This is a real question someone asked me:
Could a metal detector detect bullets flying by?
This depends on the processing speed of the machine and the tuning. Nothing off the shelf would do this, and I wouldn’t think there is any motivation to build something of the sort.
Let’s take a detector tuned to reliably detect the mass of a lead bullet within a 3-ft sphere mid-air. Assuming the bullet passed through the full diameter of the sphere, it would be in range for around .001 seconds. That’s all the time the processor, software, sensors and such have to determine something is there – IF the system had full confidence of what it was seeing… and it won’t. Electromagnetism is not easy to measure. So…
Most metal detectors use a statistical model to determine that what they are “seeing” is not a false signal. In most practical settings (e.g. outside of a shielded lab.) there are many sources of EMF (electromagnetic interference) for any detector to deal with. So, to build a confidence model, the machine would need to re-process the signal and do the math many times before building confidence that the signal is not noise.
So, let’s be sloppy and say the machine needs to see the signal 50 times before it builds minimum confidence with some of that data coming on the leading edge of the bullet’s flight and some coming as it leaves the sphere. This means the processor must process the area 50 times our original estimate in the same period of 0.001 seconds, or .00002 seconds per “sweep” and plot it over a curve representing the distance to the center of the sensor. That’s not much time!
A signal model for a metal detector
The good news is that modern computer technology has plenty of power to do this if you built a machine with a “real-time operating system” (RTOS) that is free of the overhead of a general purpose computer. This could run sweeps until the bullet was out of range and then process the signal (clean up the data) immediately after. We’re talking trillions of processes per second. Yes, we could engineer a bigger machine that has a broader detection area, but you then also increase the chances for EMF. It also increases the processing time for all the signals to build the model.
Resolution climbs with processing power. But as processing demands increase so does cost, so this would, in my mind, be a very impractical setup. You would have to build a custom RTOS, hardware, sensors and the math itself. This involves lots of systems design, testing, lab time and more. And I assume it would need to be super durable since we can’t have a machine being killed by a bullet strike.
It could be a fun project but then the MIT guys would just hold up their trillion-frame-per-second digital camera and say “why not use this instead?”
For a deep dive into metal detecting sensor data and Geophysical Classification, check this out.
Did I screw up in the math? Additional ideas you’d like to offer? Please let me know!