Silhouette extraction of a human body based on fusion of hog and graph-cut segmentation in dynamic backgrounds

• Published in: Proceedings of third International Conference on Computational Intelligence and Information Technology pp. 527 - 531
• Main Authors: Lakshmi, N.D, Latha, Y.M, Damodaram, A
• Format: Conference Proceeding
• Language: English
• Published: Stevenage, UK IET 2013; The Institution of Engineering & Technology
• Subjects: Combinatorial mathematics; Computer vision and image processing techniques; Knowledge engineering techniques; Optical, image and video signal processing; Optimisation techniques; Video signal processing; graph cut segmentation; graph theory; image fusion; object detection; SVM; feature extraction; image segmentation; graph cut method; silhouette extraction; subsequent analysis; video segmentation; gradient methods; video signal processing; dynamic backgrounds; HOG; histogram of oriented gradients; support vector machines; human regions; human body; sports videos; background information; hog fusion; support vector machine; video information; action recognition; human object detection
• ISBN: 9781849198592; 1849198594
• DOI: 10.1049/cp.2013.2641

In this paper we presents a novel and effective way for extracting a region based silhouette of a human with a moving background thereby facilitating subsequent analysis like action recognition. The system first detects objects in the video that can be classified as human or non human. For this, the Histogram of Oriented Gradients (HOG) is used as descriptors and Support Vector Machine (SVM) is used as a classifier. The localized human part also contains unnecessary background information. Hence, we propose to use Graph-Cut method for extracting the foreground (humans) information from the video. Since our goal is to extract only human regions, we propose a region-based approach that fuses Graph-cut segmentation with human object detection. Sports videos are used to test the proposed system and algorithms, and the extensive and encouraging experimental results show their effectiveness in getting region based silhouette of a player in the sports video and also supports its suitability for segmenting videos with dynamic backgrounds.