Efficient techniques for motion estimation and event detection in video surveillance

Abstract

Digital image and video processing technologies have revolutionized the world and have attained worldwide acceptance among the public in several forms of applications such as abnormal event detection, abandoned object detection, intruder detection, fall detection, traffic monitoring, crowd control, theft and fraud prevention in banks, households, business organizations etc. Digital image and video processing mainly deal with processing the image and video information for quality enhancements, achieving compression, data acquisition, extraction of useful information in the form of segmentation and so on. In addition, digital video processing application includes surveillance monitoring, frame rate conversion targeting display devices, video content analytics based on target applications, object tracking and so on. The increased usage of multimedia devices over the years have riggered several challenges in image and video processing while a few of them have been related to tradeoff among computational complexity and image quality, security and privacy concerns of the data, monitoring and maintenance f large volume of surveillance data. With several challenges ahead, the primary focus of this research is in dealing with an efficient Fast Forward Motion Estimation algorithm for video compression suitable for frame rate conversion, Video surveillance technique for the application of abandoned object detection and intruder detection. The work also involves analyzing a hybrid technique for fall detection algorithm by using a two stream classification approach. In the first research work, a Fast Forward Motion Estimation (FFME) algorithm is proposed and it aims in reducing the computational complexity as well as the risk involved in real time implementation. The algorithm proposed exploits the spatial correlation that exists with the neighboring block while predicting the motion vectors. newline