Based on the theme of beaches, we thought wouldn't it be better to know before hand how crowded the beaches are for surfing? We thought we can use the surf cams available on the most popular beaches and tell people the number of people surfing right now.
What it does
It streams live video of famous beaches(currently only 1) and tells the user how many people are there on the beach currently.
How we built it
We used Flask to handle the backend and HTML CSS frontend. For predicting the number of people in the video we extract a frame for the video and feed it through an onnx model of CSRnet, which is a Deep Learning algorithm to predict the number of people in a crowd.
Challenges we ran into
Finding a model to predict accurately and replicating it's result is very hard. Initially the model was giving very bad results and later it turned out that the pre processing code was not working properly.
Accomplishments that we're proud of
We trained the CSRnet model on the Shanghaitech dataset and exported the model in onnx format to run it on the cloud. It's pretty cool to achieve.
What we learned
The thing that we learned is that we should manage everything pretty well and work must be done in a team. Due to timezone differences, coordinating the work was particularly hard and we couldn't do what we hoped to achieve with this project.
What's next for Surf Crowd Estimator
We will add more beaches to the webapp, currently it supports only 1 beach.
PS. Our project is pretty much incomplete at this point. We couldn't solve 2 problems due to time constraints and health constraints. 1 is to extract the frames from live video and the other is to host our server.