Font Size: a A A

Light Field Depth Obtain And Algorithmof Segmentation And Reconstruction Based On Depth Data

Posted on:2016-03-06Degree:MasterType:Thesis
Country:ChinaCandidate:P C MaFull Text:PDF
GTID:2308330479450619Subject:Control theory and control engineering
Abstract/Summary:PDF Full Text Request
Light Field reconstruction is a hot topic since 2013, and is a relatively new direction in computer vision. There also appears light field related products called Light Field Camera, such as Lytro and Raytrix. But currently, there are few research and papers about light field reconstruction, and there has not yet formed a complete and effective solution in the field of depth and three-dimensional reconstruction of light field.In this paper, we have a deep research on depth recovery in high resolution light field images and human body segmentation and three-dimensional reconstruction based on depth data obtained by Kinect, mainly including the following aspects:(1) A depth reconstruction algorithm based on epipolar plane image is proposed. According to the special linear structure of EPI, a cross-detect model is proposed to detect the outlines, whose depth is then computed by combining the exponent distance function and distance weight coefficient. Then we use the contour depth as a priori to the inner flat regions, and integrate the priori and likelihood into an energy function. Finally, the contour depth is propagated to the whole depth map by minimizing the energy function.(2) A random decision forest algorithm is used for human body segmentation and gestures recognition, which is based on the depth data obtained by Kinect. In this system, We use more than 20000 labeled human body depth images as training samples, and use the maximum information gain guidelines as a standard feature selection. Finally, three decision trees are trained as a gesture recognition classifier.(3) We present a fast realistic 3D modeling system to reconstruct textured 3D model using the Kinect depth data and RGB data. In order to reconstruct the precise object, detection and extraction steps are used for object segmentation. When the azimuth between the subject and the camera changes, the system can automatically relocate the pose and fuse the new incremental frame into pre-existing model. By this method, this system finally finishes acquires the complete textured 3D reconstruction of the subject within a short time.
Keywords/Search Tags:Light field reconstruction, Gesture recognition, Three-dimensional modeling
PDF Full Text Request
Related items