Motion Segmentation

Detection of abnormal behavior in an outdoor environment is one of the most crucial tasks. There are numerous method developed to detect abnormal event that occur in our surroundings and these methods are based on segmentation, feature extraction and classification. Segmentation has to be done to extract human from the video sequence. The human characteristics such as shape, poses, and body motions are then extracted and represent it by features. Then classification approach is applied on the extracted features this is used to recognize the various events.

3.1. Motion Segmentation Approaches:

The motion segmentation is performed on each frame of the video sequences and it is used to extract the human in that video sequences. The segmentation

The camera is fixed to a specific position and angle in case of static camera. The background is stable here and it never moves, so we can design background model in advance. The moving camera is a active camera which has a dynamic location and angle.

Static camera segmentations include Background subtraction method [6][11] , Gaussian Mixture Model[13] and Moving camera segmentation include temporal differencing[4].

Moving object segmentation in video sequence is important for many computer vision applications. Segmentation of moving object is extraction of foreground from background. Segmentation process includes steps as object detection and motion detection. Currently one of the most important and active research topics in computer vision is human segmentation. Object segmentation is very useful for tracking object and for object recognition in a video.

Most Common methods for segmenting the objects which are moving are background subtraction [5][11], temporal segmentation, edge detection, contour and the combination of temporal-spatial segmentation. These methods are used in automatic surveillance monitoring a scene to detect suspicious activities in shopping malls, offices, and plaza

 Background Subtraction Method considers all the frames for computation. Non-parametric background model[17]  It considers the statistical behavior of image features to segment the foreground from the background.

 This is proposed for background modeling.

 It is effective for background subtraction in dynamic texture scenes.  The processing time is high in comparison with the adaptive Gaussian mixture model.

Hierarchical background model.

[3]  It is based on the region segmentation and pixel descriptors that is used to detect and track foreground.

 It first segments the background images into several regions by the mean-shift algorithm.

 The hierarchical models first detect the regions containing foreground and then locate the foreground only in these regions.

 It reduce the time and cost. .  A larger number of regions will cause a longer processing time of the region model.

Warping background[17]  It presents a background model that differentiates between background motion and foreground objects.

 The background is modeled as a set of warping