Calibration
As with any LPR sensor, proper post-installation camera calibration is essential to ensure optimum recognition rates, accuracy, and system performance.
Last updated
As with any LPR sensor, proper post-installation camera calibration is essential to ensure optimum recognition rates, accuracy, and system performance.
Last updated
Platforms
Rekor Discover™ for Urban MobilityRekor Command™ for TransportationRekor Scout® for Public Safety© 2024 Rekor Systems, Inc. All Rights Reserved.
This section of the Rekor Edge Pro Solutioning and Calibration Guide discusses the basic concepts of IP camera surveillance and LPR capture, as well as best practices to obtain an optimal capture scenario, field of view, and scene imaging. Understanding what factors can impact your scene, and how they should be mitigated or adjusted out of the image is a key first step in obtaining the best read rate and accuracy in your LPR deployments. If you are already familiar with LPR and IP camera concepts, skip to section 3.3 for baseline calibration values. An improperly calibrated camera can result in reduced read and detection accuracy, and can also impact vehicle identification rates, read or capture rates, and surveillance effectiveness.
When the Rekor Edge Pro is performing LPR capture at a location, at its most basic level the Rekor Scout Agent AI engine is “watching” a video stream. In order for it to process that stream and conduct LPR operations, it is ideal that the field of view contains the entire vehicle being captured for at least one second. The vehicle should be fully in the frame with some margin on all sides to accommodate larger commercial vehicles, with the license plate facing the camera and clearly visible. The license plate should also have a sufficiently high resolution at the point of aim (called “pixel density”) to ensure it is easily human-readable. There are established standards to define the minimum pixel density needed, which will be discussed later in this chapter.
Importantly, the LPR engine uses internal “confidence scores” to determine the best event data and to discard bad data. Proper camera location, scene setup, and calibration are all essential components to ensuring the engine generates correct and high confidence scores for valid vehicle and plate reads, and correct and low confidence scores for objects that should not be read.
When calibrating the Rekor Edge Pro camera, the technician must adjust the camera’s location, aim, field of view, and internal camera settings to obtain the best output in a number of key metrics. Pixel density, look angle, framing, focus, brightness, contrast, and color balance are all vital factors in obtaining the highest levels of accuracy when conducting the identification of vehicle make, model, color, plate, and state.
It’s important to note that only some of these factors are directly correctable by camera settings. Many of them are highly dependent on proper location selection, camera mounting, and aiming. Even those metrics that can be corrected with camera settings may require a combination of settings to obtain the ideal result and will require careful settings balancing to ensure that you’re not detracting from one metric to improve another.
Image resolution is the most important factor in obtaining high-accuracy and reliable LPR capture. Within a camera’s field of view, a specific object’s image resolution is measured in “pixel density”. Pixel density is defined as the number of horizontal pixels that are contained within a physical measurement of length at a specific point in the actual camera field of view. In the US, pixel density is most often measured in Pixels Per Foot (PPF). In regions that use the metric system, it is often expressed in Pixels Per Meter (PPM).
The most straightforward method to calculating pixel density is to first determine your ideal point of aim (the exact location along the roadway where you want the vehicle and license plate to be captured). Set the camera’s aim and zoom to center that location in the field of view. Once you have the image framing that you desire, physically measure the scene width at the point of aim from the far left edge of what’s visible in the camera’s field of view to the far right edge of what’s visible to the camera.
With that measurement identified, divide the horizontal resolution of your camera by the physical width of the field of view at the point of aim. For example, if you have a 1080p camera (resolution is 1920x1080 pixels), and the scene width is 18 feet, your pixel density calculation would be:
1920 ÷ 18 = 106.67
~107 PPF (Pixels Per Foot)
In the US, you can also take a still frame image of a vehicle at the point of aim, and physically count the pixels across the width of the license plate in the image. This is because US license plates are exactly 12” across. Counting pixel density in this way does require a different set of steps and software applications.
The minimum pixel density required for specific use cases is outlined through an industry standard called the DORI model. DORI stands for Detection, Observation, Recognition, and Identification. Each progressive step in the DORI model calls for a progressively higher minimum resolution to successfully fulfil the indicated mission.
In general, the highest standard on the DORI model (Identification) reflects the minimum resolution required for the successful identification of standard large-letter license plates. The Identification standard calls for a minimum pixel density of 70 PPF. The modern LPR mission, however, goes far beyond the simple identification of large letters on license plates. Advanced LPR systems are now expected to accurately identify stacked and small letters, and state of origin, and to provide vehicle make and model detection. Through our testing, Rekor has found that the minimum acceptable resolution for optimum potential accuracy for all these missions is 120 PPF.
