This is called email remarketing: detecting web abandonment and remarketing via a different medium.
First, some technical pointers.
Trying to detect when a person has stopped looking at your site, if your analytics package is not continuously sending keep-alive pings (Munchkin doesn't send such pings), is not possible in a cross-browser fashion. In modern desktop browsers (Edge, up-to-date Chrome and Firefox) you can send an exit Munchkin event on navigation or tab/browser close or navigating away from the page with mostly-reliability. But this is impossible on older browsers (that includes IE 11) and on most mobile browsers.
In the absence of a literal exit event you have what we programmers call a dead man's switch problem. This refers to the impossibility of determining that a service has stopped doing something in the absence of some kind of external observer. When that "something" is "viewing web pages," it should make sense that the same device that stopped can't tell you that it stopped, depending on how gracefully it stopped. (Compare it to the difference between a train engineer radioing "I'm about to pass out" and just passing out.)
The fact that you can't reliably track the exit itself in the browser has led to the concept of "exit intent detection." This is tech used by scripts like OptIn Monster to pop up a form just before you close the tab, in an effort to stop you from leaving. Exit intent uses mouse movement, and in some cases previous navigation patterns and site-specific knowledge, to give good-to-fair results, again faltering more on mobile than on desktop.
But really you don't want exit/exit intent detection, you want abandonment detection, which is a different matter and is extremely site-specific. Abandonment can occur even when a page has focus, if enough time has passed. Conversely, abandonment may not occur even across entire browser sessions: you may not care about someone who visits one page and leaves (they're beneath the threshold for useful abandonment) but someone who visits 3 pages each from 2 different browser sessions without converting, or who leaves an empty cart for 24 hours, may become interesting.
The good news is that abandonment detection is easier, on a technical level, than exit detection, since it doesn't have to be done in real time (nor in the browser, if used for email remarketing) and is usually done by crunching activity logs offline to see if the page navigation/timing matches a known pattern.
The bad news is that Marketo isn't the right place to detect abandonment. As Devraj rightly notes, moving tons of leads in and out of campaigns based on activity log lookbehind could kill your instance, not to mention that effectively detecting abandonment requires a query language that's far more robust than Marketo Smart Lists.
Instead I would use a third-party analytics package, then query that package and call in to Marketo via API. Examples would be Calq or Heap Analytics. Alternately, you could download the Marketo activity log into an on-premises database, but that's going to represent a lot of infrastructure overhead before you even get to querying the data for abandonment patterns.
.png)