Route Scout

• 

  We tested the app on Purdue's very own rock wall.

• 

  The model can recognize holds of all shapes and sizes.

• 

  The model still performs well in conditions with harsh lighting.

• 

  The model still performs well in conditions with harsh lighting.

• 

  The model ignores chalk imprints left on the bare wall.

Inspiration

As freshmen, we've become surprisingly passionate about rock climbing in our short time at Purdue. We realized that color blindness can diminish the experience for some climbers, and we aimed to fix that.

What it does

Route Scout uses machine learning to analyze a photo of a rock climbing wall and highlight the different routes that are available on the wall. The app requires an internet connection to access the machine learning model. Our primary goal for the app is for people who are color blind and have trouble distinguishing between the different climbs that are available. While this app was made with color-blind individuals in mind, it can also be extremely helpful for beginner rock climbers as routes can sometimes be hard to tell apart. Currently, the app allows the user to take a photo and instantly label the colors of the different holds.

How we built it

We initially focused on getting a dataset for our model. We found that there was not a lot of data that met our specifications, so we spent some time collecting, labeling, and preprocessing the data. We then trained an object detection model using this data. With the app, we used the Flutter camera package and HTTP package to get photos and run inferences on the photos. We then had to calibrate the labels and bounding boxes to get them to provide a visually appealing experience for the user.

Challenges we ran into

We originally tried to create a PyTorch model so that inference could be run locally, but the integration into Flutter was not as robust compared to the cloud-based model provided by Roboflow. We also had issues developing a user-friendly app, as our limited experience in Flutter combined with its unique widget-based syntax led to countless obstacles during development.

Accomplishments that we're proud of

This was our first attempt at app development, and the first hackathon for half of our team. We are proud of the fact that we made a functional app, let alone one that achieves its goal so well.

What we learned

As our team is composed of all freshmen, this was a huge learning opportunity for us, and for some of us, it was our first-ever hackathon. A lot of what we did was new to all of us. This includes using a machine learning model and creating an app in Flutter.

What's next for Route Scout

While what we achieved during BoilerMake was our initial goal, we do have ideas of what we can improve. For example, we could implement a way to select which route to highlight in order to decrease the visual clutter. In this way, we can decrease potential confusion and clarify routes for increased climbing efficiency. Another approach to improving the application would be to improve the user experience as a whole. As of right now, it is quite barebones but future adaptations would include: a menu system, a way to track past route pictures, a way to upload a previously taken photo, and much more. We would also like to build a local model so the app can be used when offline as well.

Built With

• flutter
• roboflow