OFGM-SMED: An Efficient and Robust Foreground Object Detection in Compressed Video Sequences

摘要

Segmenting Foreground objects from a video sequence is a key processing and critical step in video analysis such as object detection and tracking. Several Foreground detection techniques and edge detectors have been developed till now but the problem is that it is very difficult to obtain an optimal foreground due to the interference from the factors like weather, light, shadow and clutter. Background subtraction is used in many of the applications, where the background noise appears at fixed places and also, when it is used for long image sequence, there may be more accumulated error in the foreground. Optical flow is the velocity field which warps one image into another (usually very similar) image where the background noise appears randomly. It covers long distance and the background noise due to brightness change is less which results in less accumulate error percentage. However, it cannot get rid of light influences which result in background noises. This paper proposes a new foreground detection approach to overcome these issues by combining the background subtraction algorithm and optical flow along with SMED (Separable Morphological Edge Detector) to reduce the background noises. SMED has robustness to light changing and capable of detecting even slight movement in the video sequence. The proposed work is highly accurate in detecting the moving objects in various scenarios such as fast moving objects, slow moving objects and even moving objects in dynamic scenes, where both the foreground and background changes.