Human Detection from RGB Depth Image using Active contour and Grow-cut Segmentation

In modern Security and Surveillance technologies, the significance of human detection and segmentation becomes having much importance. In addition to security systems, image/video editing applications demand semantic segmentation of foreground objects. For such high-quality applications, the foreground object's boundary pixels need to be matched accurately. The proposed method aims for an automated segmentation scheme to detect and segment the human area from an image or video frame and to paste it in another frame/image using image processing techniques using a hybrid analysis of color and depth information. The approach provides a high-quality human factor segmentation scheme that can be used in Chroma keying operations in advanced multimedia editing applications. Depth based analysis along with a series of post-processing stages is employed. The video frames are taken using an RGB-Depth sensors. Hybrid depth and image analysis are used to segment the foreground human subjects semantically. The multi-level segmentation using Chan-Vase active contour detection, grow-cut segmentation and a Trimap based matting approaches have been used to achieve a fair segmentation accuracy. The results are evaluated using standard metrics and compared with state-of-the-art automated chroma keying techniques. The qualitative analysis shows the efficiency of the proposed hybrid depth based chroma keying scheme. It can be concluded here that the visual quality has been substantially attained by this method.