Below is an example license plate displayed at a range of pixel densities from 50 PPF through 150 PPF. Note how as the pixel densities are reduced, it becomes more difficult to distinguish smaller details, such as stacked letters, state of origin identifiers, and differences between certain similar characters.
It is important to note that Rekor’s accuracy observations are based on properly located, aimed, and calibrated Rekor Edge Pro camera systems with a point-of-aim pixel density exceeding 120 PPF. This corresponds to a scene width at the point of aim of 16 feet (1920 ÷ 16 = 120).
In LPR, the “look angle” is the angle of approach between the camera itself and a license plate at the point of aim. More specifically, it is the angle between a line drawn from the license plate up-road parallel to the vehicle’s path of motion, and a line drawn directly between the vehicle’s license plate and the camera itself. It is depicted as angle ‘a’ in the attached image.
Look angles are calculated for both horizontal and vertical approaches to the license plate. Vertical look angles are calculated similarly, but measure the impact of camera height rather than camera offset. Rekor specifies a maximum look angle of 30° both vertically and horizontally for optimal day/night
performance. This is particularly relevant for night-time plate read accuracy.
Angle ‘a’ depicts the horizontal "look angle"
Note that both vertical and horizontal look angles are calculated to the license plate, not some other area on a captured vehicle, or the roadway surface. When calculating horizontal and vertical look angles, keep in mind that license plates are most often mounted between 18” and 30” above the ground, and centered on the vehicle.
The look angle of a camera scene affects LPR performance in two important ways: relative pixel density losses through perspective skew, and infrared reflectivity off retroreflective surfaces during night reads.
In basic terms, the sharper the angle to the vehicle license plate, the more ‘skewed’ the license plate will appear. If you turned a license plate 90 degrees from a camera, so you were looking at it on edge, the width of the plate in the field of view would be essentially zero. As that angle decreases toward a direct, 0-degree look angle, the plate width in the field of view increases toward the full 12” width of a US license plate. As the license plate width in the field of view reduces from 12”, your pixel density also reduces in proportion. You also risk visual license plate obstruction from vehicle features such as bumpers and trunk lids as the look angle increases.
Turning vehicles, such as to or from side streets in the camera’s field of view, or around a sharp curve, can introduce dynamic look angles into the scene that can also reduce LPR performance, so camera placement to obtain a static look angle throughout the field of view is another important factor.
Night-time performance is particularly affected by look angle, in that retroreflective surfaces, such as license plate coatings, may have specific reflectivity cutoffs beyond certain angles of approach. In our testing, many states have recently introduced new license plate designs that use a novel retroreflective coating that has a sharp performance drop-off beyond 30°, both horizontally and vertically. That retroreflective coating is essential to an LPR system’s ability to see the contrast between the plate background and plate characters at night during infrared capture, particularly when there is no or limited ambient illumination.
Strong reflectivity based on look angle, combined with correct infrared illuminator strength settings and an accurate night-time focus result in an image that has strong contrast between characters and background, and a crisp, readable plate image.
When calibrating your Rekor Edge Pro for a particular location, it is very helpful to park a vehicle so the rear license plate is at the desired point of aim. If this is not possible, Rekor strongly suggests placing a license plate on a tripod in the center of the roadway at that point. The tripod should be configured to hold the license plate 18” to 30” above the roadway (approximately knee to thigh height), perpendicular to the direction of vehicle travel, and either completely vertical or slightly leaning ‘back’ so the top of the plate is leaning no more than 5° into the direction of travel. This will closely approximate the plate mounting conditions for most vehicles.
Center the license plate in the Rekor Edge Pro’s field of view by manually aiming the camera, then zoom the image in or out until the entire vehicle is in the field of view, with a margin on all sides for vehicle travel and the capture of larger vehicles. Once the license plate is centered and the zoom is set, measure your scene width at the license plate to verify that your pixel density is at least 120 PPF. If the pixel density is below 120 PPF, zoom the camera in until the correct field of view is obtained.
An out-of-focus image can also significantly reduce plate detection and recognition accuracy. In these terms, “focus” refers to the sharpness of the license plate image at the time of capture. While the focus of the camera itself is the primary factor affecting the clarity of the image, it is also impacted by the depth of field and shutter speed of the camera.
Primarily, the camera should be focused with a vehicle or license plate placed statically in the center of the frame and the aim and zoom configured for the correct field of view and pixel density, as previously discussed. The focus setting should always be the final change made prior to saving your camera settings. It’s important to note that the Rekor Edge Pro maintains separate focus settings for day and night modes, and must be focused separately for each mode. Day mode settings, including focus, should always be adjusted with a live image during a typical daytime period, and night mode should always be configured with a live image after full dark. After you make any changes to any other camera image settings, you should always re-set the focus before saving your settings.
