Puffin signal crossings are based on sensors for pedestrian detection. What are the different type of senors/image processing tools which can be used to detect pedestrians at a crossing location for installing a low cost puffin signal?
Automated digital video recognition software. Reasonably common now in research and increasingly in consulting.
Yes, or thermal cameras. May sound strange, but they have a good reason - unlike the video cameras, privacy is not the issue :-)
See for example works of Nicolas Saunier and colleagues - http://n.saunier.free.fr/saunier/
Many techniques are developed but as I have read and heard none is perfect. The problems concerns detecting the right objects (persons not tree branches) and with isolating single objects (persons staying close are detected as one object)
For a good and low cost solution I would recommend using CNN based or HoG SVM based detectors. HoG SVM is included in openCV and will run on an embedded hardware (nvidia jetson) or low cost PC with graphicscard in real time. You will get a higher detection accuracy with a CNN, but that will be a little more work (fighting with Caffe / Tensorflow)