I've got one of FLIR's cheaper thermal imagers and every time the issue of rain or snow comes up when discussing the limitations of self-driving cars I wonder w

I've got one of FLIR's cheaper thermal imagers and every time the issue of rain or snow comes up when discussing the limitations of self-driving cars I wonder why we hadn't already paired thermal sensors with machine learning. The thermal sensor might not tell you much in isolation - you can see what's warm in contrast to the surrounding environment, but moving objects give away a lot of information. A person moves a certain way and has a certain heat signature, as does a cat, another car, a person on a bicycle. I recall an interview with a (relatively) young James Cameron after he'd made the movie Aliens. He said that they spent a lot of time simplifying the Alien outfit, making it easy for performers wearing the costumes to move more freely, to help make their movement appear more alien or insect-like and less like a human struggling to stride in an outfit. To paraphrase him, the mind only needs a few pixels of movement information to understand what it's looking at. The loss of detail in the outfit was trivial compared to triggering the part of the human mind that associates certain movements with caution. An array of fairly humble low-ish resolution thermal sensors around the car paired with AI could be done with a cost in parts not much higher than USD 1k.