From Mats to Models: Applying CNN-SVM Hybrid Techniques for Yoga Pose Classification
Shiva Mehta, Savinder Kaur · 2024
This investigation report details a comprehensive investigation of a composite hybrid model of CNN-SVM developed particularly to identify the fitness patterns in yoga exercises. The model was trained and evaluated by an image dataset comprised of 2200 photos, which were labelled to five yoga poses, including class one, class two, class three, class four and class five. The study of the model tells us that it got an overall KPI of 89%. 17%. The precision rates through the machines vary between 87 machines. 93% and 90. 39% for the average rate of correct answers, and the recall rates spanned from 82 because the findings did range. 26% to 96. 86%. The performance of our model was 85%-92%—96% of measurements. The measurement gives proof of high efficacy in classifying poses, which is the point that the model has the intelligence needed to identify different yoga poses. The parameters were tweaked on the model throughout over epochs. On the other hand, the training and validation losses were consistently found to be diminishing while the accuracy was improving. It only tells us that the model can remember the patterns in the data. Anyway, overfitting constituted some problems, though. The study demonstrated the model’s capability to apply to yoga practice sessions in real-time settings. In this way, the technology can provide efficient alignments of the posture and the overall impact of the yoga poses. This can be followed up by improving the model’s generality and tackling the problems of the difference between the standing and sitting postures and identification similarity.