Dissertation > Industrial Technology > Automation technology,computer technology > Computing technology,computer technology > Computer applications > Information processing (information processing) > Pattern Recognition and devices > Image recognition device

Research on the Approch for Depth Estimation and View Transformation from a Single Outdoor Image

Author ZuoShuo
Tutor ZhangShiHui
School Yanshan University
Course Applied Computer Technology
Keywords Two-dimensional image Depth estimation Occlusion recovery Viewtransformation Image inpainting
CLC TP391.41
Type Master's thesis
Year 2012
Downloads 56
Quotes 0
Download Dissertation

In the vision field such as three-dimensional reconstruction, object recognition andmachine navigation, the problem of depth estimation from a single two-dimensional imagehas become one of the most important topics. How to estimate scene depth accurately is akey factor for three-dimensional reconstruction. Based on analysis of the present situationof the domestic and foreign research and a single outdoor image, this papercomprehensivly uses many aspects of theories knowledge and do studies on the approachfor depth estimation and view transformation based on a single outdoor image.Firstly, the overall flow of the proposed depth estimation and view transformationmethod is presented. And feature-based hierarchical segmentation method, features andclassifiers training method for region marking are introduced in detail.Secondly, a more accurate depth estimation model is defined and a more reasonabledepth estimation method is elaborated. The proposed depth estimation modelcomprehensively takes the camera’s rotation and pitch into consideration. The proposeddepth estimation method classifies standing object region into two cases of with visibleground-contact points and without visible ground-contact points to accomplishing itsdepth estimation task, thus the application scope of the proposed method is broaden.Thirdly, According to the obtained depth estimation result, an occlusion recoverymethod for determining the occlusion relation of adjacent regions and a viewtransformation method for creating the new view after any rotation, translation and pitchare presented. At the same time, image inpainting work for the missing color region isconducted with its neighbor’s color.Finally, experimental results show that the proposed method is feasible and effective,and the analysis of experimental results are made.

Related Dissertations
More Dissertations