Research On Detection And Tracking Algorithm Of Intrusion Objects In Railway Surveillance Video Image

As more and more railway lines have been run in China,the safety of train operation has become the focus of research,and the detection of foreign object intrusion in railway perimeter even becomes the top priority in railway operation safety system.With the development of computer vision technique,it is of great practical significance to study the intelligent detection technology for intrusion objects in railway perimeter.Therefore,the related research was conducted in this paper,as shown below:(1)A dataset on foreign object intrusion in railway perimeter was established.Considering that there has not been any public dataset on railway foreign object intrusion detection at present,in this paper,the demand for railway foreign object intrusion detection was analyzed,and a dataset on railway foreign object intrusion images was established through the simulated intrusion experiment and web crawler.Additionally,an algorithm for railway perimeter extraction and imbalanced sample amplification based on Canny operator was proposed to amplify the imbalanced samples.After screening and sorting,a total of 5,226 images were acquired,totally 8,183 samples.The said dataset was divided into training set and test set for validating the algorithmic model proposed in this paper.(2)Aiming at the characteristics of objects in the dataset,YOLOX algorithm was improved,and a network framework based on the new-type bidirectional feature fusion and adaptive spatial feature fusion was proposed.As such,the small object detection accuracy was improved by a new bidirectional feature fusion improvement model,and the ability of extracting effective information from feature maps of different scales was enhanced through an adaptive spatial feature fusion enhancement model.Additionally,the location prediction was incorporated into the classification as a weight by virtue of VariFocal loss.The experiment reveals that the improved algorithm can effectively improve the object detection accuracy and achieve the tracking of video image objects.In this paper,the foreign object intrusion in railway perimeter was studied.First of all,related data were collected to establish a dataset suitable for deep learning.Subsequently,an object detection algorithm framework was put forward based on the characteristics of detection tasks and the dataset.Finally,an ablation experiment was performed on the established dataset.The results reveal that the proposed algorithmic model can achieve accurate detection and tracking of the intrusion objects.