Valores para validar o algoritmo de descolamento

[pictos]

Olá a todos, vou falar em pt-BR, pois pelo que vi não tem nenhum estrangeiro - ainda - no projeto. Para validar o algoritmo de descolamento acho válido termos ao menos 3 posições conhecidas... No caso posição inicial (antes do acidente) e a posição após o acidente.

Como referência pensei em usar a posição do centro administrativo/refeitório e da pousada.
Posição inicial

• 20.1310227,-44.1236331 => escritório Vale
• 20.1395884,-44.1268049,681 => pousada arrastada (pousada Nova Estância)

Posição final
Infelizmente não consegui encontrar informações da posição final, caso alguém tenha será de grande ajuda... As informações divulgadas são muito vagas, como por exemplo "provavelmente foi arrastado pela lama e pode ter ido parar a "até quilômetros à frente"."

Espero que a ideia sirva para vocês.

[brunolcarli]

As suggested on #13

English:

To validate the detach algorithm I find valid terms at least 3 known positions ... In the case initial position (before the accident) and the position after the accident.

As a reference I thought of using the position of the administrative center / cafeteria and the inn.
Initial position

• 20.1310227,-44.1236331 => vale office

• 20.1395884, -44.1268049,681 => Traveled Hostel (Nova Estancia Hostel)

Final position
Unfortunately I could not find information on the final position, if anyone has it will be of great help ... The information disclosed is very vague, for example "probably was dragged by the mud and may have gone to" up to miles ahead. ""

I hope the idea works for you.

[panisson]

My suggestion if you want to build a solid system for giving possible coordinates for search teams is to give this high priority, since any model that might be proposed here would need a ground truth dataset to validate its performance.

I think you can build this dataset in two ways:

1. with the help of rescue teams, who can give the coordinates of known constructions/objects after the incident, and compare the coordinates with the same objects before the incident; or
2. using aerial images and crowdsourcing, where online volunteers can help to match in aerial maps objects that can be found before and after the incident.

Such ground truth data is imperative to estimate the right parameters for more complex predictive models, and to validate them.

[guidiego]

Hello guys! I love mathematics and physics and actually, I'm working with molecular modeling at a university. I think that first, we need to base our algorithm in a formula based on theoretical data like kinetic energy, velocity vectors, and density formulas after it we can apply data and improve the model in other layers after this base formula.

Right now I'm working on an issue to explain some ideas for this algorithm!

[pictos]

#1

[jsmatias]

As suggested on #13

English:

To validate the detach algorithm I find valid terms at least 3 known positions ... In the case initial position (before the accident) and the position after the accident.

As a reference I thought of using the position of the administrative center / cafeteria and the inn.
Initial position

• 20.1310227,-44.1236331 => vale office
• 20.1395884, -44.1268049,681 => Traveled Hostel (Nova Estancia Hostel)

Final position
Unfortunately I could not find information on the final position, if anyone has it will be of great help ... The information disclosed is very vague, for example "probably was dragged by the mud and may have gone to" up to miles ahead. ""

I hope the idea works for you.

Hello guys!
Some students from USP, ITA and me were working in a model based on Gradient descent for this problem.

Using these values above as initial position and an average mass equal to 80 kg, we got this result:

The map shows only the mud flow and after calculation based on drift force, we got a final position as being: (lat: -20.127900, lon: -44.120729). If one put it on google maps, he/she will get a weird result, however the initial segmentation of the map still has to be more fragmented to improve accuracy. Besides that, this is an initial model and shall be improved.
If you have any ideas, please have a look to review and validate it.

https://github.com/jsmatias/victims_location_prediction

[pictos]

@jsmatias thanks for your time. We working in your PR.