Human Activity Recognition Using Single Frame CNN

Abstract

Human Activity Recognition (HAR) is one of the most ongoing research fields in computer vision for different contexts like surveillance, military, healthcare. So Human Activity Recognition is a category of time series classification problem that needs to be solved. Here the data from a series of timestamps to accurately detect the action that is present in the video is considered. This study uses single- frame Convolutional Neural Networks as a learning method for Human Activity Recognition (HAR). Unlike the other conventional machine learning methods, which require expertise on domain-specific knowledge, Convolutional neural networks can extract the features automatically. This study comprises reading and pre-processing the dataset, constructing the Convolutional Neural Networks model as per need, Compiling and training the model and using the single-frame CNN method. This approach is implemented using various technologies like Tensor flow, Open cv.