Research On Automatic Detection Algorithm Of Railway Obstacles Based On Image

As one of the most important transportation modes,while increasing the economic benefits of various countries,the safety of railway traffic has also become a global problem.According to statistical results,the main threats to railway safety are those different types of obstacles that appear in front of the train.At present,people deal with the problem of railway obstacles by using early detection and warning.However,the commonly used methods have the disadvantages of poor environment adaptability and weak anti-interference ability,which cannot meet the needs of the real work environment.Therefore,in order to assist the driver and minimize the railway traffic accidents caused by railway obstacles,based on the actual operation condition of train,this thesis analyzes and compares the common methods for detecting the obstacles in front of the train.An automatic detection algorithm for railway obstacles is designed,which consists of image acquisition,image preprocessing,track framework extraction,infrared and visible detection modules.The infrared imaging system only depend on the radiation characteristics of the object,and the visible imaging system can acquire rich information form the object such as texture,color and so on,which is good for detecting and positioning the obstacles.Because the actual operating environment of the train involves different time periods,different weather conditions and different seasons,in order to ensure the stability of obstacle detection system and its adaptability to different environments,infrared and visible imaging system are combined together to construct obstacles detection project.The project includes image acquisition,processing and obstacle detection.According to the specific environment and the advantages and disadvantages of infrared and visible imaging systems,infrared and visible imaging systems can be manually set to perform in a parallel or time-sharing manner.For the problem of image deterioration caused by smog,noise and other factors,two image preprocessing methods are used to improve the quality of images and facilitate the obstacles detection.Because the smoggy image has the characteristics of obviously low contrast,color deviation and so on.So,the improved dark channel algorithm based on guided filtering is used to deal with the haze images.And after analyzing the simulation and overhead time of common filtering algorithms,gaussian filtering is used to filter noises in images.The experimental results have affirmed that these two pre-processing algorithms are effective to the smoggy and noisy images.Considering with the diversity and randomness of rail in real environment,the method of segmented curve model is used to extract rails.It divides the entire image into two zones including long range and close range zones.In the zone of close range,the improved Hough transform is adopted to detect the rails.And in the zone of long range,firstly,we obtain the feature points of the rails by using sliding window,then rails are extracted by the method of hyperbolic model or cubic spline function with referencing the number of feature points.For the problem of detecting obstacles in different environments,the algorithms for infrared and visible obstacle detection are studied.The information of obstacle's shape,size,location and others are more significant in infrared images.So,the obstacle detection in infrared images is achieved by calculating the multi-directional gradients of image local areas.But for visible images,they include the rich information of obstacle,such as color,texture and so on.Therefore,the edge of the image is first used to coarsely locate the obstacle.Then,the RSS(Region of Stability and Saliency)method is used for fine positioning,which includes the stable zone and the salient zone extraction.Based on the established image database,the above methods are used to detect obstacles in different images,the simulation results have verified the validity of our algorithms and the rationality of the overall system.