本論文借鏡於不同生物之雙眼視覺型態，期望以仿生方式結合超廣角影像拼接與立體視覺技術設計一種智慧型仿生雙眼視覺系統平台，藉以達到同時進行廣視域範圍監控與深度感測，改善現今許多雙眼機器視覺缺乏靈活性與適應性之問題。我們提出先以K-means演算法處理強化物件深度影像資訊於不同景深平面之特徵，再透過本研究核心演算景深影像接合方法達成任意興趣深度平面之完整影像拼接。經由一系列實驗與SIFT方法驗證，最終證實此系統能成功達成自適應專注聚焦於最為前景深度影像之物件，且該物件於影像拼接重疊區域不會出現分岔錯位之模糊情形。本研究方法結果將能順利結合交融立體視覺與影像拼接系統資訊之特色與優勢，並使得影像拼接系統將由傳統二維平面層次提高至三維立體空間作為研究與探討，以不同以往之角度審視仿生雙眼視覺平台發展之潛力與其未來之可朔性。;In this paper, we expect to design a smart binocular vision system platform combined ultra-wide-angle image stitching and stereo vision technology after learning binocular visual pattern from different creatures. It can simultaneously implement monitoring in wide field of view and depth sensing and improve the lack of flexibility and adaptability in binocular machine vision nowadays. We propose a method based on K-means algorithm to strengthen object depth information on characteristics of different image planes of depth of field, and achieving full image stitching in any interested depth plane. After a series of experiments and SIFT verification, the system is proved that can successfully implement adaptive focus on the object in the foreground of the image. Meanwhile, the object will not bifurcate or having dislocation in image stitching overlap region. Methods in this research are able to successfully combine the features and advantages of stereo vision blending and image stitching system information, and allow the system to enhance image stitching from the traditional two-dimensional spaces to three-dimensional space for research and discussion. Therefore, we can view the potential development and plasticity of binocular vision platform in a different way from the past.
