SDL-Net: A Combined CNN & RNN Human Activity Recognition Model

Abstract

Human Action Recognition is quite popular among researchers and scientists and is considered one of the most active applications in the field of computer vision. It is quite useful in modern era applications like healthcare, surveillance, sports and many more fields. Deep Learning has provided an upliftment to predict human actions in an easiest way possible. This paper proposes a combined CNN & RNN human action recognition model named SDL-Net, which generates skeletal representations using Part Affinity Fields (PAFs) and generates skeletal gait energy images. It also captures sequential patterns to generate sequential data as well. Experiments are conducted on Kinect Activity Recognition Dataset (KARD) and it shows the efficiency and effectiveness by achieving desirable results.