Car numbers under control.
From entry control systems and parking payment systems, automatic recognition of vehicle number plates has come a long way.
This article analyzes the state of the market, technological advances and common problems associated with the application and development of this technology.
Automatic Number Plate Recognition (ANPR) is a rapidly growing technology that has proven its effectiveness in many applications such as vehicle access control and speed enforcement.
Integrating ANPR with CCTV makes the system more effective as it provides video evidence.
The global market for ANPR software and integrated devices (software included in the camera and DVR with data processing in the device itself) was valued by IMS Research at $9 million in 2008 and $442.9 million by 2013, growing at approximately 30% annually.
“EMEA is the largest and most sophisticated market for ARNZ systems today,” says James McManus, a market analyst at IMS Research.
According to ARHungary, the market is roughly divided as follows: EMEA – 50%, the US – 25% (mainly due to the US), Asia – 15% and the rest – 10%.
Most applications, McManus said, include enforcement and safety monitoring, toll collection, access control and parking management, and travel time information.
Traffic violation recording, enforcement and safety monitoring, plus fees for additional services, accounted for more than 60% of revenue in 2008.
In terms of application, access control and parking management systems accounted for more than half of all sales in the market.
Erno Szucs, Director of Sales for ARHugary’s ARNZ systems, confirms these figures: recording traffic violations – 40%, fees and charges – 25%, parking and access control – 20%, travel time information – 15%.
Main developments
“There are usually several steps in the operation of an ARP system: finding the license plate, capturing the license plate data, processing the data, and then final identification,” says Vincent Chen, assistant vice president of marketing at GeoVision. “And the image processing is where the difference comes in.
This includes license plate reorientation (for images captured at an angle), X/Y difference, frame/field manipulation (to improve accuracy), pattern recognition (different states and countries), and character recognition (black and white, OCR).
Each step requires highly sophisticated algorithms, and clear and well-defined databases are key to ensuring accuracy.”
Thanks to seven years of research and development, GeoVision’s state-of-the-art fourth-generation ARNZ algorithms and technologies can “catch” vehicle license plates at 200 km/h with recognition accuracy of up to 98%.
“Recognizing the type of plate or pattern is also not an easy task.
«Country and state identification, especially in Europe and the Middle East, and database searching are critical for correct identification,» says Erno Stzuks.
Matthew Messinger, Senior Consultant, Government and Public Safety Communications, Motorola (North America), agrees: «APNZ systems must be able to identify many types of plates.
This is a challenge in North America, where each state has many variations of plates, including custom designs.»
Special characters such as Arabic and Chinese also pose a major barrier for many vendors.
Honeywell uses four different algorithms to read license plates and cross-references when reading characters.
This greatly improves recognition accuracy.
“In parallel with the number plate recognition algorithms, some developers are working on other algorithms, such as automatically detecting mismatches between the vehicle colour and manufacturer and the parameters specified in the database to identify fake numbers,” says James Somerville-Smith, head of market development EMEA at Honeywell Security.
“There is a growing interest in a number of applications from standard ARP systems used at border crossings and intercity checkpoints to systems for reading and recognizing moving vehicles such as coal wagons and public transport,” says Christian Bohn, Head of Production at Milestone Systems.
Installation Tips
“APRP systems typically consist of image capture hardware, recognition software and database searching,” says Wuning Jian, Image Recognition Engineer at Hikvision. –
For the ARNZ system to be effective, the hardware and software must be properly installed and configured.»
There are usually physical limitations associated with the license plate or camera that affect the performance of the automated system.
Paint, rust, and deformation of the license plate can affect the accuracy of the recognition process, which should be kept in mind when designing and creating the algorithm.
Other factors that come into play include the size of the license plate, where the camera is installed, the detection zone setting, exposure, camera shutter speed, and frame rate.
“Therefore, choosing the right camera is as important as choosing the right location for it.
Cameras need to be rugged and weatherproof for outdoor use,” says Matthew Messinger. “These cameras need to be specifically designed to operate in variable light conditions (either with IR or high dynamic range).”
«IR illumination,» notes Rasmus Crüger Lund, a software development engineer at Milestone Systems, «is particularly useful for detecting reflective license plates.»
Other factors related to illumination include contrast, lens and exposure.
«Determining the correct contrast level for the ARP system takes into account the level of gray (when images are in eight-bit grayscale) between the characters on the license plate and the background color,» Land says.
«When you convert an original image to an eight-bit grayscale image, the minimum color difference between the pixels in the foreground and the pixels in the background must be at least 15. However, image noise and compression can make it difficult to determine the colors of the characters and the background of the license plate.»
