Research On The Techniques Of Recognition And Distance Of Railway Signals Based On Color Image Processing

Along with the speedy development of the domestic communications and transportation comes the steady mounting of the railway transportation capability, meanwhile, civilians also confront growing railway accident numbers. At the same time, Intelligent-monitoring techniques based on Image Processing can offer information relevant to the traffic conditions ahead and alarms to the engine driver as references.According to the features of all sorts of signals in real Railway lines, this paper researches into dozens of color space and techniques utilized to discern different colors and signal lights. Along railway lines, signals can be classified as:1) lights on the ground and 2) those attached to electrical poles. Railway signals are smaller than those on roads, hence the center area of the signal lights in a taken picture appears blended and blurred, induced by multiple lights surround it range from civilian housing lights to lights around the conjunction of a road and a railroad.Based on the description above, this paper adopted median filter technique during the preprocessing step, reserving most of the useful data while sifting out the noises. Through the RGB-to-HSV color space conversion, the independent component H resolves the difficulties of discerning the three basic colors. In the following step, via another area (signal lights area) extraction by the independent component S, the general area of the signal can be acquired. The next step, the final signal lights area can be acquired by setting the operators'filling, during the process of which isolated points are sifted out by mathematical morphological calculation, moreover whether or not the signals are those supposed to be processed should be testified, via the technique of Future Knowledge in Pattern Recognition. If the area is not the one supposed to be processed, ignore the result; if it is, then approximately, the distance between the locomotive and lights can be measured according to the binocular vision principle.Based on the result of signal recognition, several simple measures realize the distance-measuring are dwelled on in this paper with corresponding demonstrations and deductive equations. According to practical needs of certain applying circumstances, the paper made a comprehensive analysis of both the superiorities and limitations of several methods and possible improvements may add to the accuracy of them.Due to the fact that the system discussed in this paper will finally be implanted into a DSP, a real-time monitoring process should be concerned. This paper, in accord with actual circumstances:utilized operations such as Windows to process the part of the image where signals situate; chose processible images; and etc.