Our inspiration stemmed from the study of how important fitness is to boost immunity during these times. When the coronavirus was declared a pandemic and people started self isolating to stay safe, an equally important issue was brought to light. With gyms and fitness centers closed, it became all the more crucial to promote workout from home. If working from home can be the new normal so can workout from home. We wanted to give trainers a chance to train and continue working during these times. A lot of trainers have lost jobs or are under income freeze and this platform will help them build a reputation with users personally. The users on the other hand get a chance to keep a regular workout routine by indulging an almost real-gym like virtual experience with added performance analytics and weekly reports.

What it does is a virtual fitness platform that integrates trainers to go live, record fun, interactive fitness sessions of different varieties and build a reputation for themselves and their fitness centers. The users can register for 2 free, live sessions per day. They will be asked to give access to their cameras and all their movements are recorded, compared to the trainer's movements and a similarity score is generated between 0 and 1 (1 being exactly similar) and further analytics are computed to give users a feel of their performances, motivation to attend another session, competition between friends and trying different categories of different levels.

How we built it

We used Tensorflow lite and posenet to build it. PoseNet is a vision model that can be used to estimate the pose of a person in an image or video by estimating where key body joints are. Pose estimation refers to computer vision techniques that detect human figures in images and videos, so that one could determine, for example, where someone’s elbow shows up in an image. The key points detected are indexed by "Part ID", with a confidence score between 0.0 and 1.0, 1.0 being the highest also called as Pose Match Score. It uses TensorFlow.js along with PoseNet Libraries to provide the confidence score for not only single person pose but also for multiple person in a single frame be it trainers or users.

Challenges we ran into

Building a pose detection algorithm was a real challenge because such models have been exclusive research papers spanning over months. We used tensorflow lite and a prebuilt posenet algorithm to test on the fitness videos. This was a new technology for us. Getting sensible similarity scores and building analytical computations was another challenge.

Accomplishments that we're proud of

Our idea was our biggest strength in inspiring us to work on this app. Our commitment to the new normal and to help fitness centers recover from losses. Our pose detection algorithm is another pioneer for in the virtual experiences domain that helps quantify workouts, something not feasible at gyms. We also are proud of our dashboard of analytics which is indeed our selling point.

What we learned

We learned the posenet architecture and the tensorflow lite platform and the usage to build our customized pose detection algorithms. We also studied the different fitness apps similar to ours and their business plans. We learned how to write our own unique business plan that makes our service stand out. We also learned how to write different computations to generate scores, reports and analysis of our users' performances.

What's next for

Custom diet plans

  1. Basic diet charts with calorie counts for different meals and suggestions based on preferences in the free version
  2. Custom diet plans generated using AI and integration of a chatbot to discuss any queries in the premium version Community Building:
  3. Building a community of bloggers, fitness enthusiasts and gym freaks to share their performance reports, diet ideas, recipes, articles and images
  4. Rate, like and comment on various fitness center profiles and each workout session so users can look at different categories and decide on what session they'd attend today

Built With

Share this project: