The accuracy of your videoNEXT License Plate Recognition analytic depends on a number of factors including the following:
2. Angles between the camera and the plate
3. Pixel counts on the target
4. Camera settings
If your system is within acceptable ranges on all of these factors, your system will deliver good accuracy. This note will address each of these factors.
In the image below, the license plate appears to be legible to a human. Probably you can read it as “GFP 3054.” The letters, however, are not sharply contrasted. Their blurry shades of gray blend into the plate background and other characters. Even with the videoNEXT LPR software module your computer system will struggle to accurately read a plate such as this.
In the next photo, the lighting was upgraded and the camera was zoomed in to put more pixels on the license plate. The same license plate is much better defined. Your LPR system will read this data with good accuracy.
Capturing a clear image of a license plate on a moving vehicle requires adequate lighting. Sometimes this requires the installation of an external illuminator. Capturing the license plate of a car moving at highway speed may require lighting that is equivalent to bright sunlight. Slowing moving cars, as in a parking facility, will operate with lower light levels.
Cameras automatically adjust their shutter speed based on the amount of available light. At high noon on a clear, sunny day, the shutter may only open for 1/10,000th of a second to capture a video frame. A slower shutter speed may result in over exposure. On a dark night, the shutter may remain open for an entire second to draw enough light.
The shutter speed directly correlates to the sharpness of the image of a moving vehicle. If the shutter opens for 1/10,000th of a second, a vehicle traveling 60 mph will move only 1/10th of an inch. This will result in virtually no motion blur on the license plate image. If the shutter opens for a full second, that same speed will result in a travel distance of 88 feet. The result will be blur that is unreadable – by computer or human.
Unfortunately, simply increasing the shutter speed will not improve results. If too little light is available, the image may appear completely black. Instead, you must increase the lighting.
If your camera is capable of infrared capture, you can use an infrared illuminator. Many cameras have built in-illuminators that are effective at considerable distances such as 150 feet. Infrared is invisible to the naked eye. Alternatively, you can install a spotlight.
Some cameras also include an IR cut filter which aid image capture at night. This is helpful when the vehicle image is captured from the front with very bright headlights pointed toward the camera. The filter will remove the headlights, leaving the dimmer license plate image clearly visible as in the image below.
When placing your camera, you should try to keep the line of sight perpendicular to the license plate in both the horizontal and vertical directions. Accuracy degrades significantly beyond a 30-degree horizontal or vertical angle.
In most cases, you will encounter fixed offsets in the horizontal or vertical directions. In order to obtain a reasonable angle, you will need to position the camera a considerable distance from the capture point. Remember the 30, 60, 90 rule from high school trigonometry – 1, 2, 3. A six-foot offset requires a distance of about 10 feet. More will not hurt.
You can select a vari-focal camera and position to camera to obtain a small angle. For example, pointing a zoomed-in camera parallel with the road will provide better results than aiming a wider-angle lens more perpendicularly.
In some cases, the angle of capture cannot be controlled. The videoNEXT LPR software model offers a feature that calibrates the camera to compensate for the angle of capture. Adjust the sliders until the rotations/angles show a more straight-on license plate in the image.
Pixels on Target
The number of pixels captured for each license plate is critical. A vari-focal lens on the camera will make it easy to optimize the field of view and the pixel count on the license plate.
The screenshots below compare a camera with a wide field of view to a camera zoomed to the area of the license plates. Both approaches will work because the pixel count is sufficient in each shot; however, plate recognition will be more accurate in the zoomed example.
Increasing the resolution of your camera will increase the pixel count on the license plate. However, increasing the pixel count will also increase the processing time. Therefore, if CPU resources are not infinite, increasing the resolution too much may decrease accuracy. We recommend setting the camera resolution no higher than 720p, making sure the plates still have enough pixels to be detected. If the camera is sufficiently zoomed, decreasing the resolution may, counterintuitively, improve accuracy.
For USA license plates, a minimum pixel count across the width of the plate of 75 (60 vertical) is required. For European license plates the minimum count is 90 (75 vertical). Counts above 250 will not noticeably improve accuracy.
Resolution/Frame Rate = 1080p (1920×1080) or 720p (1280×720) and 20 frames per second are a good starting point.
Compression = 20. A lower setting will produce better image quality.
Zipstream = Off
Camera Capture Mode
o WDR = Off. This feature adds too much noise to the image, which affects accuracy.
o Color level = 50
o Brightness = 50
o Sharpness = 60
o Contrast = 70
o Local Contrast = 55
o White balance = Automatic
o White balance window = Automatic
Wide Dynamic Range
o Enable Dynamic Contrast = No checkmark
o Exposure value = 70
o Exposure control = Automatic
o Maximum exposure Time =1/1000
o Enable backlight compensation = Unchecked
o Exposure zone = Auto
o Shutter = Fixed 1/2000 – for slow speed; 1/4000 for highway speed.
o Gain = Auto
o Max Gain = 24 db. Avoid excessive gain settings which will add noise to the image.
o Enable automatic iris adjustment = check
o IR cut filter = On (Day), Off (Night) if NOT using an Event Action rule below = Auto.
o Enable IR illumination = check
o Synchronize IR illumination with Day/Night = On if using Auto IR cut filter. Off if using an Event Action rule below.
Event Action Rules:
Enable Event Action Rules to ensure IR cut filter is off and built-in IR illuminator (if equipped) is active during the nighttime or when lighting conditions warrant its use. See the two Action Rule examples below:
Actions Day/Night Vision Mode – Sets Auto mode, day mode or night mode. When this mode is selected, the Action Rule will switch the IR cut filter to Auto mode (depending on lighting conditions), to Day mode (IR cut filter on) or to Night mode (IR cut filter off). Note that the IR cut filter setting in Video & Audio > Camera Settings page must not be set to Auto.
IR Illumination – Turns on the built-in Infrared (IR) light illuminator. It can be used in an Action Rule to turn on the IR light illuminator so that the camera can perform video surveillance in areas with low light, without requiring addition of external lighting.