基於長短期記憶模型之智慧救護車自適應比特率影像直播系統

Abstract

一台智慧型救護車會不斷地分享病人的生命徵象數據和救護車內的影像流， 讓駐院醫生可以與救護人員合作，對病人進行早期評估甚至提供治療指南。不過 要讓醫生和救護人員之間合作順暢的前提是要有高品質的影像串流。影像品質差 或延遲高的串流會大幅增加對病人的準確診斷的困難度。 由於需要從移動的救護車傳送直播視訊，因此串流的效能受到行動網路的頻 寬快速變化的影響。在這種情況下，影片串流的比特率能夠快速適應行動網路服 務品質的變化並相應地進行調整是至關重要的。若影片串流的比特率超出行動網 路的可用頻寬，就有可能會發生卡頓以及破圖。我們必須讓醫生可以透過直播來 判斷病人的病況，要做到這件事，直播的時候必須儘量避免卡頓和破圖的發生。 為了達到這件事，我們必須使所用頻寬不超過頻道的可用頻寛。 為此，我們提出了EMS-RTC，一個針對救護車服務需求專門設計的即時視頻 串流平台。EMS-RTC 訓練了一個分類器，根據在現實駕駛情況下收集的一系列信 號和網絡相關指標的特徵，推斷出最佳的比特率。我們將EMS-RTC 與目前最具 代表性的自適應串流演算法（即Google 擁塞控制，GCC）進行比較。從真實世界 的行駛情況中收集的評估結果顯示，EMS-RTC 可以在可忽略降低一點影像品質的 成本下，將卡頓事件的總時間長度減少3.85 倍。

A smart ambulance continuously shares the patient’s vital signs data and video feeds from the ambulance so that the doctors stationed in the hospital can work with the paramedics in the ambulance, performing an early assessment of the patient or even providing treatment guidelines. One prerequisite to the seamless cooperation between the doctor and the paramedics is high-quality video streaming. Video streaming with bad image quality or high latency would introduce significant difficulty in providing accurate diagnoses of the patient. As the video needs to be delivered from a moving ambulance, the streaming performance is influenced by the fast variation of the service quality of the mobile network. Compared to static scenarios, additional channel impairments, such as the shadowing effect, is introduced in the mobile scenarios. In this case, it is crucial that the bitrate of the video streaming can quickly react to the changes in the service quality and is adjusted accordingly, such that the video quality leverages full available bandwidth, but does not exceed what the channel can support. To this end, we propose EMS-RTC, a real-time video streaming platform specialized for the need for ambulance services. EMS-RTC trains a classifier to infer the optimal bitrate based on features from a range of signal- and networkrelated metrics, collected in real-world driving scenarios. We compare EMS-RTC to a state-of-the-art ABR algorithm (i.e., Google congestion control, GCC). Evaluation results collected from real-world driving scenarios show that EMS-RTC reduces the total time duration of stall events by a factor of 3.85, at the cost of a neglectable reduction of image quality.