« Back to Projects

Introduction:

In railway engineering the development of connecting components between railway vehicles, comprises the determination of the performed relative motions between two adjacent carbodies during operation. Existing measurement systems have the drawback of being extensive and time consuming regarding installation, measurement and analysis. This thesis was concerned with the feasibility study and prototype development of a robust and cost-efficient image-based measurement system, capable of tracking the relative motions between railway vehicle carbodies. The proposed system contributes a novel, genuine alternative to conventional methods applied to the particular task of measuring the relative motions between two railway vehicle carbodies. It perfectly fulfils the technical and economic requirements.

Main components of railway vehicles:


blue: carbodies, red: bogies, green: rails


Problem statement:


During operation two adjacent carbodies oscillate and therefore perform relative motions against each other. These relative motions, with six-degree-of-freedom (6DoF = 3 translations, 3 rotations), should be determined continually.


Measurement setup:


The initial problem is solved by the estimation of the pose (6 DoF) of a rigidly installed camera in the first carbody relative to a scene coordinate system. The camera records imagery data during a test ride. Then the relative motions between two adjacent carbodies is determined from the movements of the targets (white boards with concentric circles) within the images.


Videos: short version

Results:

• Verification that system complies with required accuracy.
• Successful validation of feasibility of the developed system using the prototype implementation.
• Real world appliction of system during test ride on a high speed train at the Test and Validation Centre Wegberg-Wildenrath.
• Elaboration of a best practice approach for conducting on-site measurements.

Master’s Thesis:
File only contains the abstract and the contents because public access to the thesis is restricted.
RobertHoedl_2013_MastersThesis_Abstract.pdf  [.PDF | 280 K]


The Master’s Thesis was awarded the Siemens Railway Engineering Award 2014.


« Zurück zu Projekte

« Back to Projects

Introduction:
The attractiveness of fingerprints results from their uniqueness which does not change through the life of individuals. Within the past few years the increasing demand of security and safety has led to large scale application and deployment of automated fingerprint identification systems (AFIS). New laws and regulations (US-VISIT, EU-VIS) have been introduced. These new rules require the reading of all ten fingerprints instead of one or two. The motivation is the dramatic improvement of the false accept and false reject rates if more than one finger is used to reference an individual. These policies raise the problem that using market standard (single-) fingerprint readers would absolutely be not sufficient in terms of the amount of time needed for the enrollment process and the convenience of both, visitors and customs officials. To address this problem, ten-fingerprint scanners capable of recording multiple fingers at the same time, are deployed. Most of these scanners capture so called “slap images” which are “four-finger simultaneous plain impressions”. This type of images makes it necessary to separate the biometrical interesting fingertips out of the rest of the so called “slap image”. The whole process where the slap image is divided up into four images of the fingertips is called “slap fingerprint segmentation“.
The prototype which evolved from this project was ported and integrated into the final “Siemens Homeland Security Suite” product. Furthermore the algorithm was protected by a patent specification in 2009 at the European Patent Office. Moreover the method was presented at the CVWW 2009 (Computer Vision Winter Workshop) and published in its proceedings.

Examples: (Please click on the image…)
Different slap-images with automatically segmented fingertips (dark blue boxes).

Results:

• Port and integration of the algorithm into the Siemens Homeland Security Suite
• Publication in the proceedings of the CVWW 2009
• Patent specification granted by the European Patent Office (Patent Nr: EP 2131307 A1).

Publication in the proceedings of the CVWW 2009:
This paper contains an in-depth description of the algorithm as well as a detailled discussion of the evaluation results.
Robert_Hoedl_2009_CVWW09_Slap_Fingerprint_Segmentation.pdf  [.PDF | 968 K]

Poster CVWW 2009:
Robert_Hoedl_2009_CVWW09_Poster.pdf  [.PDF | 238 K]

European patent specification:
EP2131307_patent.pdf  [.PDF | 174 K]

« Back to Projects

« Back to Projects

Introduction:
Today more than ever, reliability, convenience and speed are important factors for the commercial success of modern railways. To meet these requirements more and more automated railway monitoring systems have been deployed in the recent years. Among other things such systems visually capture entire freight trains while passing a terminal in order to identify the single railroad cars. Up to now within Europe this identication has to rely on visual information since the railroad car number according to the RIV (Regolamento Internazionale dei Veicoli) conventions is the only way to identify railroad cars operated by ‘foreign’ carriers.
In order to capture the visual information of the freight trains, a setup consisting of a line scan camera, articial lighting and inductive sensors was installed along the train line. As a preprocessing step for the vision-based identification it is necessary to divide the recorded continuous train image into the single railroad cars. So far the separation of the railroad cars is done using inductive sensors embedded in the railroad tracks which determine the axle patterns of the traversing railroad cars. These axle patterns are then queried and assigned to a specific car type which enables the system to estimate the bounds of a single railroad car. Since the implementation of inductive sensors is rather expensive and the axle pattern estimation of the bounds of the railroad cars is sometimes imprecise a solution based only on the captured visual information was needed and found!
The prototype which evolved from this project performed exceptionally well in the Railroad car separation from continuous train images.

Example: (Please click on the images…)
Original image taken from the “02_Graz_Tag” dataset.

Result image containing the automatically detected separation points (dark blue) as well as the retrieved 1D signal (bottom).

Results:
A prototype written in Matlab as well as a paper which serves as documentation.


Paper:
This paper contains an in-depth description of the algorithm as well as a detailled discussion of the evaluation results.
Robert_Hoedl_2010_HP-Railroad_Car_Separation.pdf   [.PDF | 960 K]

« Back to Projects