To solve detection-type problems, like those, you might eventually want to use some of the classical automatic statistical "pattern recognition" techniques. Basically, you get some samples of whatever types of data you have available that might be related to what you're trying to detect, then do what's called (or used to be called) "feature extraction", and then design a "discrimination function", which decides if detection has occurred or not. Then you can design a hardware/software implementation of the discrimination function and the needed sensors, to act as your "detector".
That stuff is a branch of what used to be called "artificial intelligence", back in the late 1970s. It was EE608 (IIRC) at Purdue, around 1979, with a book called "Statistical Pattern Recognition" (IIRC), by Ken Fukunaga. There were cool homework problems about things like designing the optimal system for identifying warheads and decoys, where the probability times the cost of being wrong comes into play.
Anyway, you could google for the terms in quotes, above. But you might also find most or all of what you currently need by searching for "voice recognition", and related topics. There are loads of IEEE papers about all of that stuff, and many books.
As cheap as hardware is getting, maybe later you can also use a digital camera and do image recognition. If it were me, I'd then add targeting, tracking, and automatic weapon-delivery.
Have fun!
- Tom Gootee
**broken link removed**