Railway Clearance Intrusion Detection in Aerial Video Based on Convolutional Neural Network

摘要

Aiming at the problems of dynamic background and various types of objects in railway clearance intrusion detection in UAV aerial video, a railway clearance intrusion detection algorithm in aerial video based on convolutional neural network is proposed. Firstly, the rail track region is affirmed in aerial single frame image by the linear segmentation detection algorithm, line segments merging and line segments screening; Then, the improved convolution neural network model is used to detect and classify rail track region image in single frame image; Finally, the single frame detection result is optimized by the inter-frame correlation of the video to obtain the final result of the railway clearance intrusion detection in aerial video. Experiments on a self-built dataset show that the proposed method can effectively detect various types of objects in the aerial video.