Detect an isolated camera zoom

Inspiration

Better sports analytics through computer vision and to improve sports coverage which helps in better customer engagement.

What it does

This isolated camera zoom follows one or more players and/or coaches in the frame and ends as soon as there is a camera change. Isolated frames will always have a box score graphic and at times game flow graphics that may include the following: seasonal averages, upcoming schedule, and additional promotions for future game events.

How we built it

Started with a pre-trained Inception V3 model and trained on our dataset on AWS SageMaker.

Challenges we ran into

Loading the initial dataset into the instance and creating a balanced dataset for all the classes.

Accomplishments that we're proud of

Built a full-scale CNN model and deployed it on the cloud and achieve test accuracy close to 70%.

What we learned

Usage of AWS SageMaker for training and deploying our Machine Learning models and to do transfer learning using a pre-trained model.

What's next for Detect an isolated camera zoom

Collect and augment more data and improve the test accuracy and train it on all actions to create a generalized model that can detect any type of action.

Built With

• amazon-web-services
• cnn
• inception-v3
• python
• sagemaker