Novel deep learning models for yoga pose estimator

Abstract

Abstract Yoga pose recognition and correction are artificial intelligent techniques to provide standardized and appropriate yoga poses. Incorrect yoga poses can cause serious injuries and long-term complications. Analyzing human posture can identify and rectify abnormal positions, improving well-being at home. A posture estimator extracts yoga asana attributes from properly represented images. These extracted features are then utilized directly as inputs for various neural networks and machine learning models. These models serve the purpose of evaluating and predicting the accuracy of specific yoga poses. The objective of this research is to explore multiple methods for classifying yoga poses. The LGDeep model is introduced, which combines a novel residual convolutional neural network with three deep learning approaches: Xception, VGGNet, and SqueezeNet. Additionally, the LGDeep model incorporates feature extraction methods such as LDA and GDA. Experimental results demonstrate that the LGDeep classifier outperforms other approaches and achieves the highest classification accuracy ratio.

Keywords

• Image classification
• Yoga pose deep learning
• Feature extraction
• Artificial intelligence