An Approach for Multi-human Pose Recognition and Classification Using Multiclass SVM

The aim of this paper is to recognize human activities and classify their activity using multi class support vector machine (SVM). There are many ways to predict human activity either vision based or wearable sensor based. In vision based, different types of sensors that can be used to address this task like a Kinect sensor. In this paper, a vision-based algorithm is proposed to identify multi-human activity so that we can predict the gesture Based on their action. The model is based on segmentation and filtering the environment and identifying human firstly, to identify frame based on semantic structure a dynamic distance separability algorithm is leading to divide a shot into sub shots for selecting appropriate key-frames in each sub shot by SVD decomposition. An adaptive filter is used for filtering process. Then to compose a feature vector, extraction of key poses is performed, and the meticulous Support vector Machine performs classification, and activity recognition. The application of this model is to recognize multi human and estimate their poses for classification. The future research area of this paperwork will be predicted abnormality of people’s health for medical purpose.