A Temporal Cross-Based Stereo Matching Method

碩士 === 國立成功大學 === 電腦與通信工程研究所 === 103 === In the recent year, many methods is developed for stereo matching to generate a better depth map, but lots of them consider an instant time only. It will meet a bottleneck to the accuracy because of the limited information. Moreover, the depth map of a video...

Full description

Bibliographic Details
Main Authors: Hou-ChunMok, 莫皓全
Other Authors: Pau-Choo Chung
Format: Others
Language:en_US
Published: 2015
Online Access:http://ndltd.ncl.edu.tw/handle/dqt56q
Description
Summary:碩士 === 國立成功大學 === 電腦與通信工程研究所 === 103 === In the recent year, many methods is developed for stereo matching to generate a better depth map, but lots of them consider an instant time only. It will meet a bottleneck to the accuracy because of the limited information. Moreover, the depth map of a video sequence will flicker due to the movement of the object and the neighborhood of each pixel. In order to solve the problem mention above, in this thesis an algorithm which considers the spatial-temporal information is proposed to enhance the accuracy and the consistency in a sequence. The paper is divided into three parts. The first part is to compute the disparity between the left and the right images according to the similarity. Then, we search the corresponding pixel between the before and after frame and add the temporal cost into consideration. Third, we perform post processing such as left-right check and bilateral filter to de-noise. Temporal median smooth the flickering and discontinuity as a sequence.