ROAD TO THE BEST ALPR IMAGES
INTRODUCTION Since automatic license plate recognition (ALPR) or automatic number plate recognition (ANPR) relies on optical character recognition (OCR) of images, it makes sense that a higher quality input image results in higher accuracy. Unlike what is shown on TV, you cannot zoom into a blurry image and expect to get more details. An image with acceptable sharpness and contrast must be acquired with the appropriate system from the start. This means the right image sensor, camera, optics, and lighting all combined in a reliable way. OCR ALGORITHMS WORK BETTER WITH HIGH QUALITY IMAGES FOR ALPR -2-
WHAT DEFINES GOOD IMAGE QUALITY FOR ALPR? The first step is to have reliable triggering in order to have the license plate in the proper location in the image, which can be especially difficult in multi-lane systems. After that, a good/accurate image can be described by:»»»» Good Sharpness Sufficient Contrast Free of artifacts And sometimes with accurate color The sources of these image quality issues can vary. Some possible reasons are shown in the table below, and sharpness, contrast, and artifacts are further detailed in the following sections. Image Quality Parameter Corresponding Source of Limitations Image System Parameters to Control Sharpness Limited depth of field Motion blur Variable lighting F value of lens Sensitivity of image sensor Iris control Contrast Artifacts Limited number of images Reflections of the license plate Reflections of snow, rain, flog Ghost images Bright spots and streaks from sun exposure and reflections Frame rate of the image sensor/camera Dynamic range of the image sensor/camera Alignment of filter, lens, and lighting. Channel matching in the camera Blooming and smear control in the camera -3-
SHARPNESS Sharpness is one component of image quality. It indicates the clarity of an image and therefore the amount of fine details in the image. If all of the components in the vision system are not well matched and aligned, the spatial details will be blurred. If you match these well, the total accuracy of your ALPR system can be increased. Especially in high speed ALPR systems such as open road tolling, it can be a challenge to get the required sharpness. Here are some factors that impact sharpness and how to overcome them: DEPTH OF FIELD A general definition of Depth of Field (DOF) is the distance between the nearest and farthest objects in a scene that appear acceptably sharp in an image. With an image for ALPR, the entire image needs to be sharp so a very large depth of field is required. A larger DOF is achieved with smaller iris openings versus larger openings. Two ways to allow for smaller iris openings is with a lower F number of the lens and with a more sensitive sensor. -4-
MOTION BLUR Motion blur is the fuzzy details that can appear when capturing a still image of a fast moving object, such as a car/license plate on the highway. Again, a lower F value of the lens can help here as it allows for shorter exposure times to better freeze the moving object. More sensitive sensors also mean less light is required to get a good image, thus enabling shorter exposure times. LIGHTING Different license plates have different reflection coefficients. For optimal results, the wavelength of the IR lighting should be matched to the license plate. IRIS CONTROL Having a fixed iris verses auto iris offers more control over the image. By taking multiple images of the same object with different exposure times with a fixed iris, better control over the focus and exposure is achieved. Auto iris functionality can generate a dynamic depth of field and therefore fuzzy portions in the image. -5-
CONTRAST Now that you have done what is possible to get a sharp image for your OCR algorithm for automatic license plate recognition, another critical image quality parameter that is critical is contrast. Contrast is the difference in brightness between the light and dark areas. Much finer details can be detected if the difference between the light and dark areas is more pronounced. Some suggestions on ways to improve contrast that are specific to the needs of ALPR: EXPLOITING HIGHER FRAME RATES OF A CAMERA Cameras with higher frame rates allow for multiple images to be taken of the same object with different exposure times. This way multiple images under different conditions are available, and the best one can be selected. There are now CCD cameras available with 2MP HD resolution and speeds of more than 60 frames/second. For CMOS cameras, the speeds can be more than 5 times higher. USING CAMERAS WITH A HIGH DYNAMIC RANGE Image sensors with high dynamic range can distinguish the foreground (the license plate characters) better from the background. For license plates in certain regions of the world, this is particularly challenging. If implemented properly, camera manufacturers ensure the full linear dynamic range of the sensor is available. They can even add functionality to increase the dynamic range. LIGHTING Poor reflection of light on the license plate can limit the contrast. Different license plates have different reflection coefficients. As with optimizing sharpness, for optimal results, the wavelength of the IR lighting must be matched with the license plate to be measured. Snow, rain, and fog also reflect the IR LED. Again, special attention to the IR wavelengths used will enhance the contrast of the image. -6-
MINIMIZE ARTIFACTS Reducing image sensor artifacts is not a simple thing to do, but camera manufacturers can help to remove or minimize certain artifacts that are specific to the needs of ALPR: PREVENTING GHOST IMAGES Ghost images can appear if Infrared (IR) lighting is used in combination with a visible light block filter. By using the correct filters, ghost images can be decreased as long as the filter is properly aligned with the lens, camera, and the lighting. The simplest way to prevent ghost images and lens artifacts from interfering with the system performance is to utilize a camera supplier that also has the expertise to properly integrate the filter and lens with the camera. MANAGE BLOOMING AND SMEAR Blooming and smear are challenges with outdoor vision systems, where blooming and smear (streaks) are artifacts created by saturation from very bright spots in a scene (See Figure 5). Bright spots can originate from headlights, reflections off license plates, the sun at certain times of the year, or sun reflecting on the road. Image processing in the system cannot correct these artifacts so blooming and smear must be managed in the camera through special functionality to ensure that the license plate is not obscured in the original image data. -7-
IMPROVE CHANNEL MATCHING Even with effective management of blooming and smear, direct sunlight can cause a poor image if the image sensor channel matching is insufficient in the camera. Image sensors usually have 2 or 4 readout channels that need to be stitched together in the camera to recreate the complete image. Cameras with bad channel matching can deliver images with one part overexposed and the other part underexposed. This leads to poor performance of the OCR algorithm. CONCLUSION With a higher quality of the input image, there is a better starting point for the license plate recognition algorithm, and therefore the higher license plate recognition accuracy. With proper alignment of the lens, filter, camera, and lighting, as well as specialized functionality in the camera to deal with extreme lighting conditions of traffic applications, image artifacts are reduced or eliminated. When combined with optimized sharpness contrast, the result is in high quality images. This improves the efficiency of the OCR algorithm, providing the system integrator with a better chance to win the tender contracts. In the end, the return on investment will be greater and ultimately road safety is improved. -8-
ADIMEC Adimec specializes in the development and manufacturing of high-performance cameras that meet the application-specific requirements of key market segments, including machine vision, medical imaging, and outdoor imaging. Founded in 1992, the company partners with major OEMs around the world to facilitate the creation of industry-leading cameras. The unique Adimec True Accurate Imaging technology provides new levels of precision and accuracy to vision systems. Its diverse line of camera products meet a wide range of performance, size, cost, interface and application requirements. Adimec has offices around the world focused on creating customer value and satisfaction through local, personalized support. Need more inspiration? Contact us www.adimec.com -9-