STM Journals Peer-reviewed research from Consortium e-Learning Network

Journal of Computer Technology & Applications

A Survey on Deep Learning based Detection of Abnormal Human Behaviour using Computer Vision Human Activity Recognition System

  1. Partha Ghosh
  2. Sayantan Roy
  3. Sombit Bose
  4. Avisek Mondal

Abstract

Abnormal Human activity recognition (Abnormal HAR) systems are very popular among researchers nowadays, they attempt to identify and analyze human activities using acquired information from sensors. Several papers have already been published in the abnormal HAR topics, the technologies in this field have multidisciplinary nature. They need constant updates. Our literature survey divided the approaches into three categories, the first one is about wearable sensor-based approach, the second one is about smartphone sensor-based approach and the third one is about computer vision-based approach. The first collects data from sensors using sensing devices, whereas the smartphone sensor takes input from smartphone sensors such as the gyroscope and accelerometer, and the last classifies activities using pose estimation, which necessitates the estimation of body key points using a neural network. Our literature survey attempts to review the Abnormal HAR systems from the deep learningbased computer vision perspective. We have further categorized the vision-based models into three basic types, the first one is the generative model, the second one is the discriminative model and the third one is about the hybrid ones. We have listed all recent models on various new datasets like UCF-Crime. Abnormal HAR has too many applications such as security, video surveillance, and home monitoring are highly related to abnormal HAR tasks. This establishes a new trend in abnormal HAR systems. Our survey aims to provide the reader an analysis of Vision-based HAR., it also checks the challenges and future scope of our survey.

Keywords

Support