Today, there is an increasing demand for high accuracy and high speed in processing images for Automatic Number Plate Recognition (ANPR), also known as ALPR in some regions.
In this article, we’ll share some of the performance speeds of our ANPR engine and use-cases that may require high ANPR speeds.
ANPR Inference Speeds
The average time taken per one lookup for license plate recognition:
- 0.057 seconds for Nvidia K80 Cloud with GPU
- 0.060 seconds for Intel Xeon E3-1240 v3 8 cores, without GPU
- 0.070 seconds for Intel i7-8550U CPU @ 1.80GHz × 8 Cores, without GPU
- 0.110 seconds for Intel Xeon E5-2680 v2 @ 2.80GHz x 4 Cores, without GPU
Based on the above Inference Speeds, our ANPR engine can process images up to 17 frames per second.
Situations That Require Fast ANPR
There are several use-cases that require fast ANPR.
- Highway Monitoring. Vehicles on a highway are moving at fast speeds so in order to capture the vehicle license plates, a very fast camera is needed. Once motion detection is triggered, multiple frames are taken to maximize the ability to detect the vehicles.
- Police Mobile ANPR. Unlike the stationary systems of Highway Monitoring, Police Mobile ANPR solutions are mounted inside the police car, with cameras in front, side and/or rear of the vehicle. As the police vehicle travels along the road, the cameras grab multiple frames per second and feed the images to the ANPR engine to decode the license plates.
- Toll Plaza. Tolls with several cameras across multiple lanes send the images over to a central server. The central server runs the ANRP engine to process the license plate images.
If you have done your own benchmarking of Plate Recognizer SDK on different platforms, please share your results with us!