How many pixels are required for accurate license plate recognition?
Answered by ANPR AI·Sourced
Short answer
Installers frequently mount wide-angle cameras assuming 4K resolution guarantees readability, but OCR engines require a strict Pixels-Per-Meter (PPM) density. A standard 12-inch (0.5m) wide license plate must occupy an absolute minimum of 100 horizontal pixels across the image sensor to ensure reliable character segmentation.
Explanation
One of the most common installation failures in the ANPR industry is under-zooming the camera to capture a wide overall view of a parking lot. While this is standard practice for general surveillance, machine vision algorithms require a specific optical geometry. If a camera is deployed with a wide-angle focal length, the plate's pixel density drops below the OCR engine's operational threshold, causing the system to misread visually similar characters (like '8' and 'B') or drop edge characters entirely.
Industry standards dictate that a license plate must be represented by approximately 200 Pixels-Per-Meter (PPM). For a standard US or European license plate (~12 inches or 0.5 meters wide), this equates to a bare minimum of 100 horizontal pixels spanning the width of the plate in the video frame. A 4K camera mounted too far back with a wide lens will fail to meet this threshold, while a simple 1080p (2MP) camera tightly zoomed on a choke point will read flawlessly.
Things to check
- Take a snapshot from the camera, open it in an image editor (like MS Paint or Photoshop), and literally measure the width of a captured license plate in pixels.
- Check the horizontal field of view (FOV) at the exact point where vehicles pass; for a 2MP camera, the FOV should not exceed 15 to 20 feet wide.
- Ensure the camera's horizontal angle to the plate does not exceed 45 degrees, as extreme angles distort the pixel mapping.
What to do
To calculate your pixel density before deployment, divide the camera's horizontal resolution by its horizontal Field of View in feet (e.g., 1920 pixels / 15 feet = 128 pixels per foot). Since a standard plate is roughly 1 foot wide, this yields 128 pixels on the plate, which perfectly exceeds the 100-pixel minimum. If your calculation yields fewer than 100 pixels, you must physically adjust the varifocal lens to zoom tighter on the lane choke point, or move the camera physically closer to the target zone.
ANPR Edge
Skip the dedicated ANPR camera. Read plates from the cameras you already have.
ANPR Edge is plate-recognition software that runs on a PC at your site and works with ordinary IP cameras, so plate reads no longer depend on one vendor's camera, NVR or firmware.
- Works with standard IP cameras over RTSP
- One license per location covers every camera
- Runs on a Windows PC or Mac you own