An object detection method based on adaptive background mixture models and Graph Cuts
Jinwen Tian · Chinese Journal of Stereology and Image Analysis · 2011
A new dynamic object extraction approach is proposed which is based on multi-Gaussian mixture model and Graph Cut method.In auto objects tracking system,extract the moving objects information from the image serials is the base of objects detecting and tracking.Nowadays,the methods of objects information extracting include of the temporal and spatial analysis of the image serials from video.Based on the temporal characteristics,we can extract the objects' moving information and then recognize the objects from the background.And using the image segmentation technology which is based on the spatial characteristics is also very hot in the auto object tracking research nowadays.So we propose an integrated method extract objects by adaptively fusing the temporal-spatial information,using the global optimal features of graph cuts.Firstly,to unsupervisedly learn the multi-Gaussian mixture model,build likelihood energy function,and then to use Graph Cut method to optimize the function to extract the motion objects.