approach: We measured the camera angle with a protractor. Distance from camera to vehicle was 6 feet for all images taken.
results: Plate Recognizer ALPR results were correct up to a 70 degree angle. It failed at 75 degree angle.
approach: What matters in image resolution is actually the pixel width of the license plate itself. In the original image (2880×2160), the plate is 340×140, or 12% of the original image width. We resized the images below and indicated the plate’s pixel width.
results: Our results were correct up to 30 pixels of the plate width (roughly 26 pixels per foot). It failed at 28 pixels of plate width.
approach: We applied Paint.net‘s Motion Blur feature at a standard 25 degree angle on various blur distances as stated below.
results: Our results were correct up to a blur distance of 27. It failed at 30. Some additional examples of very blurry plates.
approach: We applied Paint.net‘s Lightness levels (under Hue/Saturation Adjustments) to the original image.
results: Plate Recognizer ALPR results were correct up to a Lightness Level of -93.
approach: We applied Paint.net‘s Rotate Image based on the degrees indicated to the original image.
results: Plate Recognizer ALPR results were correct up 53 degrees of image rotation.
approach: For some of the above cases, we took images of a single vehicle (black Lexus IS250) to ensure consistency in the analysis. In other cases, we collected images online or from some of our partners (with permission to share).