The Rekor Edge Pro camera uses an optical zoom lens assembly with an adjustable aperture. The aperture size has a direct impact on the depth of field of the scene. Depth of field is the range of distance from the camera where an object will appear completely in focus. Nearer than the target depth of field, and beyond it, objects will appear out of focus. The wider the aperture, the more light it lets in, and the narrower the depth of field will be. That means a wide aperture will make the scene brighter, but will have a very small ‘target window’ through which license plates will appear in focus. In turn, a narrow aperture, while making the scene darker, will have a much larger ‘target window’. The faster the vehicle traffic is moving, the wider the ‘target window’ or ‘depth of field’ must be to ensure a properly focused read event.
Similarly, shutter speed also has an impact on focus in the form of motion blur on moving objects. Shutter speed is measured in fractions of a second. The faster the shutter speed, the less light is let through to the sensor, but the less motion blur an image will have. For example, a camera with a 1/500 shutter speed will record an “average” image over the course of 1/500th of a second for each video frame. A vehicle moving at 45 MPH will travel roughly 1.5” over the course of 1/500th of a second, so the resulting frame at the camera will show that the vehicle and the license plate characters blurred over 1.5” of motion. This is significant when you consider the average license plate character is only about 1” wide. Rekor recommends a 1/1000 to 1/2000 shutter speed to minimize the effects of motion blur on the LPR imaging.
Brightness is essentially a measurement of the average shade of luminance (called ‘gamma’) of a scene. A higher brightness will lighten both light and dark shades. Too high of a brightness will ‘compress’ light areas such that they lose contrast and clarity, and too low of brightness will ‘compress’ dark areas in the same way. Brightness and contrast must be balanced such that the vehicle is clearly visible and the plate is bright and clear, with sharp contrast. The ideal scene brightness will provide a balanced image with a bright and visible license plate, and will also deliver the largest native contrast between the plate characters and the plate background.
On the Rekor Edge Pro, the Gain setting is the primary method of setting brightness, though the Aperture and Exposure Rate (shutter speed) will also affect it. Remember that the Exposure Rate and Aperture settings also impact motion blur and depth of field, respectively, and a balance of settings is preferential to using any one setting to achieve the majority of a Brightness or Contrast correction.
Where brightness is the overall lightness or darkness of the scene as a whole, contrast is the difference in brightness between the light and dark areas of a scene. A higher contrast will result in more differences; the bright areas will be brighter, and the dark areas will be darker. A contrast that is too high can result in the license plate background ‘blooming’ over and disfiguring the license plate characters to the camera, and will also make the vehicle harder to identify at night. A contrast that is too low can result in license plate characters that do not stand out enough from the plate background, which can result in missed plates and poor accuracy.
On the Rekor Edge Pro, the Exposure Rate is the primary method of affecting contrast during the day, though the Aperture and Gain settings will also have some effect. Remember that the Exposure Rate and Aperture settings also impact depth of field and motion blur, and a balance of settings is preferential to using any one setting to achieve the majority of a Brightness or Contrast correction.
During nighttime capture, the Infrared Intensity setting will have the most significant impact on contrast, particularly on retroreflective surfaces like license plates and tail light reflectors. Too much Infrared Intensity will result in “blank” or “pure white” plate images, often with very bright reflectivity from the vehicle’s tail light reflectors, and can also cause the rest of the scene to appear particularly dark with very little dynamic range. Too little infrared intensity will result in license plates appearing “dim” with very little contrast between the plate characters, plate background, and the vehicle itself. An ideal, balanced Infrared Intensity value will provide a bright, crisp, and clear plate image with the maximum contrast between plate characters and plate background, and the clearest, most well-defined plate characters.
Properly setting the Infrared Intensity values is extraordinarily difficult without a vehicle or plate being placed statically at the point of aim during adjustment. Note that a correctly adjusted Infrared Intensity value might not deliver a bright overall vehicle image at night without adequate ambient illumination. Plate and plate character clarity is more important for nighttime capture than overall scene brightness.
Color balance refers to how accurately colors are displayed in the camera image, and has a significant effect on an LPR system’s vehicle color determination. The most common detractors to a proper color balance are sources of artificial illumination in low-light situations. For example, some street lights, particularly older designs, use sodium-vapor light modules that cast light in a very distinctive orange-yellow hue. Sodium lights can make white, silver, or light grey vehicles appear yellow, orange, or even brown in color, and can make blue vehicles appear green, among others. Location selection is the most important factor when it comes to the white balance of an LPR scene. Avoid locations with heavily tinted or unnaturally shaded artificial illumination.
