VIDEO PROCESSING AND ANALYTICS FOR SMART CAMERA BASED ON RECONFIGURABLE ARCHITECTURE

Abstract

Image-based sensing is a powerful source of information in many application areas such as surveillance, monitoring and automation. The volume of data collected by the surveillance device is enormous. Therefore, images are processed to extract application-driven data, but centralized processing is not feasible in the presence of vast numbers of cameras distributed along broad geographical regions. In this scenario, when processing limitations and bandwidth availability are important restrictions, an automated process is required to extract the meaningful information from the collected data during surveillance process and to reduce the fatigue of operator. The requirement of surveillance, machine vision and the research interest has driven the technological and product development of smart camera based intelligent vision systems. These incorporate data management i.e., collection, indexing, storage and delivery of video along with the subsequent transmission of high-quality data for web-based monitoring. Therefore, video analytics plays an important role for proactive monitoring and alert generation in all video surveillance applications. It enhances pictorial information for human interpretation and analyzes scene data through machine vision modules. These features help in using the surveillance system for monitoring the specific region of interest without human intervention under adverse weather conditions. Smart cameras combine video sensing, video processing and communication within a single device; they are key components for novel surveillance systems. Smart cameras are employed for surveillance applications, to capture high-level description of a scene and to perform real-time analysis of scene. In view of the advantages mentioned above, smart cameras are a critical component of intelligent vision systems. They serve as edge devices for the visual surveillance grid and are widely utilized in defence and counter terrorism related activities for proactive monitoring. Development of high-performance imaging systems, including surveillance cameras, requires range improvement methods, a real-time implementation framework and assessment of video analytics to improve target identification and engagement, among other considerations. The aim of this research work is to thoroughly investigate the work on video processing by smart camera for long-range surveillance, low-light operation, and utilizing video analytics for unattended border area and security. There are four main objectives of the thesis. a) Analysis of video processing algorithms for long range imaging b) Framework for performance enhancement of imaging sensors c) Design and development of video analytics for surveillance d) Architectural design for hardware implementation of video processing and analytic algorithms This thesis primarily focuses on the video processing and analytics for smart camera based on reconfigurable architecture for surveillance system. The thesis consists of seven chapters. Each chapter begins with a brief introduction to the concerned problem and its possible solution. The experimental results are summarized and concluded at the end of each chapter. This thesis begins with an introduction to video processing requirements for smart camera for surveillance applications, design challenges, and the motivation of taking up the specific problem in Chapter 1. Chapter 2 presents an extensive literature review on the latest development in video surveillance technology, video analytics, architecture, and evaluation techniques for imaging systems. Chapter 3 describes video processing algorithms for range enhancement, using high dynamic range imaging techniques and real-time implementation aspects, the proposed high dynamic range imagingbased solution can enhance the enhance the surveillance range by 10%. Chapter 4 presents reconfigurable architecture for surveillance. Chapter 5 presents a framework for performance improvement of night vision devices that suffer from the scintillation noise problem under very lowlight conditions. It is demonstrated that an 20-30% improvement in visual quality is achieved for the dataset captured under controlled simulated light conditions. Video analytics for surveillance is presented in Chapter 6 along with evaluation using standard video dataset for change detection, it is demonstrated that the proposed method is efficient and meets the real-time constraint. Chapter 7 concludes the research work, with conclusions based on the results obtained and major outcomes. Future scope of the work for further improvement is also deliberated in the conclusion.