Almost certainly back license plates are easier since the lights on the back are less intense than the front. Of course, this is impacted by where plates are in one's region. Some regions only have front plates so you are forced to do that.
In our testing, avoiding aiming directly at the headlights can help (as it reduces the amount of direct light that needs to be overcome to capture the plate), though if you have strong enough IR and a fast enough shutter you should be able to get through that.
IPVMU Certified | 11/26/15 06:36am
Thanks for finding this information Undisclosed 1, even if we're not sure if it is true!
To the right or left, or above the Lane. Within 20-30 degrees of the center line of the car at the point you want to capture the plate is typical. The more pavement the camera sees in advance of a car coming through the better. This quiescent state is needed to minimize processing, and make a dramatic difference in the scene as the car passes through the focal point of the camera. Using an IR pass through filter, ample IR illumination, coupled with a LENS/FOCAL length that has a very short depth of field (IRIS FULLY OPEN) is the key to getting good results.
IPVMU Certified | 11/28/15 11:26am
Thanks very much Andrew. That's great information!
"Dramatic Difference" The LPR process will work best when it is idle as it looks at the pavement. Then when a car passes through it processes that scene looking for the shape of a plate and the edges of fonts.... Then back to looking at pavement. The fewer good frames the better.. (short focal length)
With pass through IR filters, its all the same.
Using IR pass through filters, and ample IR illumination, you're not likely to need much gain. However, every camera is going to behave differently, and you will not need day/night settings.
I can't share entire plate captures, but did we discuss the camera being in B/W Mode?
Since I am only experienced with Avigilon LPR, you mileage may vary... however,
I did a screen grab from a client of the first three digits from a plate that is really washed out, but all of the digits are reading correctly. Two images of same plate at different times are shown. The day images are almost identical with a deep IR pass through filter. Regarding the sample above, and reviewing some installations the horz. width and height of the plate is between 1/7th and 1/8th of the screen width / and height. I think these look good, and would expect good results.