More broadly, the area of roadway infrastructure tech brings up a pet peeve of mine. The street I live on ends in a T-intersection with a traffic light onto a main thoroughfare in our mid-sized suburb. The cross-traffic on the main thoroughfare can be quite dense but only during peak commute hours. Understandably, the wait for the green arrow to turn left onto our street is quite long after triggering the in-road sensor because it's stopping a lot of traffic which they don't want to do too frequently. Late at night the wait timer is set much shorter since there's very little oncoming traffic. However, quite consistently, there are other periods with equally little oncoming traffic, such as a couple hours every weekday mid-morning and mid-afternoon.
This leads to residents in our neighborhood waiting three or more minutes to turn left in the middle of the day when there aren't even any oncoming cars to stop, and often no cars at all on the thoroughfare within half a mile. Regularly leaving turning cars idling for so long when no other cars are even in sight is environmentally wasteful as well as super annoying. I assume a Raspberry Pi-level SBC with a camera could easily make the traffic signal efficient for cars waiting to turn left without adding any delay for oncoming traffic over the current timer system (or digging up roadway for more sensors).
In this not-uncommon scenario, the "smart light" doesn't need to even be that smart or bullet-proof since it can always fall back to the basic timer if the situation isn't clear and can be gated at the other extreme by a max frequency limit for acting. Which is why I'm puzzled I don't see any such solutions deployed anywhere. It seems like a simple way to improve worthy metrics that's easy to deploy, low cost and has no downsides.