Object Detection and Classification of Videos for Indexing &Retrieval Using Machine Learning

Abstract

In today’s world, people have access to a huge amount of videos, both on the internet and television. But the amount of video data is large and it is increasing day by day as the technologies grow. So video data storage, classification for indexing, filtering,, and retrieval is a big issues. So it is very difficult for a human being to go through all the videos to search for a particular video and it is time-consuming, but the user wants a particular category of video or video genre within a short span. So that there is a need to classify the videos based on their genre, and subgenre, research has begun on automatically classifying videos.

Many works has been done for the classification of videos in certain categories or genre and sub-genre, by filling the semantic between low level features of video and high level concepts, so that user can find their specific interest of video within a narrow domain. There are different techniques have been developed for good understanding of video content and various video features have been recognized for best representation of videos. There are different techniques have been developed for a good understanding of video content and various video features have been recognized for the best representation of videos. Such as Gaussian Mixture Model (GMM), Neural Network (NN), Bayesian, Hidden Markov Model (HMM), and Support Vector Machine (SVM).

Object-based video classification method has been adopted in our work. The video classification system can be roughly divided into two major components: a module for extracting video features from key frames and another module to find feature similarities between video frames and the object from the database. Here the proposed project work is to classify the videos from the repository by the method of extracting regions of interest from each key frame of an input video clip. Then these regions of interest kept as features, are compared with features of objects for video recognition using the MSER algorithm. Finally, we retrieve the videos from the repository by understanding the constrained user text query is presented. The method is evaluated by experimentation over a dataset containing different types of videos i.e. cricket, football, cartoon, news, and movies.