The problem poseAI solves

poseAI is your personal AI-powered Yoga instructor, allowing you to harness the benefits of yoga from any place you want.

poseAI uses computer vision and a convolutional neural network to give you the ultimate yoga experience! It recognizes your voice and accordingly guides you through the yoga asana pose you want to perform.

Its time for people from all walks of life to start paying attention to physical as well as mental fitness.

In this hectic lifestyle, we all get so busy with our daily schedule & this is where we always forget to take care of ourselves. It is important for us to take care of our body and mind as well. And this is where yoga comes into play.

Practitioners tout yoga for its mind-body benefits—flexibility, toned muscles, reduced stress, among others. More recently, scientists have begun to test yoga's effect on serious medical conditions. The results have been impressive enough that investigators expect yoga will soon become part of the standard treatment for a number of disorders.

poseAI will make sure that from fitness experts to railway workers, who are exceptionally busy, do not have to travel to practice yoga. They can do so at any convenient place.

• Yoga can significantly improve mood and reduce the symptoms of depression and anxiety.
• Women who do yoga during and after treatment of breast cancer experience less physical discomfort and stress.
• Yoga can lower blood pressure, cholesterol, and resting heart rates, and help slow the progression of atherosclerosis—all risk factors for heart disease.
• Menopausal women who took two months of a weekly restorative yoga class, reported a 30 percent decrease in hot flashes.
• Yoga during pregnancy have found it is also effective for reducing depression, particularly when it is started early in the pregnancy and can be combined with standard depression treatment.

Challenges we ran into

Getting a large amount of data and cleaning is was one of the challenges we faced. We used Beautiful Soup to scrap image links out of XML and HTML files. Beautiful soup is a library in python.

Making use of deep learning required knowledge on the pose estimation. We initially tried with open pose and then started with the posenet model. The posenet model recognises the joints of a human and connects them in the form of a stick figure. Making use of this stick figure to train our model was really a challenge. We finally decide to go for a sequentitial model and added several layers to make it work.

Getting the model to be accurate was really hard. We kept adding new images to the training data as well as test data. Our model initially had the prblem of over fitting, but by using a cross validation set, we reduces the overfitting considerably to improve the model's accuracy.

Docker was a new service to get started with. As we kept working, it got clearer. We faced challenges along the way, mostly error in the system regarding our file formats and everything, but eventually it worked.

Hosting the docker container on an AWS ec2 instance, we ran into issues and errors at first. We had to install TensorFlow in cache free dir to make sure that the memory limit did not exceed. After a lot of effort, we succeeded in getting all the requirements installed into the container before hosting it on AWS.

We also faced certain challenges while sending each camera frames to the Deep Learning model using Flutter. It being a new framework does have good plugin support for an app like this, where live camera frames are to be captured and sent to AWS and receive the output accordingly.