Human Body Poses Detection and Estimation Using Convolutional Neural Network

This study introduces a unique method for human body pose detection and estimation by combining convolutional neural network (CNN) and grab cut segmentation techniques. The suggested technology is meant to aid in the detection and estimation of human pose, which is important for many real-time applications. Features are extracted from pictures of human pose using a grab cut and create a human silhouette. Furthermore, the convolutional neural network is applied to classify the human pose. When tested on a dataset consisting of photographs of human pose, the suggested system achieved an accuracy of 93.89% in 6 human pose classifications. A total of 1181 pictures were used in this analysis, including six different human poses (down dog, warrior, tree, prank, goddess, and handshaking). There are 237 test photographs and 944 full-size images throughout all categories. An 80:20 ratio is maintained for training and testing. An F1 score of 93.75%, a recall score of 93.89%, and a precision score of 93.89% were all obtained using the proposed strategy. Based on the obtained data, it appears that the proposed method achieves good accuracy in pose detection compared to the state-of-the-art methods. It will be beneficial for yoga pose detection, patient detection systems, etc.