Railway turnout recognition method and system based on track profile point cloud data

By using a recognition method based on track profile point cloud data, and employing a classification network and sliding window queue, the problem of low turnout recognition accuracy was solved, achieving high-precision and real-time turnout recognition.

CN118334428BActive Publication Date: 2026-06-23CHENGDU NAT RAILWAYS ELECTRICAL EQUIP
View PDF 1 Cites 0 Cited by

Patent Information

Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
CHENGDU NAT RAILWAYS ELECTRICAL EQUIP
Filing Date
2024-04-15
Publication Date
2026-06-23

AI Technical Summary

Technical Problem

Existing turnout identification methods suffer from several drawbacks: structured light and infrared light are greatly affected by the environment, depth resolution is limited, they are not sensitive to transparent objects, neural network classification models are not accurate enough, and the problem of left and right turnouts is not considered, which leads to a decrease in identification accuracy.

Method used

A recognition method based on track profile point cloud data is adopted. By establishing a classification network and a sliding window queue, combined with the ONNX model, real-time classification judgment is performed. The starting point of the turnout is accurately identified by using the sliding window queue and statistical analysis.

Benefits of technology

It achieves high-precision, real-time turnout identification, improves the accuracy and stability of identification, and provides an efficient means of track inspection and maintenance.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN118334428B_ABST
    Figure CN118334428B_ABST
Patent Text Reader

Abstract

The application discloses a railway turnout identification method based on track profile point cloud data, and comprises the following steps: establishing a classification network through an intelligent service module, classifying normal profiles and suspected turnout profiles of single-track point cloud profile graphs through the classification network, and obtaining a classification model file; exporting the required onnx model file through the intelligent service tool, calling the exported onnx model to process the collected track profile point cloud data in real time, and classifying and judging the turnout rail profile; classifying the rail profile of the point cloud data of the left and right tracks, if any side of the left and right tracks is a turnout, the profile is regarded as a turnout profile; establishing a sliding window queue, the rail profile classification result of each profile is added to the queue in real time, each queue of the sliding window queue is statistically analyzed, and when there are a set proportion of turnout profiles in the sliding window, the sliding window is a turnout; hysteresis processing: at the moment when the sliding window is judged as a turnout, a window length is rolled back to determine the starting point of the turnout.
Need to check novelty before this filing date? Find Prior Art