Background subtraction algorithm is a kind of main moving object detection framework

but it is too difficult to build a model with short establishing period

high reliability and good robustness.From the perspective of scene perception

a technique for object detection based on the framework of background subtraction is proposed.To improve the reliability of initial model

the potential foreground pixels in background

which are obtained on the basis of stable structural information in the former two frames

are replaced by the nearest neighbor pixels belonging to background

when the initial model is being established in the second frame.Integrating color information with binary feature

a two-stage classification decision mechanism is proposed

meanwhile the local decision threshold and update frequency are adaptively adjusted in accordance with the texture complexity of pixels neighborhood.Subsequently

a feedback mechanism for misclassification is presented in the update model phase.Experimental results using challenging public video sequences show the effectiveness and superiority of the proposed method

compared with other state-of-the-art tracking approaches.