基於智慧影像分析模式之道路積淹水自動辨識系統

曾建澔(Chien-Hao Tseng) ； 林孟緯(Meng-Wei Lin) ； 謝佳蒨(Chia-Chien Hsieh) ； 吳志泓(Jyh-Horng Wu)

《TANET2019 臺灣網際網路研討會》 2019年 (2019/12) Pp. 92-96

https://doi.org/10.6924/TANET.201909.0016

摘要

因應全球氣候變遷劇烈變化，使得都市內積淹水現象災損嚴重。本研究提出基於智慧影像分析技術於深度學習架構進行自動化積淹水辨識模式及水位高度計算之研究。一般而言，當都市街道或道路發生間歇性暴雨，容易導致市區道路造成嚴重的積淹水事件。固然，現有低窪或是易發生災情地區皆裝有水位計或感測設備，但是對於廣泛部屬成本、設備妥善及人力維運在長時間作業環境下，皆是相當沉重的負擔。因此，本研究將整合影像分析與深度學習架構進行市區積淹水自動化物件偵測與水位高度計算，同時，設置不同淹水高度之層級警報，以有效提升防汛人員進行災情分析與決策之參考依據。影像分析驗證結果顯示，使用實際災情之道路積淹水影像進行系統測試與演算，可有效辨識水位高度並發布不同層級之警示。

關鍵字

監視系統 ； 影像分析 ； 深度學習 ； 物件偵測 ； 防災

並列摘要

In the serious global weather's situation, the urban areas have many puddles which are dangerous phenomena extremely. This research proposes a based on a smart image processing deep learning framework to execute the model of auto-identical detection and to calculate the water stage. In general, it is common puddled events that the intermittent rainstorm appears in urban streets and roads. However, the places which low and dangerous were set sensors and nilometers. But in some exceptions, the widely deployed areas are a heavy burden such as the cost, devices, and human resources. In addition, this research integrates the image processing and deep learning framework to execute the detection which the puddled objects in cities and calculating the water stage. Also, we prepare the alarm which can accord the water stage to use. This alarm can improve researchers' analysis and decision. According to image processing's result, this is a useful function to release alarm by testing and calculating the actual roads.

並列關鍵字

surveillance system ； image analysis ； deep learning ； object detection ； hazard mitigations