Research On Detection And Tracking Method Of Correlation Filtering Railway Foreign Object Invasion Based On Depth Features

The rapid development and wide coverage of my country’s railways have brought challenges to the efficiency and safety of railway transportation.However,one of the main factors affecting the safety and efficiency of railway transportation is the frequent intrusion of foreign object such as pedestrians and light objects within the railway boundary.Our country mainly uses the form of fence to physically block the invading foreign object in order to obtain the protective effect.However,foreign object still often invade the railway boundaries,causing emergency trains to brake,disrupting the train line operation plan and delaying trains on related lines.At the same time,most of the railway lines are in a semi-enclosed environment,and the sky area on the railway becomes a blind zone for protection.The kites,plastic bags and other foreign object near the railway line often drift into the railway boundary or fall on the key parts of the catenary to damage the power supply device of the bow network,causing the train to stop and the related lines to be delayed.These incidents have a potential impact on the safety and efficiency of railway transportation.Therefore,it is a hot issue in the field of railway transportation to study an accurate and robust detection and tracking method for railway foreign object intrusion,and to ensure the safety and efficiency of railway transportation through intelligent video analysis technology.For this reason,this paper studies detection and tracking method of correlation filtering railway foreign object invasion based on depth features.The main research work of this paper in the detection and tracking stage of foreign object intrusion limit is as follows:1)The optimized pixel-level visual background extractor is used to detect invading foreign object.Among them,in order to improve the visibility of complex scenes within the railway boundary,the defogging process of the video sequence is integrated in the detection link.And a new background Local Ternary Pattern texture feature model is established in the visual background extractor.When the pre-segmentation result of the background color and Local Ternary Pattern texture feature model are both the foreground spot,the point is judged as the foreground spot.It can eliminate noise sensitivity and reduce the influence of illumination changes on the foreground segmentation of foreign object intrusion.2)Aiming at the problem that the railway foreign object intrusion tracking in foggy weather is easily affected by the image quality of the collected video sequence and the changes in the imaging scale of the intrusive foreign object,which makes the algorithm learn the wrong target information,a scale-adaptive foreign object intrusion defogging tracking algorithm is proposed.The algorithm integrates the defogging link to obtain the best fidelity adjustment factor suitable for special application scenarios in the railway integrated video surveillance system,and optimize the transmittance.And a new scale estimation link is added,and the scale pyramid is used to enrich the scale samples to complete the scale prediction of the intrusion foreign object.In the foggy weather,the accurate and effective tracking of invading foreign object with scale changes is realized.3)Afterwards,in order to solve the problem of tracking failure caused by background interference,rotation deformation or partial occlusion,a robust correlation filtering tracking algorithm based on depth feature decentralization and fusion is proposed.According to the complementary advantages of traditional features and depth features,the algorithm effectively fuses the multi-layer convolution feature decentralization with traditional features,optimizes the sample set and the number of iterations,and introduces a strategy of adaptive low-frequency update of the model.Effectively solve the problem that the tracking algorithm has weak anti-interference ability when foreign object invade within the railway boundary are affected by other interference.Finally,multiple sets of video sequences collected by a railway integrated video monitoring system on a certain railway test line are used to test the proposed algorithm.The experimental results show the effectiveness of defogging in the tracking,and show that the tracking accuracy and robustness of the proposed DFAL algorithm in the complex environment within the railway boundary reach 0.603 and 0.828.While ensuring the accuracy of the tracking algorithm,it further improves the robustness of the tracking algorithm.