Research On Video Detection Of Intrusional Objects In Railway Perimeter

In recent years,Travelling safety of railway is also increasingly concerned by people as the railway transportation gains a much more rapid development.One of the important causes of railroad safety accidents is the invasion of foreign objects within the railroad perimeter.The traditional railroad perimeter protection system is constructed by setting barriers at the railroad perimeter or using human patrol.However,with the continuous growth of China’s railroad construction mileage,the above two methods are difficult to meet the requirements of people.In recent years,video surveillance technology is developing fast.Railroad perimeter foreign object detection method based on video surveillance technology has become a hot spot for research.However,the complexity and variability of the railroad environment is high.When foreign object intrusion occurs in the railroad perimeter,how to accurately detect the intrusion target is an urgent problem of ours.Therefore,this paper proposes a system for video-based foreign object intrusion detection for railroad perimeter.The system is created as follows.Firstly,delineate the railroad perimeter area;secondly,use the improved ViBe algorithm to extract the foreground motion target in the surveillance video and uses a feature fusion method based on YCbCr color features and CLBP texture features to remove shadows from the detected foreground target;finally,judge the intrusion target in the limited perimeter range and issue a warning when the intrusion target enters the limited area.The main research of this paper reflects the following aspects.(1)To improve the accuracy of the algorithm for foreground motion target extraction under the sudden change of illumination environment,in this paper,an improved ViBe algorithm for sudden changes in illumination is proposed.This algorithm detects illumination mutation frame in YCbCr color space and determines whether there is a foreground motion target in the current illumination mutation frame.Use different methods for foreground extraction according to the detection results.When the foreground target exists in the illumination mutation frame,the improved three-frame difference method is used.When there is no foreground target exists in the illumination mutation frame,the current frame is used for the background initialization of ViBe algorithm.The experiments show that the improved algorithm can suppress the "ghost" phenomenon of subsequent target extraction due to the existence of foreground targets in the current frame,which greatly improves the accuracy of the detection performance.(2)In order to solve the false detection phenomenon caused by shadows in the process of foreground motion target extraction,this paper proposes a feature fusion shadow removal method based on YCbCr color features and CLBP texture features based on the ViBe algorithm for foreground motion target extraction.The motion target region extracted by ViBe algorithm is considered as the shadow candidate region for shadow removal.The algorithm reduces the computational overload and improves the efficiency of detection.On the one hand,shadow regions are extracted in YCbCr space with fixed thresholds based on shadow color features in candidate shadow regions;on the other hand,CLBP texture feature extraction method based on shadow texture features in candidate shadow regions also is used to extract shadow regions again.The final shadow regions are produced by “AND” operation between two shadow regions detected by the above two methods.The experiments show that the shadow detection algorithm proposed in this paper can effectively detect and eliminate shadow regions while preserve the original motion target feature information with adaptation to the railroad complex environment.(3)In order to verify whether the extracted m otion target has encroachment behavior,firstly,the railroad perimeter area is limited by the manual marking method,which can adapt to the different requirements of area delineation in different railroad environments.The original image of the delimited area is converted into a binary image,where the pixels within the area are set to 1 and the pixels outside the area are set to 0.Secondly,the foreground motion target image using foreground detection and shadow removal operated with the binary image of the delimited perimeter area using “AND” operation to achieve the detection of foreign object intrusion in the perimeter area.Finally,the intrusion target is judged by the detection results.If there is an intrusion behavior,an alarm is raised.The algorithm is verified by several sets of video of railroad environment and the results show that the detection accuracy of the algorithm in this paper is more than 89%.