Research On Detection Of Foreign Intrusion In Railway Perimeter Based On Surveillance Video

With the continuous increase of railway operating lines,the safe operation of trains has become increasingly important.If foreign objects invading the railway perimeter cannot be found in time,it will cause great economic losses and even hidden safety hazards.Therefore,the safety protection of the railway perimeter has always been valued by researchers.Taking into account the sparseness of the actual foreign body intrusion on the railway site,the improvement of video image resolution and the development of target detection technology,video-based detection technology has become the main research method due to its all-weather and high cost-effective characteristics.Therefore,relying on the on-site requirements for detection of intrusion targets in the railway perimeter,this paper carried out a research on the detection of foreign objects in the railway perimeter based on surveillance video.Firstly,a foreign body intrusion data set on the railway perimeter was constructed.Through the demand analysis of the foreign body intrusion data set on the railway perimeter,foreign body intrusion simulation experiments were performed in different railway scenes to obtain video image data.After data amplification,data classification and annotation A dataset of foreign body intrusion in the railway perimeter is established by other methods,which contains a total of 8000 images,which can be used for training of the improved YOLOv3 model in this paper.Secondly,this paper proposes a method for detecting foreign objects in the railway perimeter based on a dual detection algorithm.The overall idea of the model is: first use the Gaussian mixture model to screen the video for foreign objects,and if there are foreign objects,start the second method based on the improved YOLOv3 model.It does not start during the second detection,otherwise it will not start,which greatly reduces the computer consumption.Among them,in the initial screening of foreign objects,image enhancement methods combined with morphological processing and median filtering;in the secondary detection of foreign objects,the YOLOv3 model is improved by resetting the prior frame and changing the activation function.Finally,using the railway perimeter field data,the detection accuracy and detection speed are compared with other related target detection algorithms.The experimental results show that the double detection algorithm in this paper has fast detection speed,high accuracy and excellent performance.This article proposes a new railway foreign body intrusion detection method through innovative ideas.This method considers the sparseness of railway intrusion activities,weighs the relationship between detection accuracy and detection speed,and can not only inherit the advantages of the above algorithms but also weaken the disadvantages.This algorithm has high economic benefits and feasibility,and provides a new idea for the detection of foreign objects in the railway perimeter.