Check into anti-pattern and humanization techniques. Simple random checks and camera turns aren't enough to create the small randomization in human movements. That being said, fagex uses a combination of human checking combined with machine learning to determine human v bot behavior. I highly suspect that a machine learning algorithm is behind this, flagging accounts to be reviewed and banned. Potentially, it even bans accounts that it flags with a certain level of certainty (tin foil hats boys n girls). That being said, its better to find ways to make the mouse motion itself seem as human as possible, such as implementing planned randomization into specific actions and motions which would normally be much more exact in a script and less human.. You could also create a small scale neural network, from here you could train said network with a mixture of inputs both human and bot. By teaching it to differentiate between the two, you could then teach it to behave like a human and not a bot. However, this would take considerable resources, both time and asset wise.
Also, if you have a poor or rudimentary understanding of programming, ProjectPact actually has an excellent tool which he suggested above. I can attest that it does in fact have the majority if not all of the API capabilities. However, it will always fall short in its ability to be robust beyond the API being implemented directly into a script via an IDE. Through the this, you could gain quite a bit of use out of not only the OSBot API, but potentially others as well.