平交道事故中死傷人數最多之原因以擅自闖越平交道為主,但亦有因安全裝置故障或氣候不佳、視線不良等因素造成人員或車輛無故闖越並停靠於平交道上所造成。除交通意外之外,過去台鐵亦在鐵路平交道發生多起工安意外。 本研究之目的?d降低平交道事故的發生率,欲建立鐵路平交道淨空視覺安全系統,可對平交道上之障礙物進行辨識並發出警報供火車司機員暸解平交道即時狀況。研究之方法包含系統建置及系統偵測正確率,系統建置的部份將使用攝影機攝取平交道模型之影像,錄製成多段不同狀況的汽車擋道片段,經由影像處理程式,自動辨識障礙物,系統並進一步發出警示。系統偵測正確率將探討不同照度環境下比較系統偵測到車輛之時間間隔與車輛實際進入ROI之時間間隔比值。 研究結果建立鐵道模擬環境、影像處理模組,透過發送及接收系統之流程,可成功辨識出平交道之障礙物並予以示警,門檻值的設定與環境照度不同皆會影響障礙物偵測的正確率。以門檻值為80/1380的參數設定下,系統偵測正確率最高可達92%。
The incidents at the railway crossings have caused many casualties all over the world. The reasons include that the road users entered the railway crossings without authorization, the pavement conditions deteriorated, the safety installations failed, bad weather came in, and even poor views happened, which caused people or vehicles to get caught, stranded and hit inside the railway crossings. Besides traffic accidents, there were also occupational accidents at the railway crossings in Taiwan. The purpose of this research is to develop a machine visual safety system used at the railway crossings in order to decrease the accident rate therein. The system can detect obstacles at the railway crossing and send a warning and images to engineers who are then able to understand the front conditions immediately. The research method includes both system establishment and precision rate of detection. System establishment is the installation of a camera to take images at the crossing and the recording of them into clips in different situation. The system uses these clips subsequently to recognize obstacles by image processing algorithms. We will discuss in different illuminations, time interval of detection as compared with the actual time of a car entering the ROI. We have constructed a railway miniature which is integrated with the image process module. With the transmitting and receiving procedures our preliminary tests show that the system can detect obstacles successfully at the railway crossing and be able to send alarms to those in need. The precision rate of detection was affected by parameters setting in different illuminations. In the thresholds 80/1380 for setting parameters, the precision rate of detection can reach up to 92%.
為了持續優化網站功能與使用者體驗,本網站將Cookies分析技術用於網站營運、分析和個人化服務之目的。
若您繼續瀏覽本網站,即表示您同意本網站使用Cookies。