The problem Workplus solves

Fitness is one of the most cautiously monitored aspects of life and in these modern times of heavy workloads and Pandemic driven laziness the fitness aspect is almost neglected. Hence we have introduced WorkPlus, an AI-powered fitness application that motivates people to exercise daily. People can set their fitness goals and the application prompts the user to complete the target each day and also suggests new workout plans. The main focus of the app is to help people do the workout properly and maintain the perfect posture during it since not maintaining proper posture can strain several parts of the body and will decrease the efficiency of the workout. to achieve this we use the smartphone camera and AI to analyze the body postures in real-time and tell the user to correct the posture if the accuracy is poor. We have also integrated a tool to count the number of times the person has done the exercise. the counter counts the movement using reference points of the body and the screen.
This application provides a complete package for fitness tracking and analysis and we intend to further develop this application and add more relevant features in the future.

Challenges we ran into

One of the main difficulties we faced was to integrate TFlite with flutter and we faced quite a lot of errors in this process, but eventually, we managed to solve the issue by using better documentation of the process. Another error we faced was the accuracy of the model. It was terribly poor and laggy so in order to counter this, we focused on extracting only key coordinates, necessary for the particular exercise, instead of retrieving all the body coordinates. This drastically improved the speed of computation and also increased the simplicity. Hence the accuracy issue was solved. the last error we faced was in the counter because we had no viable way of making the counter not go haywire and count everything just by slight changes. So to solve this issue we used reference points in the screen to help us track any ascend or descend in the image with respect to the reference point. These points were found by trial and error basis and works pretty well for all types of body.