Eyedea Recognition

Number Plate Reading

Automatic Number Plate Reading (ANPR)

Eyedea Recognition develops and delivers own Automatic Number Plate Reading software (ANPR). We specialize mainly in high-end ANPR solutions optimized according to customer specifications.

Eyedea Recognition produces a license plate reading software both for static and moving (hand-held) systems.

Software for static cameras

product_03_large

Function:

The ANPR software detects number plates within a given image and reads a given text. The system can be completed by a car category/make/model recognition module and by a face detection module.

Main features:

  • Extremely effective and fast number plate detector
  • OCR engine designed especially for number plate reading
  • SDK enabling a simple integration into a customer's solution
  • All the engines are optimized for customer's specific conditions (i.e. for specific cameras, CPU, lighting conditions, localization etc.)
  • The software can be configured for arbitrary number plate type (shape, colour etc.) and different fonts including both Latin and non-Latin characters

Applications:

  • Toll-systems
  • Average speed limit monitoring
  • Traffic monitoring
  • Traffic data collection
  • Travel time measurement
  • Traffic law enforcement (speed and red-light violation cameras)

Product:

The ANPR software is distributed as a SDK containing a dynamic-link library, a simple example, a demo-application and documentation.

References:

The Eyedea ANPR engines power the CAMEA UNICAM ANPR system installed in many parts around the world.

Eyedea Recognition provides the high-end Number Plate Reading software. The engines are based on adaptive algorithms enabling automatic optimization for integrator's specific conditions (i.e. for specific cameras, CPU, lighting conditions, localization etc.).

How does it work?

01
02 02 02 02

Technical info:

The number plate reading software consists of two basic routines:

  • Number plate detector which detects plates within an image and
  • OCR engine which reads the text

Number plate detector:

02
The extremely effective appearance based detector is used. The detector is formed from a sequential classifier automatically trained using the WaldBoost algorithm. This algorithm processes huge training image sets and automatically builds the optimal sequential detector, which maximizes the detection success rate and also the detection speed.

The appearance based detector enables to detect number plates in different resolutions and in low intensity conditions. The detector is also able to distinguish number plates from other text areas and to minimize false alarms.

Figure 2 shows performance of a detector designed for CAMEA in 2010. The ROC curve plots true positive rate vs. false positive rate (false alarms rate) for varying detection threshold. The graph shows close to 100% detection success-rate with low false alarm rate (~0.03 false alarms per image). The detection speed strongly depends on a scanning area size and scanning parameters configuration. Typical average detection speed for scanning area 580x1280 px is 0.09s (running on one 1.8GHz core).

OCR engine:

02

The used OCR engine was developed especially for reading number plates respecting a low text resolution (from 10px text height) and a plate syntax. The OCR classifiers are trained on huge image data sets, which are continuously increased daily by new incoming image data. The classifiers can be prepared for any font and both for Latin and non-Latin alphabets.

A typical long-term average success-rate is between 95-98% depending on image quality. A typical OCR execution speed is around 0.05s (running on one 1.8GHz core).

ANPR package configuration and integration:

Commonly, the complete ANPR package configuration is provided by Eyedea Recognition on the basis of example data sample received from a customer / integrator. The example data sample usually consists from 5.000 to 10.000 images. The set should reflect given project specification (expected cameras, localization etc.). The customer will then receive optimally configured ANPR package with proved success-rate statistics measured on given data sets.

The next ANPR package integration into customer's system is very simple. Basically just three functions are used:

  • Engine initialization
  • Number plate detection function
  • Number plate text reading function

The ANPR software is distributed as a SDK containing a dynamic-link library, a simple example, a demo-application and documentation.

Input:

bitmap image (with one or more cars, see Figure 1)

Output:

  • Detected number plate positions
  • Normalized number plate image
  • Plate number in ASCII / Unicode
  • Plate number confidence
  • Alternative hypothesis

Accuracy: >95%

Supported platforms:

  • Windows
  • Linux/Unix
  • Solaris

For more see developer's guide. (PDF, 283 kB)

ANPR package maintenance:

Eyedea Recognition also provides a contractual service, which includes long-term ANPR package maintenance consisting of regular updates and statistics evaluation.

Software for moving (hand-held) cameras

Real-time number plate detection and recognition up to ±60º in any direction from frontal view.

Applications:

  • Hand-held systems
  • Mobile systems for police cars
  • Mobile LIDAR systems
  • Security monitoring

Input:

bitmap image (with one or more cars) or image stream

Output:

  • Detected number plate positions
  • Normalized number plate image
  • Plate number in ASCII / Unicode
  • Plate number confidence
  • Alternative hypothesis

Accuracy:

> 90%

Supported platforms:

  • Windows
  • Linux/Unix
  • Solaris

Product:

System is delivered as a stand-alone application or as a dynamic-link library.

References:

The Eyedea ANPR engines power the CAMEA UNICAM system installed in many parts around the world.

The automatic number plate reading software consists of two basic routines:

  • Number plate detector which detects plates within an image and
  • OCR engine which reads the text

Number plate detector:

A number plate detector invariant to viewpoint changes is used. The detector is able to detect plates viewed from acute angles and a high range of scales. This enables to use the system in mobile (e.g. hand-held) systems.

OCR engine:

The used OCR engine was developed especially for reading number plates respecting a low text resolution (from 10px text height) and a plate syntax. The OCR classifiers are trained on huge image data sets are continuously increased daily by new incoming image data. The classifiers can be prepared for any font and both for Latin and non-Latin alphabets.

A typical long-term average success-rate is between 95-98% depending on image quality. A typical OCR execution speed is around 0.05s (running on one 1.8GHz core).

ANPR package configuration and integration:

Commonly, the complete ANPR package configuration is provided by Eyedea Recognition on the basis of example data sample received from a customer / integrator. The example data sample usually consists from 5.000 to 10.000 images. The set should reflect given project specification (expected cameras, localization etc.). The customer will then receive optimally configured ANPR package with proved success-rate statistics measured on given data sets.

The next ANPR package integration into customer's system is very simple. Basically just three functions are used:

  • Engine initialization
  • Number plate detection function
  • Number plate text reading function

For more see developer's guide. (PDF, 283 kB)

ANPR package maintenance:

Eyedea Recognition also provides a contractual service, which includes long-term ANPR package maintenance consisting of regular updates and statistics evaluation.

Would you like to know more?

Contact us on info@eyedea.cz