关键词: 近景运动目标检测, 多模态均值, 前/背景分割, 最大流/最小割

Abstract: In intelligent video surveillance field，the moving target detection is one of the fundamental tasks and an active topic for image sequence analysis.When detecting big target moving slowly in nearby view，aiming at the shortcoming of the temporal detecting algorithm which can only detect edge and internal holes of complete segmentation problems，it proposes a method of close-range moving target detection based on the space-time information.This algorithm uses spatial information of image to research foreground/background division technology，based on the multimodal mean of motion history at temporal domain.Then，by minimizing energy model，the structure of network and the theory of network flow，it converts the target detection to the problem of max-flow/min-cut.Experiment shows that the proposed algorithm can overcome gradual changes of the illumination，background disturbances and slight shaking of the camera，and update foreground/background effectively in the multimodal environment.It can detect moving targets accurately and fully.

Key words: close-range moving target detection, multimodal mean, foreground/background segmentation, max-flow/min-cut