Swamp Vision

• 

  Key deer

• 

  Manatee

• 

  Alligator

• 

  Panther

• 

  Turtle

Team leader discord ID- Atri#2832

Inspiration:

Our inspiration for the project comes from the idea of doing something for the endangered animal species of Florida that helps to conserve them. Florida is home to quite some species native to it, and many of them face the danger of extinction. Some of those species are Florida panthers, Key deers, Manatees, Sea turtles, Alligators, and a few more. We wanted to make a computer vision model that can detect the sightings of these animals wherever they are found so that it can be used to take conservatory measures to save them.

What it does:

Our trained model can be used to detect 5 endangered species (for now), Florida panthers, Alligators, Manatees, Key deers, and Florida sea turtles. It takes images or video footage as its input and finds if there are any of the 5 species in it and detects them with a rectangular box around it and saves the output.

How we built it:

We used YOLO version 5 architecture for training our object model. YOLO is the most popular algorithm for object detection so it was a natural choice. We had to collect the dataset to train our model and since we didn't find a public dataset for our requirements we collected our own database of close to 1000 images and annotated them. We used this dataset to train our model in Google Colab GPU and were able to get good accuracy. Other technologies used in this project are Roboflow, Python, Pytorch, OpenCV, Numpy, and a few others mentioned in the requirements file in our repo.

Challenges we ran into:

Finding a good dataset was a huge challenge so we had to make our own dataset and make annotation labeling ourselves. Finding a GPU to train the model was also challenging as we ran into a lot of problems but were able to get everything done with Colab.

Accomplishments that we're proud of:

That we were able to get good accuracy and confidence scores on the output and we are trying to address a problem that Florida faces which is the extinction of its endangered species.

What we learned;

We learned a lot about Computer Vision and Object detection models. Using Git for collaboratively working together virtually and about using free GPU resources. How to build good datasets for Computer vision models. Finally, some good teamwork lessons.

What's next for Swamp Vision:

It has a lot of scope. The data obtained as a result can be used to create a database and store information about animal sightings, their location , time and the count of animals and use these details to take conservatory measures by lets say the wildlife conservation authorities. This is something we couldn't add in our project now due to time crunch. We can also have web application that the authorities can use to maintain and access this data and also some educational content can be made public for the users on the same website.

Built With

• git
• google-colab
• matplotlib
• numpy
• opencv
• python
• pytoch
• roboflow
• yolo