請用此 Handle URI 來引用此文件:
http://tdr.lib.ntu.edu.tw/jspui/handle/123456789/8005
標題: | 視覺感測系統之資訊密度的解析度控制與資料獨特性的傳輸量控制 Region-of-Interest-Based Quality Control and Keyframe-Based Quantity Control in Visual Sensing System |
作者: | Yi-Chun Lin 林意淳 |
指導教授: | 連豊力 |
關鍵字: | 解析度控制,傳輸量控制,資料獨特性,動態取樣,量化回饋穩定性,變焦控制, Quality control,Quantity control,Keyframe extraction,Dynamic sampling,Quantized feedback stabilization,Zoom control, |
出版年 : | 2014 |
學位: | 博士 |
摘要: | 視覺感測器因其具有競爭性的價格和豐富的感測資訊量,因此,在過去幾年間被廣泛的運用在特定的區域,並且有目的性的收集感興趣目標物的影像資訊。在許多應用上都可見到其提供協助或是監視功能的蹤跡,例如: 工業機器人、軍事防禦和監控系統。隨著視覺感測系統使用的快速增加,也產生了越來越多需要被傳輸影像資料。然而,要在一個共享且頻寬有限制的網路上進行大量的影像傳輸,是非常困難且具有挑戰性的。其引起的嚴重延遲以及封包丟失會大大的降低系統表現和影像分析結果。為了兼顧系統表現和可靠穩定的傳輸,在本論文中基於資訊密度、資料獨特性和系統動態性,提出針對影像資料的解析度和傳輸量的控制方法。就理論面而言,影像資料的解析度控制被轉換成量化回饋穩定性問題並且以 Lyapunov 的方法加以證明。就 實際應用上,所設計的影像解析度控制策略被實現在 PTZ 相機上,並且在室內和室外的實驗場景都得到極佳的表現。另一方面,影像資料的傳輸量控制可視為影像摘要問題,為了確保在經過資料減量過程後,系統表現依舊維持在令人滿意的範圍內。因此,在本論文,提出基於感知運動能量來設計針對獨特性資料的取樣策略並藉此移除重複性高的資料。接著,將其實現在大量且豐富的實驗場景,不僅可以得到將近50% 的優秀資料減量結果,更重要的是,系統表現也維持在可接受範圍內。再著,在本論文中所提出影像資料解析度跟傳輸量的控制方法也與其他方法做比較,並藉此展現其卓越的優勢。 Visual sensors are widely used in the significant area, such as industry, army and public to collect the abundant video-related data about the objects of interest in the past years due to reasonable price and unique sensing capability. They have been found in various applications such as industrial robotics, military defense and surveillance for assisting and monitoring purposes. The rapid use of visual sensing systems has led to the increased amounts of video data that impose implicit difficulties on video transmission task over a shared and bandwidth-limited communication network. Moreover, control performance or video analysis results would be greatly degraded in the presence of constraints such as severe delays and packet dropouts induced by excessive transmitted video data. For taking desired performance and reliable transmission into account, quality and quantity control of video packet data are proposed based on information density and system dynamics in the dissertation. Quality control is modeled as the quantized feedback stabilization problem and is proved in the sense of Lyapunov. In practical applications, designed quality control policies are implemented in the camera with zoom functionality and are experimented on indoor and outdoor environment to clearly demonstrate the better human tracking and detection performance. On the other hand, quantity control is modeled as video summarization problem. In order to prevent the system performance from being influenced during data reduction process, the designed keyframe extraction rules based on perceived motion energy are proposed to remove the similar frames. The exceptional and near 50% data reduction ratio and acceptable tracking, detection and transmission results are presented with abundant typical and experimental videos. Furthermore, the proposed quality and quantity control are also compared with the other approaches to present its outstanding advantage. |
URI: | http://tdr.lib.ntu.edu.tw/jspui/handle/123456789/8005 |
全文授權: | 同意授權(全球公開) |
電子全文公開日期: | 2024-03-22 |
顯示於系所單位: | 電機工程學系 |
文件中的檔案:
檔案 | 大小 | 格式 | |
---|---|---|---|
ntu-103-1.pdf | 7.47 MB | Adobe PDF | 檢視/開啟 |
系統中的文件,除了特別指名其著作權條款之外,均受到著作權保護,並且保留所有的權利。