Intelligent Video Analytics For Human Action Detection: A Deep Learning Approach With Transfer Learning

Abstract

Human actions consist of a sequence of similar patterns which are difficult to classify using traditional image processing algorithms. Video analytics is a major research area that adds brains to eyes which means analytics to the camera. It monitors the video contents and extracts intelligent information from it. The human action analysis and its detection is a challenging task. The proposed method focuses on detection of normal human activity using Long-Short Term Memory (LSTM) as a deep neural architecture. The pre-processing technique of redundant frame detection along with pre-trained Convolutional Neural Network (CNN) is implemented for classifying the activities efficiently. Transfer learning approach is used followed by Long-Short Term Memory (LSTM) network to generate hybrid framework which further enhances the activity detection. Proposed method shows improvement in accuracy as compared to reference method. This method can be further implemented for on edge processing in embedded platforms for real time applications.