available platforms that allow us to perform our application development much more efficiently, including ScaleAI and Mage's API integrations. In a broader sense, we learned how an idea can be used for a variety of applications, as described in the next section.

What's next for PoachNot

Detecting gunshots and identifying the relevant authorities in real-time is not unique to poaching. It has many other applications in urban society. For example, Berkeley had a shooting scare on Friday which resulted in many people getting hurt. If the police had been notified immediately, they would have been able to proactively prevent people from panicking. Additionally, alarms could be installed in public spaces that respond to a shooting alert, this way people nearby the scene of incidence can be notified about the danger without the need for someone to manually report the incident once they reach a safe space. Moreover, the authenticity of a real gunshot vs a mock noise can be distinguished using our platform.