Nombre: ANDRÉ STANZANI FRANCA
Fecha de publicación: 29/05/2017
Supervisor:
Nombre |
Papel |
|---|---|
| RAQUEL FRIZERA VASSALLO | Advisor * |
Junta de examinadores:
| Nombre |
Papel |
|---|---|
| FLAVIO GARCIA PEREIRA | External Examiner * |
| PATRICK MARQUES CIARELLI | Internal Examiner * |
| RAQUEL FRIZERA VASSALLO | Advisor * |
Sumario: The railway is an important engine of the world economy. This means of transport more than 200 years old is efficient, safe and of great capacity and speed but still suffers from the difficulties of maintenance, mainly due to its assets of great extent, quantity, weight and geographic dispersion. In view of this, some initiatives in automatic inspection of railway assets have been developing. In particular, the inspection of railway sleepers (railway ties), which is sometimes done manually, needs development and consolidation. This dissertation presents a method for inventorying, identifying the type and some defects in sleepers based on image processing, heuristics and feature fusion in an unsupervised way. It uses Haar transform and integral images, as well as
other image processing techniques such as edge detection and entropy
computation along with aspects of railroad topology to achieve the proposed objectives. The algorithm was developed using real images of daily railway, previously unclassified, and that were subject to various noises and variations of a real railway operation. The method was validated through experiments with an image set that have approximately 33,000 sleepers. The results are promising and reach 97% accuracy in identifying the type of sleepers and reach 93% accuracy in identifying visible defects in sleepers.
