Name: FLÁVIO MACHADO
Publication date: 08/10/2020
Advisor:
Name |
Role |
|---|---|
| RAQUEL FRIZERA VASSALLO | Advisor * |
Examining board:
| Name |
Role |
|---|---|
| DANIEL CRUZ CAVALIÉRI | External Examiner * |
| PATRICK MARQUES CIARELLI | Internal Examiner * |
| RAQUEL FRIZERA VASSALLO | Advisor * |
Summary: Draft survey is the process used to determine the amount of bulk cargo of a ship. Its main step is the draft reading, and although many have tried to automate this process in controlled conditions, it remains manual and labor-intense. This still happens because past automatic reading methods ignored non-ideal conditions, which are extremely common to the task, such as barnacles and sludge on the ships hull, weather conditions, along with
others. This paper presents a new complete pipeline for this process automation using two common deep learning architectures for the waterline identification and draft marks reading, and a new draft comprehension heuristic to unite these information. In contrast to other works, we compare these techniques to the state of the art and also captured a dataset with
various non-ideal conditions, which are common to the real scenarios, to train and test our solution. The results presented are promising, with the pipeline obtaining a median error of 7, 0 cm and an error inferior to 20 cm in 84% of the 800+ test images. These results are a step closer to the complete automatic draft reading in real situations, which will provide more security, agility and transparency to the execution of this task by those professionals.
