shin_getter
ACCESS: Top Secret
- Joined
- 1 June 2019
- Messages
- 941
- Reaction score
- 1,198
Historically camouflage is designed around human vision. With machine learning and integration of electronic sensors to many aspect of life, computer vision is the new constraint to be built around.
With computer vision, I think it is possible to have a camouflage patterns that:
1. Is an effective camouflage to the "naive" observer without data on the camouflage pattern algorithm or at least a large, high res data set of the camouflage patterns.
2. Also contains regularities that can be detected by computer vision if one has the camouflage pattern generation algorithm, for identification.
Or more extreme still, each individual camouflage patterns gets transformed into a specific identification algorithm (by training an ML algorithm on the pattern within the spanning set of environmental conditions), so upon observation you can identify not only which side the camo pattern applies to, but which exact vehicle it is. It is not too much memory to save the camo pattern of every vehicle on your side.
Now there is a lot of information security considerations to take into account here, and it would be like codebreaking vs crypto in previous wars. There is much to consider like what happens if a vehicle database is captured, what happened if you suspect a leak, what happens if opponent has obtain goo visual data of a subset of patterns, what happens when field modification of vehicles is to be done.
The integration of computer vision, AI, camouflage, information security, IFF just seems self evident, if to find optimal designs with different trade-offs for different applications. (processing, camouflage effectiveness, IFF effectiveness, code-break feasibility and so on) I am not seeing too much literature on the integrated system of all the parts, have people just fail to imagine this or its secret?
With computer vision, I think it is possible to have a camouflage patterns that:
1. Is an effective camouflage to the "naive" observer without data on the camouflage pattern algorithm or at least a large, high res data set of the camouflage patterns.
2. Also contains regularities that can be detected by computer vision if one has the camouflage pattern generation algorithm, for identification.
Or more extreme still, each individual camouflage patterns gets transformed into a specific identification algorithm (by training an ML algorithm on the pattern within the spanning set of environmental conditions), so upon observation you can identify not only which side the camo pattern applies to, but which exact vehicle it is. It is not too much memory to save the camo pattern of every vehicle on your side.
Now there is a lot of information security considerations to take into account here, and it would be like codebreaking vs crypto in previous wars. There is much to consider like what happens if a vehicle database is captured, what happened if you suspect a leak, what happens if opponent has obtain goo visual data of a subset of patterns, what happens when field modification of vehicles is to be done.
The integration of computer vision, AI, camouflage, information security, IFF just seems self evident, if to find optimal designs with different trade-offs for different applications. (processing, camouflage effectiveness, IFF effectiveness, code-break feasibility and so on) I am not seeing too much literature on the integrated system of all the parts, have people just fail to imagine this or its secret?
