Fine-tuning the VT number plate detection tool

To fine-tune the detection tool, do as follows:

1. If necessary, enable Advanced image analysis (1). If enabled, it improves the quality of recognition under adverse conditions (for example, if the characteristics/settings of the camera do not fully meet the requirements, or in bad weather). The processing time of the frame increases by 20-30%, depending on its size. Under normal conditions, this setting does not affect the quality of recognition.

   
2. To recognize the numbers at sharp angles with respect to the camera, select Yes in the Algorithm of recognition of distorted LP image field (2).
3. Set the image contrast threshold (3). The default value is 40. On a high-quality image, increase this value to 50-60. If the image has poor contrast, lower this value. 

4. If you expect different sizes of number plates within FOV, activate the LP advanced search algorithm (4). This parameter accounts for max and min width of number plates, and, in some cases, may help increasing recognition quality and optimizing CPU load.

5. To set up a number plate recognition event:

   1. Specify the minimum percentage of similarity between the recognition result and the corresponding number plate template, for positive LPR result (7). Use this parameter to filter the results by reliability.

   2. By default, a number plate recognition event is registered after a track containing a number plate disappears from FOV. You can set the moment of registration to a Timeout in seconds (8), or to reaching a similarity percentage specified in the LP display quality field (5). If both parameters are set to non-zero, the recognition event will be registered upon matching the first condition.

      Note

      The LP display quality parameter value must be no lower than the Minimum similarity value.

6. Specify the maximum number of recognition threads (6). If the value is 0, the recognition process will occur in the same thread that starts it.

   Important

   The cumulative value of this parameter across all NPR detection tools must not exceed:

   • the licensed number of recognition threads (check your license with the lsvpwcutility);
   • the number of CPU cores;
   • 100.

7. In the Tracker timeout, sec field, set a time interval in seconds after which the tracking of a vehicle is reset (9).

   Note

   Use this setting to eliminate double-triggering in such cases as, for example, another recognition of the same number after it has been obstructed for some time, and then reappears in scene.

   If you set Tracker timeout to a value greater than a probable time of obstruction, the detection tool will not double-trigger.

8. If necessary, set additional parameters. For more details, refer to the table.
   

   Parameter	Description
   VodiCTL_VPW_DYNAMIC_ENABLE	Enable/disable the number recognition dynamics (by default, the dynamics is enabled). If the value is Yes, then tracking is enabled, and the number is recognized by the set of frames. If the value is No, then tracking is disabled, and the number is recognized by each frame separately without taking to account the previous ones, and the quality can vary from 0 to 100%
   VodiCTL_VPW_DYNAMIC_OUTPUT_PERIOD	Time period (in microseconds) over which the recognition result is to be displayed to the user. This parameter can be used only if the VodiCTL_VPW_DYNAMIC_WITH_DUPLICATE parameter is set
   VodiCTL_VPW_DYNAMIC_OUTPUT_TIMEOUT	The minimum time required to monitor the license plate (in microseconds) before displaying the recognition result to the user. This parameter can only be used when the “Dynamic” mode is on. In this mode, the trajectory of the vehicle is monitored, and the user does not immediately receive the recognition result of the license plate, but after the time specified for this setting. In this case, the first recognition result will be replaced by the result of higher quality and subsequently displayed to the user. If parameter 0 is set for this value, the user gets the first result of recognizing the detected license plate.  After the time specified in this parameter expires, the monitoring of the trajectory of the license plate continues until it disappears from the frame
   VodiCTL_VPW_DYNAMIC_WITH_DUPLICATE	Enable/disable the periodic output of license plate recognition results
   VodiCTL_VPW_IMAGE_BLUR	The parameter for internal use. The recommended value to set is 13
   VodiCTL_VPW_LOG_SETTINGS	Enable/disable logging of all recognition parameters
   VodiCTL_VPW_PLATE_FILTER_RODROPFACTOR	The license plate filter coefficient by the so-called image density—ratio of white pixels to total pixels (second strategy). The type is unsigned. This coefficient is used for image threshold and has the optimal values, which are determined by developers using their own test samples. The parameter is considered as a service one, and its value should be set according to the recommendations of technical support specialists
   VodiCTL_VPW_PLATE_FILTER_ROFACTOR	The license plate filter coefficient by the so-called image density – ratio of white pixels to total pixels (first strategy). The type is unsigned. This coefficient is used for image threshold and has the optimal values, which are determined by developers using their own test samples. The parameter is considered as a service one, and its value should be set according to the recommendations of technical support specialists
   VodiCTL_VPW_PLATE_FILTER_SYMCOUNT	Enable/disable the simple license plates filter algorithm by the minimum number of recognized symbols on them. If the algorithm is enabled (the value of the parameter is greater than 0), the base search for symbols on the prospective license plate (geometry, proportions) is performed. If less symbols are recognized on the prospective license plate than specified in this parameter, this prospective license plate is not considered a license plate. That is, the value of this parameter is the minimum characters that must be present on the prospective license plate when the basic algorithm is in use
   VodiCTL_VPW_PLATE_STAR_MAX	Maximum unrecognized symbols on the license plate, at which the result will still be considered the result of the recognition of the license plate

9. Click the Apply button.