Skimming of video analytics

Abstract

Video content analysis is an emerging technique to easily redact video footage for public disclosure and to identify events and objects in surveillance cameras. The proficiency of this analysis depends on various crucial parameters such as area under exposure, content of surveillance, time frame and prior knowledge on statistical tool to enhance the streaming analysis. Video Surveillance has been a widespread research avenue and the focus of research has been rapidly changing. In the current state of art, video processing system observes a recorded video, analyze frame size, quality of rendering etc., later on video processing worked around analyzing activities in the video. At the advent of CCTV cameras researchers moved ahead to identify human behaviors and pattern of motion. Human behavior recognition algorithms aids in detecting incidents and preventing abnormal behaviors which quantifies its importance. There is large computational overhead in handling the quantum of data in traditional machine learning approach. The systems used for supervision are efficient in analytics and they use simple methods like rule based analysis or analytics using keywords and metadata. Next, as the volume of data increases, surveillance systems provides only an infrastructure to capture, store and retrieve data but prediction of threats is a major setback. Data stored in warehouses are a combination of structured and unstructured data. Combining all these formats of data for analysis is a huge task, also adherence of machine learning algorithms to predict patterns in real time video seems to be a complex task. The need of the hour is a scalable newline