Estimation of human pose by tsallis entropy-based feature selection with ensemble machine learning model

Social concerns about how to care for the elderly who live alone have grown as a result of the growing problem of ageing. The use of video cameras in the monitoring of the elderly has been widespread. Despite this, these methods are difficult to use on a daily basis due to the camera's large storage capacity, sluggish processing speed, sensitivity to light, blind spot in eyesight, and the potential for privacy leakage. Initially, two public dataset such as Olympic sports dataset and UT-interaction dataset are used in this research work for pose estimation. During pre-processing, background subtraction, body part detection and filtering techniques are used to improve the input noisy images. After that, two techniques called distance based and angle based feature extraction techniques are used to extract the valuable information about the images. To improve the classification accuracy, Tsallis Entropy-Based Feature Selection (TEFS) is used and prediction process is carried out by proposed ensemble models such as Support Vector Machine (SVM), Naive Bayes (NB), Probabilistic Neural Network (PNN) and K-Nearest Neighbor (KNN). The experimental analysis proves that the PNN achieved better performance than other models. The PNN model achieved mean precision of 96.25% and KNN model achieved 92.30% of mean precision on UT-interaction dataset. The PNN achieved 92.56% of accuracy, KNN achieved 91.02% of accuracy and SVM achieved 90.53% of accuracy on Olympic sports dataset.