Abstract |
: |
The labelling of image sequences with action labels is termed as human action recognition and the main goal of this process is the series observations on actions. Human activity recognition aims to make better representation using deep and shallow learning techniques. For humans moving in the scene, we use techniques for tracking, body pose estimation, or space-time shape templates and categorize activities based on the video’s over- all pattern of appearance and motion using spatio - temporal interest operators and local descriptors to build the representation. It can be support for different fields of studies such as human computer interaction, or sociology, medicine and other applications. The system uses Binary Motion Image (BMI) to perform human activity recognition after preprocessing and used as input for Convolutional neural network (CNN) algorithm to recognize the human activity. The CNN combined with support vector machine (SVM) algorithm can be used as the classifiers to recognize the human action. The algorithm has three kind of inputs test samples, train samples, train labels. This project proposes a joint learning framework to identify the performance of Deep and Shallow Classifiers in training videos. |