Research Of Railway Defect Detection System Based On Embedded Machine Vision

In the railway detection, technological products not only improves the work efficiency, but also improves the quality of detection, including the detection of surface defects, but at present, the defect detection equipment is not mature enough especially in the high-speed detection field and very expensive, according to the current situation, putting forward a defect detection system based on embedded machine vision, the aim is to design a defect detection equipment that have high speed, stable performance.Through the study of the advantages and disadvantages of two kinds of image sensor CCD and CMOS using most widely today, CCD used as image sensor of the system. In the image processing module, based on the analysis of popular embedded technology, CPLD+DSP of the embedded system have combined.In the hardware design of the system, describing the realization of the function of each module of the system and the specific design scheme. Due to the special working mode of CCD sensor, the drive circuit use CPLD as the core processor, Conversion circuit designed about different drive pulse level; in the signal preprocessing module, the amplification filter circuit and digital-analog conversion circuit designed about the original acquisition signal, optimize circuit designed to eliminate the interference of the zero drift and temperature drift; before the signal enters the image processing module, designing a FIFO buffer device, the main purpose is to solve the problems that because the different speed of before and after modules lead to slow down the processing speed, this design alleviate the problem and greatly improve the working speed of processing module; In the image processing module with taking the DSP processor as the core, Data interface and power supply and reset circuits designed about DSP; In the data transmission module, Interface circuit designed about MAC layer and PHY layer of Gigabit Ethernet. In the driver design, the timing of each module is analyzed and using QUARTUS II for software simulation.In the rail defect detection experiment, through the research about the image processing algorithm, An image segmentation algorithm proposed, which combine an improved weighted median filter and improved Canny edge detection. Experiment indicates that it can extract the defect information clearly. Finally by the research of LVQ neural networks based on the feature vector of the orbital defect information complete accurate classification of defect information.The new design of the defect detection system of hardware and software reached the preliminary design goal: The system can collecte the detect object image, through the image processing, to realize accurate classification of defects.