Leveraging a wearable for activity recognition in salat

Abstract

Salat, the most important worship of Muslims and the second pillar of Islam, is an integral part of the Muslim community. Salat involves a series of steady and transitional activities to be performed in a specific sequence. In the HAR literature, such activities are termed as complex activities [6, 7], which is the most challenging category of activities to recognize [6]. On top of that, Salat has variations based on time, priority, school of thought, etc. The activities in Salat, the postures of the same activity, etc., differ greatly based on these aspects. There also exist chances that people make different mistakes in Salat, such as forgetting an activity or adding an extra activity while performing Salat. Considering all these facts, it is understandable that recognizing activities in Salat is indeed a challenging task. Existing research studies related to recognizing individual activities in Salat either demand capturing images by a camera for image processing or carrying a smartphone (sometimes in inconvenient places) while praying for capturing sensor signals. Both of the demands are not convenient or applicable in real cases. Besides, the existing studies lack user-independent accuracy analysis and fine- grained prediction. Therefore, the literature is yet to provide a convenient, and user- independent solution for activity recognition in Salat that is applicable in general. To address this gap, in this study, we first assess the requirement and acceptability of technological solutions for activity recognition in Salat by conducting an online survey. Here, the objective is to analyze whether people need and are willing to explore such solutions for their Salat. Our key findings from the analysis are that mistakes in prayer are common, and people, in general, are eager to explore convenient technological assistance for improving their Salat through corrective measures. Upon establishing the requirement, to go further in addressing the gap in the literature, we propose an activity recognition methodology using a smartwatch to recognize different activities in Salat, such as standing, bowing, prostrating, etc., so that a user can assess different aspects of his prayer such as correctness, completeness, etc., of his prayer. We prepare a Salat activity dataset through collecting data from 30 subjects using a smartwatch. Upon preprocessing the collected raw data, we separate the steady and transitional states of Salat following two different ways - 1) using machine learning, and 2) using Signal Magnitude Area (SMA). Afterwards, through applying some semantic rules derived from domain knowledge, we recognize a couple of steady states, namely bowing and standing, with perfect accuracy. Next, we develop a pattern database for the transitional activities in Salat. We classify the transitional activities using Dynamic Time Warping (DTW) algorithm based on the best-matching template set. Finally, a

postprocessing algorithm detects misclassifications and applies domain knowledge again to correct them to provide the final predictions. In the process of performing all these tasks, we propose a new pipeline of methodology for activity recognition in Salat. We perform user-independent analysis over the performance of our proposed methodology and achieve a near-perfect accuracy (99.3%). Along with this very high accuracy for the first time in the literature, our model offers fine-grained recognition of the individual activities in Salat. Besides, our methodology is robust enough to overlook the extra transitional activities a person performs while praying, which does not nullify Salat. To demonstrate the robustness of our methodology and the implication of each of these steps of the methodology, we perform rigorous experimentation and analysis. Findings on the experimentation and analysis demonstrate step-by-step advancement towards achieving near-perfect accuracy in real-life settings. Therefore, this research, covering the ground study on user requirements and developing a new methodology to achieve near-perfect accuracy in activity recognition in Salat, is expected to lead towards a comprehensive solution for monitoring Salat. The solution, in the future, could provide a detailed summary with appropriate feedback to a worshipper to help him improve his quality of Salat.