When setting up lenses and shutter speed for ARNZ systems, Land suggests paying attention to the automatic aperture, IR illumination setting, and the expected speed mode. If you use a lens with an auto-iris, you should always set the focus with the aperture wide open.
For this, you can use ND filters or, if the camera supports manual shutter time settings, make the shutter time as short as possible.
When using IR illumination, the focus may change when switching between the visible and infrared spectrum.
This can be avoided by using spectrum-shifted lenses or IR filters, and when using the latter, the IR illumination must be turned on even during the day.
When detecting a moving vehicle, the shutter time must be short enough to avoid blurring the image.
«When using the system on a motorway, a car traveling at 200 km/h travels 56 m per second, or 5.6 m per 100 milliseconds,» says Erno Stzuks from ARHugary.
«If the camera covers a sector of 6 m, you need to be sure that the processing will be done in this time frame, so that it is possible to catch every car in real time.»
«For high speeds, the devices also need to be installed on the road to capture video frames at the right time,» adds his colleague Christian Bohn from Milestone Systems.
“Overexposure is a common problem both during the day and at night, as there may be strong glare from the sun or car headlights, and usually results in an overly whitish image,” Land continues.
“To avoid this, you should use cameras with WDR (wide dynamic range) and/or a lens with an auto iris.”
On the other hand, underexposure results in a dark image with low contrast.
«This can be avoided by using additional external lighting and/or a camera with high ISO that can operate in low light without amplification,» Land explains.
«Blur is also a known problem, adding unwanted vertical stripes to the image,» Land also notes. «This is often due to a slight imperfection in the CCD. In general, CCDs with a larger area (diagonal) are less sensitive to this kind of interference.
CMOS cameras are also less sensitive to blur than CCDs.»
The physical placement of the cameras also plays a role. “To get the perfect image of a license plate, the cameras need to be mounted so that the license plate is in the center of the image when the frame is captured,” says Land. “The maximum vertical viewing angle of the camera used in ARNZ systems is 30 degrees, and the horizontal viewing angle is 25 degrees. In most systems, the horizontal viewing angle is between 15 and 20 degrees.”
In reality, the camera has a number of undesirable features, such as automatic gain adjustment and compression.
“Some cameras use algorithms to enhance edges, contours, and contrast to improve the visual perception of the image by the human eye,” explains Land.
– Also, the ARNZ system will not work on cameras using MPEG compression. When using strong compression, a higher resolution is required to achieve optimal system performance.”
Further development
“With the help of intelligent algorithms, we can track the trajectory and direction of a specific vehicle through a graphical interface,” says Szonja Balogh, Marketing Manager at Intellio.
– Using the data received, today's smart systems can analyze events on the road in seconds and take action to alert the appropriate services.
We can receive data not only from individual cameras, but also from a group of cameras, which allows us to conduct deeper and better data analysis.»
ARNZ technology continues to advance by leaps and bounds, allowing providers to serve 50 to 80 countries with a single solution.
«To be fully automated, well-designed software should not require human intervention until a vehicle of interest is detected,» Messinger notes.
James Somerville-Smith of Honeywell Security believes that research and development is focused on improving the accuracy (currently 95-98%) of recognition, detail, matching and increasing the degree of integration of ARNZ systems with other security and business management systems.
“If we use a monochrome five-megapixel camera, we can force the algorithm to read and recognize text,” says Dave Tynan, vice president of worldwide sales at Avigilon, which is betting on higher image detail.
“This eliminates background graphics, which can potentially negatively affect the accuracy of character capture.”
With a color megapixel camera, collecting eyewitness testimony and evidence (vehicle color, identifying features, description of occupants) becomes easier, providing lossless video quality and a higher success rate for investigations.»
Vincent Chen of GeoVision advocates open integration: “It would be good to have an open SDK (developer’s kit) for easy integration with CCTV, access control and other systems.” For example, Honeywell and Siemens Building Technologies’ ARP systems are used at airports to manage personal parking spaces in the parking lot, allocating spaces to cars in real time.
James McManus of IMS Research sees the future in integrated, all-in-one ARP devices.
There will also be an increase in fixed systems that operate 24/7, and integration with third-party technologies such as facial recognition.
The benefits of wireless technology will make installation easier and cheaper. Essentially, such systems will be deployed anywhere that license plate reading is required. These could be shipping containers, trains, trucks or boxes on a conveyor belt.
«Our company has seen annual growth of more than 40% for the last four to five years. We expect these results to continue in 2009,» concludes Erno Stzuks from ARHugary optimistically, predicting a bright future for ARNZ systems.