|
Rail transportation in Taiwan has been developing over 100 years since 1887. About 600 to 800 train accidents, however, happened each year recently and resulted in about 80 to 100 casualties each year. Though the accidents and the deaths caused by them decrease year by year, the delay and the large amount of compensation to the passengers during the rush hours become more serious.
In the past, the rail tracks were laid down in the least populated area. With the rapid development of urban area nowadays, people gradually settle down along the railway and the intrusion of the tracks become inevitable. In order to reduce this risk and enhance the train speed, the construction of the elevated or the underground tracks in many cities has become a trend. Not only does it straighten the railway by cutting the curves, but it also reduces the railway crossings. Both the train speed and the safety are increased effectively.
Because of the steep topography in the eastern part of Taiwan, it makes the railway construction, including many tunnels and curves, tough and arduous. With the absence of an effective monitoring system of routes, all supervisions depend only on engineers' vision to evade obstacles on the track. The trains of Taiwan Railway Administration (TRA) can run currently up to 130 Km/hr, and it needs a distance of at least 600 meters to stop the train under some urgent situations. The delay of 1 to 2 seconds to put on the emergency brake caused by hindrance of eye sights before the railway curves may trigger very serious accidents.
In this thesis, we claim that an effective image recognition system which allows the engineers to predict the risks in front of the tracks is necessary and feasible. We investigate if the current image recognition system can be used in TRA and simulate a standard operating procedure of railway regulation after accidents have happened. We propose our model which can improve the operation efficiency and expect that it can enhance the safety in TRA.
|