Features cameras for license plate recognition

Hello my name is Anthony I am developing my college thesis with Nvidia Jetson embedded systems (Nano, TX2 and Xavier NX). My project consists in the development of a vehicle license plate detection and license plate character collection system. Please help me with the characteristics that must have a camera that works at a height of 5 meters and at a distance of 7 meters. I must mention that the system will work in real time all day and night.
I want to know what features a camera should have that will work with the needs I have mentioned. I would also appreciate if you could suggest a camera for the Jetson Nano, Jetson TX2 and Jetson Xavier NX embedded systems.

I have developed custom baseboards for Jetson modules for several camera projects.

Your main problem is low light. Your camera

  • must be able to use good lenses. C or CS mount for smaller cameras and something similar for larger ones. Think of DSLR cameras.
  • must have a decent sensor size. The Raspberry Pi camera modules have sensors that are only a few mm wide and high. That is not enough. You need something bigger. Watch out for the size of a single pixel.
  • must have Global Shutter so the whole frame is captured at a specific time. This is extremely important for capturing moving objects, which cars usually are. Most sensors have rolling shutter and capture imanges line by line, which can lead to distorted pictures on fast moving objects. You absolutely dont want this.
  • should be b/w with near IR sensitivity. You might need to consider infrared illumination for the night. 808nm is a common wavelength. Colour sensors have an IR blocking filter and are not useful for your application.
  • Pixel number/resolution is not a top priority. Don’t go for super high resolutions unless you need to. Sensors with high pixel count often have smaller pixels that capture less light.

My suggestion:

This is a PCIe camera. PCIe is the interface with the lowest latency and highest bandwidth into the system. Ximea has got excellent support for Jetson modules. This is a 4.2MP model. There are also versions with much higher pixel count (up to 150MP), but this also would mean much higher CPU/GPU requirements. Don’t go for more pixels than you absolutely need. This also saves you tons of money.


1 Like

Any monochrome camera can capture infrared light or do you have to look at something else?

Thank you for the previous answer, it has helped me.

Some monochrome cameras and all colour cameras do have IR blocking filters in order to match the wavelength sensitivity of the human eye. Even for monochrome cameras you have to ensure that there is no such IR blocking filter. The suggested camera has the keyword “NIR”. Or look in the datasheet. The suggested camera is sensitive up to 1000nm - visible light ranges from 400nm to 700nm.

PS: The suggested camera is also available with an USB3 interface. The USB3 models have a much higher latency and - worse - the data transfer causes a very noticable amount of CPU load. The PCIe version has almost no CPU load for the data transfer since the PCIe device can write the frames directly into main memory without using the CPU. Plus the PCIe interface has proven to be much more stable.

1 Like

This topic was automatically closed 14 days after the last reply. New replies are no longer allowed.