A lot would depend on your environment and what is being protected by the security cameras. For example, if you have multiple cameras and different filenames for each camera, then potentially, you could track the movements of someone in your environment. This could compromise other aspects of your security - for example, tell someone where the security patrol is or what path/pattern they follow and frequency of patrols etc. It may also be possible to perform some analysis of the file sizes to derive some other information as well - for example, perhaps number of people etc. One of the limitations of this type of metadata analysis is that you typically require quite large data sets in order to find meaningful metadata information. This often means considerable effort over an extended period of time. It also needs additional information about the environment i.e. number of cameras, site layout, file formats etc. Humans are also very creative and the right individual with the right mindset may find ways of deriving information which nobody else was able to conceive. This is why it is usually wise to limit the amount of information, meta or otherwise, attackers can access. The challenge is getting the balance right - at what point is the amount of effort used to hide this information more effort than the risk it is protecting against.
Then of course, there is the danger of actually cracking your gpg - while gpg is pretty good, it is also possible to either misconfigure it to use poor ciphers or too small a key etc. Sometimes, with applications like this, which may potentially need to encrypt large data sets frequently, options which increase encryption speed/efficiency are selected. In general, speed/efficiency and security are at opposite ends and if you don't get the balance right, you can compromise the security of the encrypted files. don't assume just because you are using gpg that your data is safe - it depends on what configuraiton options you are using.
Personally, I'd be more worried about the server compromise than the metadata such a compromise is able to reveal. Unless you are simply assessing the potential impact from a compromise, you need to consider what can be done to prevent the initial compromise first and then assess what measure would be approrpiate to mitigate the impact shold the compromise occur.
Having said all of this, I think it also needs to be highlighted that when it comes to physical security, more often than not, the real weakness is something far more simple. The effort and amount of information that can be gained indirectly via metadata usually takes considerable effort. You often require large amounts of input data to find useful patterns and it is generally a lot of work. This effort will only be worthwhile when the value of whatever is being protected is fairly high and even when it is, you will likely get a faster and less resource intensive result using simpler social engineering techniques or just finding overlooked physical security weaknesses.