Real Time Human Activity Detection using Improved DCNN Based on Transfer Learning

Abstract

Nowadays the world become a digitalization. That's why now security problems are rising every day. We have too much concern about the security system. So human activity detection’s significance is also increased argent. Human activity detection (HAD) is a significant series of data in the computer vision community. Nowadays we are utilizing CCTV cameras, Smartphone cameras for reliability plan. Human action acknowledgment is worried about recognizing various kinds of human developments and activities utilizing information accumulated from different ways. In this project report, we are proposing an improved DCNN that can recognize eating, walking, working, playing, fighting, and other types of real-time human activity are being from images. Using the Deep Convolutional Neural Network model (DCNN) images were fed for image classification. Then, by means of joining the DCNN model with a custom human action identification dataset, limits and new attaching is one great step. The benefit of an improved Deep Convolution Neural Network or (DCNN) is its capacity to separate attributes from the data. Transfer Learning was used to feature extract the images and also the methodology we used is transfer learning. Used Keras framework to train the images. In our Project, we used Keras and TensorFlow as a framework. Moreover, we have contrasted the further improved DCNN model and other conventional techniques, and here the improved DCNN model accomplished an accuracy pace of 98.82% and outperforms different models.