Student athletes often have a hard time affording the time and money it requires to attend classes in order to maintain their skill. TRACKer aims to solve this issue. TRACKer uses CMU's open pose to compare an individual's form with that of someone who they aspire to be like (a pro, for example). We built TRACKer using tensorflow-gpu and NVIDIA's CUDA and CUDnn driver libraries with a deep neural net based off CMU’s openpose for accelerated pose estimation on frames of a video. The most difficult part of the project was getting the flask call for python to work, as transferring the results from a python file to our website was a severe headache. We are proud of successfully implementing not only flask but also CMU's openpose, the latter of which was only possible due to prior experience in tensorflow-gpu use. We learned about the various different ways of storing data on a website and the limitless applications of tensorflow. In the future, we hope to host TRACKer on a public site and share our product to the masses, as it is currently limited only to a local host.

Built With

Share this project:
×

Updates