From gate control and parking lot management to toll collection, automatic license/number plate recognition has come a long way. This article examines the latest market status, technological developments and common installation problems associated with this technology.
Automatic license/number plate recognition (ALPR/ANPR) is a fast developing technology that has been proven very effective in many applications such as vehicular access control and speed regulation. Combining ALPR/ANPR with video surveillance is even more useful as it provides context information as evidence.
The global market for ALPR/ANPR software and integrated devices (software embedded in cameras and DVRs with onboard processing) was estimated by IMS Research to be US$119 million in 2008 and will grow to $442.9 million by 2013, at a compound annual growth rate of approximately 30 percent. “The EMEA market is currently the largest and most sophisticated one for ALPR/ANPR usage,” said James McManus, Market Research Analyst at IMS Research. And the market split, according to A.R. Hungary, is roughly: EMEA 50 percent, the Americas 25 percent (mostly the United States), Asia 15 percent and others 10 percent.
Major applications, said McManus, include security and law enforcement, tolling and congestion charging, access control and parking management, and journey time information. “Security and law enforcement plus tolling accounted for more than 60 percent of the revenues generated in 2008. In terms of units, more access control and parking systems were sold, accounting for more than half of the market.”
These numbers were seconded by Erno Szucs, Sales Director of ALPR/ANPR for A.R. Hungary: law enforcement 40 percent, tolling 25 percent, parking and access 20 percent, and travel time information 15 percent.
There are a few steps involved in ALPR/ANPR: plate locating , plate capturing, image processing and final identification, said Vincent Chen, Assistant VP of Product Marketing for GeoVision, “of which, image processing is where differentiation arises.” It encompasses plate reorientation (for images captured from an angle), X/Y-axis determination, image/frame comparison (to enhance accuracy), pattern recognition (multiple countries or states) and character recognition (black/white, OCR). Each step requires highly sophisticated algorithms, and comprehensive, clearly defined databases are key to ensuring accuracy. “With seven years of R&D and experience, GeoVision's fourth-generation ALPR/ANPR technology can capture plates traveling at 200 kilometers per hour and has achieved an accuracy rate exceeding 98 percent.”
Pattern or plate type recognition is no small feat, either. Country and state recognition, especially in Europe and the Middle East, and database referencing are crucial in correct identification, said Szucs. This was echoed by Matthew Messinger, Senior Communications Consultant, Government and Public Safety Communications, Motorola (North America): “ALPR/ANPR has to be able to identify many types of license plates. This is particularly a challenge in North America, where every state has a variety of license plate designs — including vanity plates.”
Special characters, such as those from the Arabic and Chinese languages, also pose an entry barrier for many suppliers, Szucs cautioned. Honeywell uses four different algorithms to read license plates and cross-reference the characters read. This greatly increases the accuracy of the readings. Aside from plate recognition, some developers are working on new algorithms, such as automatic verification against registered color and make, helping detect false plates, said James Somerville-Smith, EMEA Market Development Leader for Honeywell Security.
From standard ALPR/ANPR systems implemented at border controls and toll stations, to recognition and reading of moving vehicles such as coal wagons and public buses, Christian Bohn, Head of Product Management for Milestone Systems, is seeing a growing interest in a range of applications. “The area of forensics is also expected to drive the uptake of ALPR/ANPR, through storing metadata in an IP format that enables more advanced data analysis and sharing.”
ALPR/ANPR systems typically consist of image capturing hardware, recognition software and back-end/database management, said Wuning Jian, Image Processing Engineer at Hikvision. Both software and hardware need to be properly configured for an ALPR/ANPR system to be effective.
"There are usually physical limitations as far as plates or cameras are concerned, which could affect the outcome of a supposedly automated system,” said Chen. "The paint, rust or frame of a plate can easily interfere with the recognition process; as such, these need to be taken into consideration during the software engineering/design phase.” Other influencing factors include plate size, camera positioning, zone setting, exposure, shutter and frame rate.
Hence, picking adequate cameras and positioning them right seem to be equally important.
Cameras must be rugged and weatherproof for outdoor use, said Messinger, and according to Motorola and A.R. Hungary, these cameras should be specifically designed to accommodate varying lighting conditions (such as through IR illumination or wide-dynamic functionality). “IR illumination is particularly useful in detecting retroreflective license plates,” Rasmus Crüger Lund, Software Architect for Milestone Systems, further explained.
Other lighting-related issues include contrast, lens and exposure. “When determining the right contrast for ALPR/ANPR, consider the difference in gray value (when images are converted to eight-bit grayscale) between the license plate's characters and its background color,” said Lund. “When you convert your input image to an eight-bit grayscale image, the minimum pixel color value difference between a pixel in the foreground and a pixel in the background should be at least 15. Note that image noise and compression can make it difficult to determine what the colors of a license plate's characters and background are.”
