Real-time human segmentation from RGB-D video sequence based on adaptive geodesic distance computation

• Published in: Multimedia tools and applications Vol. 78; no. 20; pp. 28409 - 28421
• Main Authors: Kim, Yeong-Seok, Yoon, Jong-Chul, Lee, In-Kwon
• Format: Journal Article
• Language: English
• Published: New York Springer US 01.10.2019; Springer Nature B.V
• Subjects: Algorithms; Color; Computer Communication Networks; Computer Science; Data Structures and Information Theory; Low noise; Multimedia Information Systems; Noise; Noise levels; Real time; Segmentation; Spatial smoothing; Special Purpose and Application-Based Systems; Video data; Geodesic matting; Depth video processing; RGB-D video; Natural background substitution; Real-time foreground segmentation
• ISSN: 1380-7501; 1573-7721
• DOI: 10.1007/s11042-017-5375-5

In this paper, we propose a method for extracting humans in the foreground of video frames using color and depth information. To ensure real-time performance and to increase accuracy, we classify a video frame into two parts by degree of noise: head region with high noise level, and non-head region with low noise level. Then, we apply a high-computational geodesic matting algorithm to the noisy head region that includes hair, and a low-computational hole filling with smoothing method to other regions. Additionally, we modify the traditional color-based geodesic segmentation algorithm to consider additional depth information. Then, we apply temporal/spatial smoothing to the blended foreground mask in order to enhance the coherence between video frames. Experimental results show that the proposed method outperforms a previous approach by accuracy and performance.