When recognition accuracy disappoints, optimise the imaging conditions first — it is far more effective than repeatedly tuning algorithm parameters.

• Insufficient pixel density: licence plate recognition needs the plate to occupy enough pixels (roughly 100 across or more); face recognition needs at least 60 pixels between the eyes. Excessive distance or an over-wide lens breaks both.
• Camera angle: for plates keep the horizontal offset under 30 degrees and the tilt under 20 degrees. For faces, use a frontal view with a tilt under 15 degrees; profiles sharply reduce accuracy.
• Lighting: backlight, strong glare and unlit night scenes are the main sources of failure. Use wide dynamic range models at entrances and gates, with dedicated illuminators.
• Motion blur: high vehicle speed or a long shutter produces smear. Prefer shutter-priority settings or add light.
• Watchlist and thresholds: low-quality enrolment images and over-tight thresholds both present as recognition failures.

Commission against real conditions: validate with actual vehicle or pedestrian footage before locking in the mounting position and parameters.