Target detection is difficult to be realized in complex scenes when there are moving background objects such as trees. In this paper

a new target fusion detection method is proposed based on background model. Firsty

by combining temporal information of per-pixel and the spatial information in the local region

we introduce a variant of multimodal mean model called the spatiotemporal multimodal mean model that is well suited for the non-stationary scenes. Then

the proposed background model is separately used to extract foreground pixels in visible and infrared image sequences

and a fusion detection method based on the confidence map is proposed to get the target detection result. The multi-sensor information can improve the detection precision and handle different environmental conditions. Experiment results demonstrate the effectiveness of the proposed method.