Islamic Multi-State Salat Posture Dataset (IMCSPD): A Vision-Based Dataset for Comprehensive Prayer Position Recognition Across Standing, Sitting, and Lying Conditions

The Islamic Multi-State Salat Posture Dataset (IMCSPD) is a structured and domain-specific dataset developed to support research in vision-based human pose estimation and intelligent prayer assistance systems. Unlike conventional Salat datasets that primarily focus on standing prayer performed by healthy individuals, IMCSPD is designed to capture the diversity of prayer postures across multiple physical conditions, making it suitable for inclusive and assistive AI applications.

The dataset consists of a total of 1,267 labeled images collected from 10 different participants to ensure variability in body structure, posture execution, and natural movement patterns. All images are manually annotated and categorized into 25 distinct Salat posture classes, organized into three primary modalities based on body orientation:

1. Standing Modality (9 classes): Represents the standard Salat sequence, including Qiyam, Takbir, Qiyam Recitation, Ruku, Sujud, Jalsa, Salam Right, Salam Left, and Dua. This modality emphasizes full-body skeletal dynamics such as joint angles and vertical alignment.

2. Sitting Modality (8 classes): Designed for elderly individuals or users with physical limitations (e.g., joint pain or mobility constraints), this modality captures seated prayer postures. It focuses on upper-body movements, torso inclination, and center-of-mass distribution.

3. Lying Modality (8 classes): Specifically included for bedridden or physically impaired individuals (e.g., post-surgery or ICU patients), this modality represents prayer performed in a lying position. It emphasizes head orientation, limited limb movement, and symbolic gesture recognition.

The dataset includes 571 images for standing postures, 425 images for sitting postures, and 271 images for lying postures, ensuring a balanced representation of real-world prayer conditions. The data collection process was conducted under controlled yet natural conditions to preserve realism while maintaining consistency in posture labeling.

IMCSPD is intended for applications in computer vision, human activity recognition, assistive healthcare technologies, and intelligent religious guidance systems. The dataset can be used for training, benchmarking, and evaluating machine learning models for multi-state posture recognition.

To ensure ethical compliance and privacy, no personally identifiable information is included in the dataset, and facial details can be optionally anonymized depending on usage requirements.

This dataset is publicly released to encourage further research and reproducibility in the field of inclusive human-centered AI systems.