Future Development
1. Exercise Database
At present, we can detect 3 types of exercise for users --- squat, pushup, hip bridge. Compare to some other detection apps which use built-in sensors to count exercise, our project can detect pose more accurately with help of CV library which is well trained by Machine Learning.
However, this project is limited to only these 3 types. In the future, we think it is possible to build another machine learning model to recognize more exercise types and do the counting separately for different types of work. Many ongoing research is focused on pose identification, and we think those techniques can be further developed to solve this problem.
2. Correction System
For different kinds of exercise, we did some researches on most common mistakes and analyzed the specific mistakes to workout their features. We can detect at least one common mistakes for our provided exercise and design different correction system for different types of exercise.
We can detect our proposed mistake very well but fail to detect other common but not defined problems. The main limitation is 2D video detection library we chose. As this detection library uses phone camera to do the detection in 2-dimension, many important data are lost to built a 3D model in real world which may leads to many analytical error. Apart from that, as we analyze motion in a 2D map, this correction system is strictly limited to a certain perspective of view.
3. Interactive Design
At present, we only have limited cities and the contents are also limited as we do not have enough time for searching and settling. So, in the future, we think it is possible to explore more cities and more interesting information for users. Also, there could be a delete function of destinations, which was not made because of the limitation of software.
As our reward system has basic settings mostly, in the future, we may add more attracting things, e.g., different virtual figures to travel with users. We may also insert map tools, giving users better understanding of the positions of cities, also helping them to plan their tour routes and explore interests by themselves, e.g., write or draw by the traveling routes, and share to others or enjoy themselves.
4. App Integration
For this final demo, we only use one computer connected with one mobile device which may limited the number of users. We think here is 2 potential solution to enlarge number of users. One is to build a cloud server which connected with local device on unique user account. For this method, constructing and maintaining a cloud server may be a problem. Another is to integrate detection and correction system directly with a local app. But the computational ability of mobile device needs to be strong enough to run machine learning model. Large storage may also be required.