Salat Guide

Project Title

Salat Guide

Project Area of Specialization Artificial IntelligenceProject Summary

Salat, the Islamic prayer, is an obligation on all Muslims and must be performed five times a day. Vital to Salat is a sequence of poses and gestures that must be performed in a specific sequence. These poses include Takbeerat, Al-Qiyam, Rukku, Qiyam, Sujood, Julus, Tashahhud, and Salam. Often, a person forgets his current state in the sequence and skips some of the steps. It could be due to forgetfulness, lack of concentration, or maybe because the person is new to praying. In this project, using modern Computer Vision techniques, we aim to develop a smartphone-based solution that will guide the person through all the gestures and poses in the right sequence. A worshiper when praying will place his smartphone near him such that the camera will be able to capture his whole body from the ground up. The worshiper can then start his prayer while our app monitors the state of the worshipper in that prayer.

Project Objectives

The objective of the project, as mentioned previously, is to assist worshippers in offering their Salat correctly. Specifically, we aim to produce the following deliverables:

• An android based smartphone app.
• An app that uses its front camera and identifies major poses, such as Al-Qiyam, Ruku, Julus, and Sujud.
• An app that allows the user to choose between different Fard Salats and based on the chosen Salat, guides the worshipper through the sequence of poses.
• An app that provides visual and audio hints of the next pose to perform to the worshipper.
• In case of a mistake, the app should dynamically modify the sequence as required by the Salat.

Further objectives:

• An app that further classifies gestures, such as Takbeerat, Salam, and Tashashud.
• An App allows customization of the Salat sequence, so a person can perform Salat according to his/her Islamic school of thought.

Project Implementation Method

This project is composed of two components, of which the first is Android Development, and the other is the Machine Learning part. The Android application development will have its own development cycle, while the machine learning part will have its own cycle. Naturally, the two components will be developed in parallel with very loose coupling.

Producing Training Data:

As this app makes use of Machine Learning, we require data to train our classification model. Consequently, we need to produce this data:

• We will use smartphones to record individuals performing Salat because the data distribution of training and test data should be as similar as possible.

• Some related data is available in a repository [10]. This data lacks some poses and also doesn't differentiate between some poses.

• More data is definitely needed and will be generated using smartphones and from videos and images available online. 

Preparing Data:

• Annotating and preparing more training data will require time and energy. This will process will continue till the end as the more data we have, the better the model will be.

• We will make use of python scripting to convert data into the appropriate structure for the model.

• We also aim to develop a JavaFX-based desktop application to assist us in data preparation and labeling.

Choosing the Machine Learning Model:

We will start with a baseline image classification model, which is just a regular ConvNet, and proceed to more complex methods as needed.

Keeping the limitations of the smartphone hardware in mind, we require our model to be as less resource hungry as possible all the while maintaining good accuracy.

In summary, model selection will be based on experimentation subject to constraints.

Training Machine Learning Model:

We aim to make use of GPUs to help us speed up the training process, as deep Learning-based computer vision techniques can take up a lot of time to train a model.

The Andoid Application:

For the application development, broadly, our android application will go through the following phases:

• Designing User Interface.
• Implementing the User Interface in the app.
• Add the programming logic to the app.
• Add the classification model to the app.
• Testing and improvements.

Benefits of the Project

This app can help Muslims in practicing and performing their Salat Correctly. This includes:

• Children who are learning to pray.

• Adults and children who are careless and/or lose concentration and therefore forget the progress of their prayer.

• Aged, forgetful, or people with memory loss issues. About 40% of people aged 65 or older have age-associated memory impairment—in the United States, about 16 million people.[2]

• New Reverts to Islam. According to a study [3], the number of American Muslims has been growing by around 100,000 annually.

Technical Details of Final Deliverable

The final deliverable will be a smartphone-based Android application that will:

• Allow the worshipper to choose between different Salat.
• Based on the chosen Salat, our app will guide the worshipper through a sequence of poses.
• make use of the front camera to capture the images of the worshipper for the classification of poses.
• Classification of the poses will be done by the classification model trained using powerful GPUs.

Final Deliverable of the Project HW/SW integrated systemCore Industry HealthOther IndustriesCore Technology Artificial Intelligence(AI)Other TechnologiesSustainable Development Goals Good Health and Well-Being for PeopleRequired Resources

Item Name	Type	No. of Units	Per Unit Cost (in Rs)	Total (in Rs)
			Total in (Rs)	79700
GPU-GeForce GTX 1660 Ti	Equipment	1	70000	70000
SD card	Miscellaneous	3	500	1500
USB cables	Miscellaneous	2	1000	2000
card reader	Miscellaneous	1	200	200
motherboard interface for GPU	Miscellaneous	1	6000	6000