NumberOk offers advanced recognition settings to compensate camera installation disadvantages, bad weather conditions, lack of lighting, and number plates defects.

This manual describes ANPR Analytics Setup in details to allow the user to set the best combination to achieve the best recognition results under particular conditions. 

The input video stream is decoded and analyzed frame by frame and number plates are recognized from each individual frame. Navigate to Settings > Analytic to tune up the recognition results processing.

When checked analytical settings will be applied to multiple recognitions to produce a summarizing result.

Soft comparison

Off - the result from each individual frame recognition will be shown as a standalone event.

On - results of recognition will be combined into summarizing results following the rules as set below.

Levenshtein distance

Recognitions from every individual frame that differs in no more than N symbols as specified are considered equal and combined into a single event.

Example. NumberOk has recognized license plate number from 6 consequent frames. Some recognition results are different from real ones due to weather conditions, lighting, and image defects. Depending on Levenshtein distance parameter, recognition results will be grouped as follows:

Use raw recognition results

When checked raw (uncombined) recognition results will be delivered to external applications via integrations.

Do not show first N results

(Allowed range is 0 to 10)

Initial recognition as specified is analyzed before any result is produced. Efficient under poor environmental conditions. For example, when license plates are poorly lightened or vehicles enter the frame at steep angles from sidewards.

Ignore region

When checked turns OFF region code recognition for license plates from Russian Federation.

Decision rules

Definition of the rules applied to recognition results for combining into summarizing results. Interim results can be combined per highest confidence level or per highest number of completely equivalent results (statistical).

Best result on confidence

When checked the final result is determined based on the highest confidence level from among consequent results.

Best result on statistic

When checked the final result is determined based on the highest frequency among consequent results.

Minimum and maximum count of symbols in a license plate. Recognitions, where symbols count lies beyond the range, will be ignored. The closer the range boundaries to realistic counts the finer results are produced.

Minimum symbols count in the plate number

Defines the min. count of symbols in recognized plates, shorter counts will be ignored. The practical recommended minimum is 5.

Maximum symbols count in the plate number

Defines the max. count of characters in recognized plates, longer sequences will be ignored. The closer the value to realistic counts the finer results are produced.

Soft comparison. Levenshtein distance

Any number plate that differs from any other in the database in no more than N symbols will be deemed as the one from the database. Might be useful for general purpose when camera installed under poor environmental conditions.

Strong comparison

The number will be considered appropriate only when exactly matches any number from car database (black, white or other lists). Recommended when access control is critical, however requires good environmental conditions and camera being installed following recommendations. 

Soft comparison. Levenshtein distance

The number plate of the car on exit will be matched against any number plate from among those that have entered before, a difference in N characters is allowed.

Used in Results > Grouped by number mode.

Strong comparison

The number of the car on the exit will be matched against any number plate from among those that have entered before only when the match is exact. 

Used in Results > Grouped by number mode.