A high-speed rail scene sensing method based on deep learning and structural information

Abstract

The invention discloses a high-speed rail scene perception method based on deep learning and structure information, and the method comprises the following steps: 1, obtaining track images, dividing the track images into a training set and a test set, and carrying out the marking of the images in the training set, so as to form a data set; 2, constructing an SSD network model, and constructing a loss function; 3, performing iterative training on the network obtained in the step 2 by adopting the data set formed in the step 1 to obtain a training model; 4, inputting a video needing to be detected and perceived into the training model obtained in the step 3 by frames, extracting features to obtain positions and category information of the fasteners and the shoulders, and distinguishing the turnout from the common rail according to the positions and the category information of the fasteners and the shoulders; 5,clustering position information of the positioning results in the step 4, and completing perception of the steel rail and the sleeper; rail parts in a turnout area can be detected and semantically segmented, the detection precision is high, and the detection speed is high.

Description

technical field [0001] The invention relates to an image processing-based track component detection method, in particular to a high-speed rail scene perception method based on deep learning and structural information. Background technique [0002] As an important support for social and economic development and an indispensable means of transportation in people's lives, railway transportation plays a pivotal role in the development of the whole society; especially for factors such as my country's vast territory, large population flow, and uneven distribution of resources; Railway transportation occupies an absolute advantage in all kinds of public transportation due to its advantages of large transportation capacity, low transportation cost, and relatively small area; The operating mileage will reach 30,000 kilometers, and the high-speed railway network will cover more than 80% of the big cities. [0003] The maintenance and testing of rail equipment is a concern in railway tr...

AI Technical Summary

Problems solved by technology

[0003] The maintenance and inspection of rail equipment is a concern in railway transportation technology, and it is also one of the projects with the most investment in research and development at the present stage; Among them, the high-speed rail track is mainly composed of fasteners, rails, sleepers, track slabs, etc. Traditional inspection methods It is completed by experienced road patrol workers. Workers regularly inspect along the line to find and report abnormal parts; manual inspection is labor-intensive and the working environment is very harsh. Vehicles driving on the line also pose a potential threat to the personal safety of inspection employees; traditional The railway inspection method has been difficult to meet the development needs of today's high-speed railway operation

[0004] With the development of machine vision, more and more attention has been paid to the image-based line visual inspection technology, but the traditional visual inspection method usually only detects the problem of a single component, and does not separate fasteners, rails and sleepers, etc. Perception; however, each track component is organically combined, which is very necessary for the perception of each track component; usually, the traditional visual inspection method uses template matching. At the turnout of the high-speed rail track, due to the different structure of the track components, Basically impossible to detect and perceive

Images