Motion segmentation of surveillance video is the basis and premise of intelligent video processing such as video synopsis and behavior recognition,and is a research hotspot in the field of computer vision.Nevertheless,existing motion segmentation methods usually suffer from cumbersome steps and mass calculation, and their application is restricted in the field of limited computing power.To address these issues,we propose a method called motion segmentation of surveillance video by combining dichotomous and spatio-temporal tube.Firstly,the initial spatio-temporal flow is calculated using nested elliptical spatio-temporal tube model to judge the completeness of target trajectory (CTT).Secondly,we adjust dynamically the elliptical sampling line by combining the bisection to capture adaptively the moving target in the sampling area.Finally,the pixels on the sampling line are extracted to form the adaptive spatio-temporal tube for motion segmentation.Experimental results demonstrate that the proposed method outperforms the state-of-the-art methods in terms of both computing speed and accuracy,which has strong robustness and is also suitable for surveillance scenarios with changing motion.