YogaPoseVision: MobileNetV3-Powered CNN for Yoga Pose Identification
Poonam Shourie, Vatsala Anand, Deepak Upadhyay, Swati Devliyal, Sheifali Gupta · 2024
Yoga is a popular type of meditation and exercise that has several health advantages. Recent developments in deep learning and computer vision have made it possible to automatically recognise and classify yoga poses, providing instructors and practitioners with useful resources and increasing life expectancy. This article suggests a unique method for classifying yoga poses using the MobileNetV3 architecture and convolutional neural networks (CNNs). The durable and lightweight model that is appropriate for real-time applications is developed by utilising the MobileNetV3 architecture, which is widely recognized for its efficiency and accuracy. Through the use of transfer learning techniques, the model is fine-tuned on a specialized dataset of photographs depicting yoga poses after being pretrained on a sizable dataset of different images to extract generic features. The collection is made up of annotated photos of numerous yoga positions in a range of variations and difficulty levels. In terms of precisely categorizing yoga postures, the suggested CNN architecture with MobileNetV3 shows encouraging results, attaining high classification accuracy while retaining computational efficiency. To evaluate the model's ability to identify between various yoga positions, performance evaluation metrics like precision, recall, and F1-score are used. In general, the incorporation of CNNs into the MobileNetV3 architecture signifies a noteworthy development in the automation of yoga posture classification, hence further advancing computer vision technologies related to health and wellness.