卷積神經網路於擁擠指標之研究

Analysis of Congestion Index using Convolution Neural Network Approach

目前國內外交控中心常用不同顏色用來描述路況資訊，主要在於簡單、畫面易懂；國內外的交通控制中心的路況擁擠程度，目前主要仍以道路速限為準，主觀地將速率高低分為幾種級距，以反應用路人對道路擁擠感知的等級，常會發生與用路人主觀之行車擁擠感知經驗不一致的現象。過去學術研究爰用進階之分類方法，雖有改善，仍有諸多改善空間。由於用路人係以視覺感知來判讀交通擁擠狀態，有鑑於此，本研究嘗試以圖像辨識之卷積神經網路技術，預測擁擠指標類別，並以路段固定偵測器之即時交通參數為基礎，本研究以高速公路為例，比較過去使用轉換之判讀方法，分析結果顯示，準確度大幅提升，高達82.9%。

"Congestion Index with color remarks a common way in traffic control area, mainly for its easy application and simplicity. Currently, the traffic congestion index are grouped into different levels in terms of speed range addressed for its inconsistency with the congestion perception of driver's experience. Although emerged advanced technologies improve its accuracy, it remains rooms for refinement. Since road users identify the states of road congestion by perception, which is obviously equivalent to sets of image extraction processing. It motivates this paper applying convolution neural network to predict the states of congestion classification. Compared with real-time traffic parameters from roadside detector by previous works, the proposed approach yields promising results with over 82.9% accuracy."