Video object segmentation based on adaptive background and Wronskian change detection model

In computer vision, detection of moving objects from a complex video scene is an important and challenging problem. It finds application in many computer vision and artificial intelligent systems. Background subtraction is a very popular and powerful technique in computer vision for moving object detection in the presence of stationary camera. In the proposed scheme, Wronskian change detection model (WM) is used to find out the change between the constructed background and the incoming video frame. In this paper we have used WM in the Gaussian distribution for video object segmentation. We have presented a new equation for variance updation in the neighbourhood. The parameters of Gaussian (i.e., the mean and the variance) are updated for linearly dependent pixels using a Gaussian weight learning rate in the neigbourhood. The result of the proposed scheme is found to provide accurate silhouette of moving objects in presence of illumination variation and unstationary backgrounds like fountain, ocean, curtain and Train. We compare our method with other modelling techniques and report experimental results.