Utilizing a Hybrid Deep Learning Architecture For Salat Posture Detection

Abdul Salam Shah
Farhan Akbar
Muhammad Adnan Kaim Khani
Adil Maqsood
Fahad Shah Bukhari

Journal of ICT, Design, Engineering and Technological Science
Volume 9, Issue 1

A lot of Muslims have trouble getting their daily prayers right. You know, Salat with the movements and the recitations. It disrupts their religious duties. They do not get quick feedback on how their form looks. So we put together this system. It grabs images right as they happen. Then it checks them out using a convolutional neural network. That is CNN for short. It spots and confirms the basic postures in Salat. The thing covers six main positions. Takbir. Qiyam. Ruku. Sujood. Tashahhud. And Salam. Pretty much opens it up for tons of people to use. We tested how well it works. Looked at pose detection accuracy. Response time, too. And what users thought about it. Turns out the system helps a bunch. Folks can improve their Salat quality with it. Shows how computer vision and deep learning fit into something like this. Not your usual setup.