Subject

Drones equipped with high-resolution cameras often present a promising alternative to total stations for surveying railway infrastructure. They offer the ability to quickly acquire data over large areas. However, a major challenge remains: georeferencing UAV images with the accuracy required for (sub-)centimeter measurements.

Kind of work

In this thesis, the student examines the achievable accuracy for UAV photogrammetry. The images are aligned using RTK GPS and ground control points (GCPs). Prior datasets have been collected. The student will also explore new methods of fluently and automatically processing the raw images towards orthographic images and mesh models. Here software will be tested and automation of some steps will be performed, such as the detection of GCP points.

In a first phase the student will use existing datasets and research pipelines of processing. For this several software may be used, such as Metashape, Colmap, Alicevision and others.. Secondly the student will assess the achieved accuracy and processing time of his pipeline.

Finally the student may envise new methods of flying technique or placement of GCP to improve his pipeline. This may include following changes:

Use of a different drone (to be seen if feasible)

Flight speed

Flight patterns

Intersection angles

Quantity of GCP

Usage of id-encoded GCP

Framework of the Thesis

The main objectives of this thesis are:

Examine potential software and libraries towards building the pipeline:?It is important to analyse all possible software and libraries to get an idea of the building blocks for the pipeline.

Perform various tests on different software:?Accuracy is an important factor but also ease of use and fully automated processing.

Automation of some pipeline steps. Once a general workflow for processing is presented some steps may be automated. Main ideas here are research worthy objectives such as GCP detection or automated accuracy assessment.

Research potential on-site optimizations. Possible on-site optimizations for your pipeline? The student is free to explore this in discussion with his supervisors. Examples he can research may include the addition of encoded markers for GCPs or different flight methods/drone for increased accuracy or processing time.

Number of Students

1

Expected Student Profile

Basic understanding of railway infrastructure and components (e.g., signals, switches, panels).

Familiarity with 3D mapping techniques (e.g., photogrammetry, LiDAR) and geospatial software (e.g., QGIS, CloudCompare).

Knowledge of UAV operations and flight planning for data acquisition.

Programming skills in Python or MATLAB for algorithm development and optimization.

Analytical skills for data processing, accuracy assessment, and system validation.

This thesis will provide a comprehensive framework for leveraging drones and computer vision technologies to address railway-specific challenges, paving the way for more efficient and automated railway infrastructure monitoring and maintenance.

The scope of this thesis can be more focused on certain aspects in agreement with the student.