- 學位論文
具自動化對焦功能之立體視覺影像追蹤
Stereo Vision-Based Image Tracking System with Auto-Focus Capabilities
摘要
摘要 由於大部份的影像監視系統常常受到攝影機解析度不足、拍攝角度固定、目標距離太遠及環境光源不足的影響，導致記錄的影像品質不佳，或遺漏關鍵畫面，或無法清楚辨識，雖有監視設備，卻無法提供有參考價值的影像。本文的研究目的在發展一部由雙攝影機構成之立體視覺影像追蹤系統，改善上述的缺點。第一，此系統可偵測環境中的移動目標物，進行追蹤，不會遺漏關鍵畫面。第二，本文提出一混合式被動對焦方法，乃結合立體視覺演算法及基於影像清晰度的對焦點預估法，其中，立體視覺演算法能求出移動目標物與攝影機間的距離，估算攝影機的放大倍率及縮小對焦點的搜尋範圍；而基於影像清晰度的對焦點預估法搭配多重閥值顏色篩選，能將影像清晰度函數簡化成單波峰狀，再利用雲形線內插，即可快速且準確的預估對焦點位置，以記錄清晰且適度放大目標物的影像，有效解決目標物距離太遠時影像太小或放大失焦的缺點。
並列摘要
ABSTRACT Most of video monitoring system is often hampered by lack of camera resolution, the shooting angle fixed, objective effects of distance and lack of ambient light, led to record images of poor quality, omit key images or cannot be clearly identified. It can't provide worthy video frame, notwithstanding have surveillance and monitoring system. In order to improve the above shortcomings.This research aims at developing a mobile imaging tracking system which consists of dual camera stereo vision. First, the system is able to detect moving targets and tracking it, but it would not left out key images. Second, the purpose of this dissertation present a mixed type passive focus method which is combination with stereo visual algorithm and focus point estimation basing on image sharpness function. Stereo Visual algorithm can not only seeking out distance between target and camera, but also estimateing cameras of zoom rate and narrowing on focus of search range. To clarify the record target image and moderate magnification. We can use focus point estimation basing on images sharpness function mixs color multithresholding filter method. This method can simplify into one peak on image sharpness function. And using cubic splines interpolation can quickly predictives focus location very accurate. To resolve the image is too small when the traget is too far or when you zoom in traget defocused a result of nonpoint.