Training Video Data for Semantic Content Extraction in Surveillance using Tensorflow

Authors

  • A. Manju
  • P. Valarmathie

Abstract

In recent research, Video-based applications requires techniques to modify the objects in video recordings. The field of computer vision has since quite a while ago and strives to extricate understanding from pictures and recordings sequences. Video information is universal, happening in various ordinary activities, for example, surveillance, traffic, motion pictures, sports, and so on. This enormous measure of video should be investigated and prepared effectively to remove semantic highlights. Street crime is growing in a fast rate, which has requested progressively solid and smart open moderate framework. Such capacities could profit surveillance, video analytics and visually challenged individuals. While watching a long video, people have the uncanny capacity to sidestep pointless data and focus on the imperative events. These key events can be utilized as a more elevated amount depiction or outline of a long video. Propelled by the human visual cortex, this research manages such capacities in computers utilizing neural systems. Helpful or intriguing occasions are first separated from a video and after that profound learning techniques are utilized to extricate natural language summaries for every video sequence. Past methodologies of video portrayal either have been domain specific or utilize a layout based way to deal with fill detected objects, for example, action words or activities to comprise a linguistically right sentence.  This work proposes strategies to create visual outlines of long recordings, and furthermore proposes methods to train and produce textual summaries of the videos utilizing recurrent networks like Tensorflow. Fascinating fragments of long video are separated dependent on picture quality just as cinematographic and customer inclination. This tale approach will be a venturing stone for an assortment of inventive applications, for example, video recovery, programmed synopsis for visually impaired persons, automatic movie review generation, video question and answering frameworks.

Downloads

Published

2020-01-19

Issue

Section

Articles