Segregating and recognizing human actions from video footages using lrcn technique

Abstract

Computer vision is a vast area of research that includes extracting useful information from images or sequence of images. Human activity recognition is one such field undergoing lots of research. The practical application for this model is vast in various kinds of researches as well as actual practice. This paper proposes a two-model approach using a combination of a convolutional neural network using transfer learning and a long short-term memory model. CNN network is applied to gather the feature vectors for each video, and the LSTM network is used to classify the video activity. Standard activities contain benchpress, horse riding, basketball dunk, etc. A high accuracy level of 94.2% was achieved by the proposed algorithm.