No abstract

Last decade has witnessed an ever increasing number of video surveillance installations due to the rise of security concerns worldwide. With this comes the need for video analysis for fraud detection, crime investigation, traffic monitoring to name a few. For any kind of video analysis application, detection of moving objects in videos is a fundamental step. In this paper, an efficient foreground modelling method to segment multiple moving objects is implemented. Proposed method significantly reduces noise thereby accurately segmenting region of interest under dynamic conditions while handling occlusion to a large extent. Extensive performance analysis shows that the proposed method was found to give far better results when compared to the de facto standard as well as relatively new approaches used for moving object detection.

Digital video is becoming an emerging force in current computer and telecommunication industries for its large mass of data. Video segmentation and key-frame extraction have become crucial for the development of advanced digital video systems. Key frame extraction is a very useful technique to provide a concise access to the video content and is the first step towards efficient browsing and retrieval in video databases. Existing approaches are either computationally expensive or ineffective in capturing salient visual content. The proposed system extracts key frames from input videos using two distinct, cost-effective algorithms namely reference based key frame extraction and clustering. It uses multiple characteristics such as co-relation, optical flow and mutual information to identify and extract key frames. The proposed system is able to extract the key frames efficiently for any video format & the extracted key frames can satisfactorily represent the salient content of the video. Storage is reduced by one-eighth of the total space required by the original video and the original content can be represented in one-fourth the time of the input video achieving very high compression efficiency & hence can be used in any video retrieval applications.

No abstract

Abstract