What factors affect the accuracy of license plate recognition?
From the installation angle, fill light, lens to algorithm parameters, the key factors and optimization methods that affect the license plate recognition rate are systematically sorted out.
Installation parameters (the most critical)
| Parameters | Recommended values | Description |
|---|---|---|
| Lane width | 2.5~3.5 meters | Excessive width will cause vehicles to drift out of the lane |
| Camera height | 1.2~1.8 meters | Special machine installed close to the lane |
| Trigger distance | 4~6 meters | Leave a capturing distance for the recognition algorithm |
| Horizontal angle | ≤ 30° | Excessive large angle causes character deformation |
| Depression angle | ≤ 25° | Excessive large angle causes character compression |
| License plate pixel width | ≥ 120 px | Below this value, the recognition rate drops significantly |
Equipment selection
- Using the Special License Plate Camera, it has strong light suppression and license plate-specific algorithms, and the effect is far better than that of a general-purpose camera.
- Select a 6~12mm lens with manual zoom, and fine-tune the license plate to the appropriate proportion of the screen on site.
- Large target area sensors (1/1.8″ and above) perform better in low light.
Fill light and backlight
- Fill-in light is necessary at night: Use strobe fill-in light (very bright), Always-on lights can cause driver glare and cause complaints.
- Install the fill light on the same side as the camera to prevent light from shining directly on the lens.
- For strong light from car lights, turn on Highlight Suppression and limit the upper limit of the shutter speed.
- When the entrance faces west, the sunset in the evening will shine directly on the lens, so you need to install a hood or adjust the installation position.
Algorithm and parameters
- Enable license plate duplication and multi-frame comparison to avoid repeated identification of the same vehicle.
- Configure the unlicensed vehicle processing strategy (such as manual confirmation or scan QR code to pass).
- New energy license plates (gradient green, yellow and green) need to confirm that the algorithm version is supported.
- The trailer/tractor must lock the main license plate to avoid recognizing the front of the vehicle.
- Defaced license plates and mud and water are the most difficult situations. It is necessary to check whether the recognition rate meets the business requirements.
Acceptance method
Collect traffic data continuously for 7 days, covering day and night and peak hours, and calculate the recognition rate. Common industry goals: ≥ 99% during the day and ≥ 98% at night.If it is lower than expected, check the installation angle and fill light first. Instead of blindly increasing the algorithm sensitivity.
Related content
Linkage integration of video surveillance, access control, and alarm systems: interface methods and implementation points
09-19
Moiré and false color in surveillance images: lens and sensor matching issues
09-18
Progressive transformation from analog to network: reusing coaxial and hybrid networking solutions
09-17
Lens dirt and protective window cleaning: invisible causes of image quality degradation
09-16
Surveillance video as evidence: retrieval process, preservation and legal effect
09-15
Permission classification and desensitization of video data: who can watch and who can export
09-14
Security system level protection: What requirements should the video surveillance part meet?
09-13
Computer room dynamic environment monitoring: temperature and humidity, mains power, water leakage and smoke detector access
09-12
