Gym Companion uses surveillance cameras installed at the gym to help people track their workouts automatically. It consists of several subsystems as follows:
- Tracking People: Gym count utilizes existing deep learning-based human detection models to track the people in the gym.
- Exercise Recognition: It uses a machine learning model using sparse Spatio-temporal features to recognize the type of exercise. As a proof of concept, the current model is trained to recognize four exercises i.e. squats, dumbbell, running on a treadmill, and using an elliptical machine.
- Exercise Counting: As most of the gym exercises are periodic, so this problem reduces down to counting the number of repetitions in periodic motion. We built upon the existing research works using deep convolutional networks for counting repetitions in periodic motion. For non-periodic exercises such as running on a treadmill, we record the time duration.
More details can be found in the attached report.