Note that during night operation, the Rekor Edge Pro uses an infrared filter and infrared illuminator to obtain scene imagery. Infrared is interpreted by the camera image sensor as a level of intensity of infrared reflectivity; more reflectivity is brighter, and less is darker. Object colors when viewed in the visible spectrum do not correlate to their shade when viewed under infrared, and no LPR system will provide visible-spectrum color determinations during infrared operation.
Environmental factors play a significant role in proper camera calibration. For example, in scenes that have dynamic lighting (either frequent lighting changes from artificial sources or areas where the amount of sunlight versus shade varies drastically from morning to evening), it is important to ensure your calibration values are balanced to provide clear license plate imaging with good contrast in all real-world lighting conditions at that location. Capture can also be impacted at night by obstructive light sources, such as low-lying security lights or street lights whose emitters are in the field of view, or a high volume of oncoming traffic, particularly if the roadway is curved or crowned such that the headlights aim directly into the camera lens. Additionally, reflective objects such as roadside reflectors and reflective highway signage can cause an excess of infrared blowback and make calibration of the Infrared Intensity value difficult to balance.
Proper or improper location selection, and the associated environmental factors, can significantly impact the terminal performance potential of any LPR system. Visual obstructions, such as trees, bushes, and infrastructure such as gantries, poles, and buildings will limit vehicle and plate visibility to the Rekor Scout Agent, which can reduce read / capture rate and accuracy.
The best way to mitigate environmental and calibration challenges is proper site selection. If a camera cannot be relocated, changing the height, zoom, and aim of the camera (within the recommended ranges), or adjusting the camera hood, may allow the user to adjust the field of view such that the environmental challenge is no longer visible in the scene.
It’s important to keep in mind that landscaping does change size and shape over time, so just because a young tree or bush does not obstruct the view of the target location when installed, or during the winter when tree branches are bare, does not mean it will not become an issue in the future. If a tree or other landscaping object needs to be trimmed during the installation to obtain a proper field of view, it will likely need to be maintained seasonally to ensure the field of view does not become impacted in the future.
Note that prior to finalizing a site location and installing the camera, Rekor recommends connecting each Rekor Edge Pro to a network connection in a controlled environment to validate camera functionality and to configure basic camera settings such as network interface and user values.
NOTE: The calibration values listed in this section are recommended baseline settings only. The Rekor Edge Pro camera will need to be calibrated specifically for each individual deployment environment, both during the day and night, to balance brightness, contrast, and infrared intensity at night as described in this document.
Gain
40%
Note: Rekor does not recommend exceeding 40%
Aperture
40%
Note: Rekor does not recommend exceeding 40%
Exposure Rate
1/1600
Note: Set between 1/1000 and 1/2000
Infrared Intensity
0%
Gain
40%
Note: Rekor does not recommend exceeding 40%
Aperture
40%
Note: Rekor does not recommend exceeding 40%
Exposure Rate
1/1600
Note: Set between 1/1000 and 1/2000
Infrared Intensity
50%
Note: Typically will be set between 40% and 70%
The user should reference the Rekor Edge Pro User Guide for operation instructions on logging into and using the Rekor Edge Pro user interface. Once logged in and viewing the Camera Configuration screen, proceed with the settings configuration in the following order:
Click “Stop LPR” to stop the LPR service.
Note: Camera settings cannot be configured or stored while the LPR service is running.
Select “Day Time Mode” or “Night Time Mode” to set the mode you are configuring. Note: Each mode must be fully configured and stored independently, including focus. “Night Time Mode” must be configured at night after full dark.
Adjust the “Gain” value. Note: A higher percentage results in a brighter image. Rekor recommends avoiding exceeding 40% in most scenarios to prevent image graining and poor performance.
Adjust the “Aperture” value. Note: A higher aperture reduces the light entering the camera, affecting both brightness and depth of field. Rekor recommends avoiding exceeding 40% in most scenarios.
Adjust the “Exposure Rate” value. Note: Set between 1/1000 and 1/2000 for optimal brightness and reduced motion blur. Rekor recommends 1/1600 as a typical starting value.
Adjust the “Infrared Intensity” value. Note: When configuring Day Time Mode, set Infrared Intensity to 0%. When configuring Night Time Mode, normal correct values range from 40% to 70%. Rekor recommends 50% as a typical starting value. 100% Infrared Intensity will result in a poor nighttime image and system performance
Use “Auto Focus” to focus the image. Note: As detailed in this document, Rekor strongly recommends using a parked car or a license plate on a tripod during the calibration and focus process. Attempting to use Auto Focus on a moving vehicle will almost always result in a non-ideal focal point and poor image clarity.
Click “Save Day” or “Save Night” to save your settings.
When calibration is complete, click “Start LPR” to re-start the LPR service.
