Random sampling-based background subtraction with adaptive multi-cue fusion in RGBD videos
Jianwei Huang, Hefeng Wu, Yongyi Gong, Dongfa Gao · 2016
Background subtraction is a fundamental and critical component in many high-level video processing applications. It remains a challenging task in complex scenes due to difficult factors such as illumination variations, cast shadows and dynamic backgrounds. In this work, we propose a novel background subtraction method for coping with RGBD videos. Our method is built upon a recent random sampling-based model and employs two consecutive procedures of multi-cue fusion to achieve foreground detection. In the first procedure, we obtain coarse foreground detection result by weighted fusion of separate outputs of color-based and depth-based background subtraction. Then multi-cue based adaptive refinement with spatio-temporal consistency is performed in the second procedure to produce the final foreground detection output. Experimental evaluations on several datasets demonstrate that the proposed method can handle challenging difficulties in complex scenes and it outperforms state-of-the-art methods both qualitatively and quantitatively.