Feasibility Study on Real-time Image Recognition for Visual Safety of the Highway Crossing Clearance

• Main Authors: Chih-Pi Tsai, 蔡之弼
• Other Authors: 蒲永仁
• Format: Others
• Language: zh-TW
• Published: 2008
• Online Access: http://ndltd.ncl.edu.tw/handle/75473053251943651666

碩士 === 長榮大學 === 職業安全與衛生研究所 === 96 === This research studied the feasibility of the image recognition for visual safety of the highway crossing clearance by real-time analyzing the CCD images of the crossing. In detecting the existence of obstacles on the crossing, the safety of the engineers, passengers on the trains and the road users can be further assured. An accident of the highway crossings takes place when an approaching train collides with other intruding pedestrians or vehicles on the tracks of the crossing. The incident may cause many results of different extent including schedule behind, traffic jams, damage on the vehicles, injuries, fatalities or even derailment of the trains. The government and railroad companies all over the world pay much attention to the railway safety and take measures in many aspects to prevent any incident on the crossings. The accident rate is decreasing according to the statistics, but still far from the extermination that we are pursuing. An accident on a highway crossing often causes many casualties that leave miseries to their families and society. The cause of the incident is mostly on the fact that the emergency of vehicles stuck for some reason on the tracks of the crossing can not conveyed to the engineer of the train immediately. Until the engineer realizes the situation, everything is too late. In order to avoid the foregoing accidents we conducted a preliminary research on developing the visual safety system on the highway crossing. We studied the feasibility to recognize if the crossing was clear of obstacles through the processing of CCD images. We constructed a crossing miniature on which the simulations were conducted under various illumination and other influencing factors. These experiments were videotaped, digitized and analyzed subsequently through our image processing program in which the obstacles would be distinguished as possible by adjusting the thresholds of ROI image binarization as well as the number of varying pixel values. Both thresholds regarded as the software parameters were optimized to increase the ability of obstacle recognition and to reduce the possibility of mistakes. The results showed that the optimal threshold of the binarization was found to be 80 and the optimal one of the number of varying pixel values was 1500 to determine the presence of obstacles. Our visual safety program gets to the extent of real-time, stable and dependable and will be one of the most important bases in developing the future visual safety system.