An Adaptive Nature Inspired Technique for Event Detection in Video Sequences

Abstract

Nowadays, the increasing popularity of video analysis (VA) among scholars can be newlineattributed to its broad range of uses and essential social impact. In multimedia and newlinecomputer vision (CV) uses, automatic detection (AD) of strict actions in internet video newlineobtains growing research attention from investigators. Video event detection (VED) newlinesupports screening human events and other graphical activities in videos, which is helpful newlinein domains such as military, commercial, public security, etc. Video surveillance (VS) has newlinegained considerable attention recently and is a primary research focus within the CV newlinedomain. Generally, the VS system framework gives the following steps: (i) Background newlinesubtraction, (ii) Environment modeling, (iii) objection detection (OD), (iv) detection, and newline(v) track moving objects (MO) and explaining of actions. The VS system mainly aims to newlineclassify and detect events using supervised and unsupervised methods. newlineSeveral techniques have been implemented for video event detection, such as graphical, newlineknowledge-based, significant margin-based approaches, etc. Nowadays, the generally used newlinemethod defines a video as a worldwide bag-of-word (BoW) vector. This approach can be newlineseparated into different steps: newline1. Local features {audio, attributes, and visual} are removed from sections of a video. newline2. The extracted feature sets are then quantized based on the knowledgeable dictionary. newline newline