When configuring lenses and shutter speeds for ALPR/ANPR, Lund suggested taking heed of auto-iris, IR setting and expected traveling speeds. "If using an auto-iris lens, always set the focus with the aperture as open as possible In order to make the aperture open, you can use ND filters or — when the camera supports manual configurations of the shutter time — the shutter time can be set to very short. If using an IR light source, focus may change when switching between visible light and IR light. You can avoid this by using an IR-compensated lens, or by using an IR pass filter. Note that when using an IR pass filter, an IR light source is required — even during daytime. When detecting moving vehicles, shutter time should be short enough to avoid motion blurs."
In case of a high way application, “a car driving at 200 kilometers per hour moves 56 meters in a second, or 5.6 meters in 100 milliseconds,” said Szucs. “If the camera covers a section of six meters on the highway, then we need to make sure the processing is done within this timeframe to be able to catch every car in real time.” For high speeds, devices also have to be installed in the road to capture video frames at the correct time, added Bohn.
Overexposure — a common problem both during the day and at night, as there could be strong sunlight or car headlights — usually results in an overly white appearance, continued Lund, addressing more lighting issues. “To avoid overexposure, it's recommended that you use a camera with wide-dynamic functionality and/or use an auto-iris lens.”
Underexposure, on the other hand, results in a dark image with hardly any contrast. “It can be avoided by using external lighting and/or by using a camera which has sufficient sensitivity in low-light environments without using gain,” explained Lund.
Smear is also a common problem, leading to unwanted vertical strips in images, said Lund. “It's frequently linked to slight imperfections in CCD imagers. In general, CCDs with larger surfaces are less sensitive to smears. Cameras with CMOS imagers are also less sensitive to smears than those with CCDs.”
Physical positioning of cameras plays a vital role during installation as well. "Mount the camera in such a way that an ideal image of the license plate is captured when the license plate is in the center of the recorded image,” said Lund. "The maximum vertical view angle of a camera used for ALPR/ANPR is 30 degrees, and the maximum horizontal view angle is 25 degrees. In most systems, the horizontal angle is somewhere between 15 and 20 degrees.”
In reality, there are also undesired camera features, such as automatic gain adjustment, enhancement and compression. “Some cameras use contour, edge or contrast enhancement algorithms to make images look better to the human eye. However, such algorithms can interfere with the algorithms used in the recognition process,” explained Lund. “Also note that ALPR/ANPR will not work if cameras use MPEG compression. When a high compression rate is used, more resolution is required to achieve optimal performance.”
With the help of intelligent features, “we can follow a specific vehicle's trajectory and direction with a graphical interface,” said Szonja Balogh, Marketing Manager of Intellio. Using the data collected, today's smart systems can analyze events happening on the road in seconds and take action to alert the appropriate authorities. “We can extract data not only from individual cameras but also from groups of cameras, enabling us to create more intricate and meaningful statistics.”
ALPR/ANPR technology continues to move forward in leaps and bounds, allowing suppliers to service 50 to 80 countries with one single package. To be truly automatic, “well-designed software should require no human interaction until a vehicle of interest is found,” said Messinger. Development work will focus on further improving accuracy (currently at 95 to 98 percent), providing greater detail in what is matched and increasing seamless integration of ALPR/ANPR with other security and business systems, said Somerville-Smith.
"If we utilize a monochrome, five-megapixel camera, we can concentrate the algorithms on the font-matching exercise,” said Dave Tynan, VP of Global Sales for Avigilon, expanding on greater image detail. “This eliminates the background graphics that could potentially corrupt the process of capturing the characters clearly and precisely. With a color, megapixel camera, triangulation of evidence (vehicle colors, distinguishing marks, occupant descriptions) is made easy, providing visually lossless video for a higher percentage of successful investigations.”
Chen agreed with the open integration argument: “A clear trend is to have an open SDK for easier integration with surveillance, access control and other systems.” For example, ALPR/ANPR systems from Honeywell and Siemens Building Technologies are used at premises like airports to manage individual parking spaces in parking lots that allocate spaces to cars in real time.
McManus is seeing a t rend toward all-in-one ALPR/ANPR integrated devices. "There will also be growth in fixed, 24/7 systems and integration with third-party technologies such as facial recognition.” Advances in wireless technology also bring about easier installation and reduced cost. Essentially, such software can be deployed in any environment where there is a need to read numbers. “They could be cargo containers, trains, trucks or boxes on a conveyer belt,” said Bohn.
"Our company has experienced annual growth of more than 40 percent for the past four to five years. We are expecting to keep this pace in 2009,” Szucs concluded confidently, boding well for ALPR/ANPR systems.