隨著許多民眾透過電腦、平板、手機等裝置，進行線上影片觀賞的服務，裝置可適性遞送研究近年來已成為最熱門的主題。在確保使用者清晰收視的前提下，根據裝置能力，有效的節省頻寬，將是本篇論文研究的議題。我們利用尺寸和解析度的數值計算出裝置的像素密度，再根據視網膜解析度概念計算出在特定距離下，人眼所能辨識的像素密度。將裝置像素密度與極限值比較，以超出的比例作為降低影音解析度參考的依據，進而降低影音傳輸所需的頻寬。另一方面，隨著行動裝置的硬體越來越高階，加上觀看線上影音只需透過瀏覽器外掛的播放器，因此我們不需實際偵測硬體與解碼器，僅需偵測播放器與計算即時頻寬即可。除裝置端能力外，網路的頻寬狀況也會影響觀看影音的品質，因此我們也研究影音伺服器位置對下載速度的影響，系統將判斷影片來源存在於哪些伺服器上，然後網頁上的播放器再重新定向影片的位置開始播放。最後，我們提出一Heuristic Bandwidth Estimation演算法，在使用者等待時間與精準偵測下載頻寬中取得一平衡點。在實驗展示中，我們變動裝置與網路環境，即時偵測裝置能力，並給予適當解析度影片。實驗證明本系統技術，確實根據即時頻寬與裝置能力給予不同影片，達到節省頻寬與清晰收視的目的，突破舊有方法的限制，突顯優越性。

More and more people use devices such as PC, tablets, and mobile phones to watch online videos.Adaptive streaming data delivery method based on device capability is getting attention in recent years. Themain topic of this research is saving bandwidth effectively based on device capability with clearly watchingexperience. We calculate Pixel per Inch (PPI) of device with a size and resolution of screen, and thenaccording to retina concept, we calculate suitable PPI that people eyes can identify. The difference betweendevice PPI and retina PPI will be referred to adjust the quality of streaming video and further reducenecessary bandwidth. Besides, with the evolution of hardware and software, there is no need to physicallydetect hardware capability, but web browser plug-in to evaluate playing of online videos. By the way, it can beeasily achieved to examine real-time bandwidth by web programming. In addition to the capability of devices,the condition of Internet traffic also affects the quality of watching videos. Therefore, we discuss the relationsbetween server positions and download speed. Our system will determine the source of videos is from, andthen the player plays the re-directed part on the web. Because of the uncertainty of network traffic, aHeuristic Bandwidth Estimation algorithm is proposed to evaluate a precise download bandwidth withinconstant time. In the experiment, results show great effort of saving bandwidth with clearly watchingexperience with distinct device and Internet environment.