Abstract: To effectively resolve the problem of moving objects detection in complex background scenes and overcome the disadvantages of the traditional single-pixel based modeling method,spatial information of video image sequence was introduced into background modeling process.Several methods were studied,including neighborhood updating,two-dimensional joint histogram information entropy judgment,spatio-temporal smoothing etc.Firstly,video images were pre-processed with guided filter algorithm to remove noise and preserve edge information of the interested objects.And then,implementation procedure and processing steps were presented.Finally,some qualitative and quantitative experiments and comparison analysis were carried out based on different benchmark datasets and reality video frames.Results show that,the proposed method outperforms other traditional methods for moving object detection.The method provides a novel approach for the problem of moving objects detection in complex scenes.