Benchmarking

Drive Rekor Scout® on all CPU cores to benchmark speed for various video resolutions.

Example Benchmarks

Benchmarks demonstrating the power of Rekor Scout are presented below. If you find yourself thinking, “These numbers look too good to be true”, we understand. Don't take our word for it.

‍Try it out yourself!

Plate number accuracy

Benchmark	License plate state identified	Correct in 1 of 10 estimates	Correct on 1st estimate
OpenALPR Open Source	0.00%	65.69%	42.16%
Rekor Scout®	96.83%	100.00%	99.02%

CPU hardware video performance

Benchmarks were conducted on Ubuntu Linux using Intel systems and integrated graphics.

System profile	1080p	720p	480p
Xeon E5-2666 v3 @ 2.9 GHz	12.7 FPS	15.0 FPS	16.8 FPS
Core i5-5250U @ 1.6 GHz	17.4 FPS	20.4 FPS	22.9 FPS
Xeon E7-8880 v3 @ 2.3 GHz	20.1 FPS	23.4 FPS	26.2 FPS
Core i7-7700K @ 4.2 GHz	52.4 FPS	61.8 FPS	69.9 FPS
Core i7-8750H @ 2.2 GHz	53.3 FPS	60.8 FPS	67.3 FPS
Xeon Platinum 8124M @ 3.0 GHz	203.8 FPS	236.0 FPS	275.5 FPS

NVIDIA GPU video performance

Benchmarks were conducted on Ubuntu Linux using varied systems with NVIDIA GPU Acceleration enabled.

System profile	1080p	720p	480p
NVIDIA Jetson TX-1	6 FPS	15 FPS	44 FPS
NVIDIA Jetson TX-2	14 FPS	34 FPS	86 FPS
NVIDIA Tesla M60	99 FPS	164 FPS	202 FPS
NVIDIA GeForce GTX 1060	165 FPS	191 FPS	206 FPS
NVIDIA Tesla V100	184 FPS	329 FPS	364 FPS
NVIDIA GeForce RTX 2080	221 FPS	297 FPS	351 FPS

Prerequisites

• Rekor Scout® Basic or Pro license/subscription

• Ubuntu 22.04, Ubuntu 20.04, Ubuntu 18.04, Windows 10, or Windows 11

• Python (2 or 3)

Installation

Generic

1. Download the Vehicle Recognition SDK
2. Clone this repository git clone https://github.com/openalpr/speed_benchmark.git

3. Install the Python requirements pip install -r requirements.txt

Docker

docker run -it --rm -v /etc/openalpr:/etc/openalpr/ openalpr/commercial-agent /bin/bash
apt update && apt install -y curl python-pip git
git clone https://github.com/openalpr/speed_benchmark.git
cd speed_benchmark/
pip install -r requirements.txt
bash <(curl https://deb.openalpr.com/install)  # Select SDK

Usage

1. View all command line options by running python speed_benchmark.py -h

2. Select your desired resolution(s) - vga, 720p, 1080p, and/or 4k

3. Benchmark using the default flags (1 stream and no minimum CPU threshold) by running python speed_benchmark.py

4. Check the average CPU utilization (see sample output below). Resolutions with utilization of less than 95% are bottlenecked on decoding the video stream (typical for higher resolutions). These should be rerun with additional streams for a better estimate of maximum performance

5. Set the --thres to a non-zero value. This causes the program to add streams until the threshold CPU utilization is achieved. We recommend using 90 < thres < 95. On large systems where the CPU utilization for a single stream is much lower than your desired threshold, you can reduce the granularity of the search by setting --steps > 1

6. Estimate the number of cameras for a given total FPS value by using the following per-camera rules of thumb

   • Low Speed (under 25 mph): 5-10 fps

   • Medium Speed (25-45 mph): 10-15 fps

   • High Speed (over 45 mph): 15-30 fps

Sample Output

Using default options

user@ubuntu:~/git/speed-bench$ python speed_benchmark.py
Initializing...
	Operating system: Linux
	CPU model: Intel Core i7-8750H CPU @ 2.20GHz
	OpenALPR version: 2.7.101
	Runtime data: /usr/share/openalpr/runtime_data
	OpenALPR configuration: /usr/share/openalpr/config/openalpr.defaults.conf
Downloading benchmark videos...
	Downloaded vga
	Downloaded 720p
	Downloaded 1080p
	Downloaded 4k
Testing with 1 stream(s)...
	Processing vga
	Processing 720p
	Processing 1080p
	Processing 4k
	Lowest average CPU usage 81.4%
+---------------------------------------------------------+
|        OpenALPR Speed: 1 stream(s) on 12 threads        |
+------------+-----------+-----------+-----------+--------+
| Resolution | Total FPS | CPU (Avg) | CPU (Max) | Frames |
+------------+-----------+-----------+-----------+--------+
|    vga     |    52.9   |    81.4   |    99.4   |  479   |
|    720p    |    49.6   |    84.9   |    99.5   |  479   |
|   1080p    |    44.4   |    88.8   |   100.0   |  479   |
|     4k     |    23.8   |    93.7   |   100.0   |  479   |
+------------+-----------+-----------+-----------+--------+
Saving results to /home/user/git/speed_benchmark/speed-bench-20190618.csv

Starting with 3 streams and incrementing by 2 each time 95% CPU utilization is not achieved

user@ubuntu:~/git/speed-bench$ python speed_benchmark.py --thres 95 --streams 3 --step 2
Initializing...
	Operating system: Linux
	CPU model: Intel Core i7-8750H CPU @ 2.20GHz
	OpenALPR version: 2.7.101
	Runtime data: /usr/share/openalpr/runtime_data
	OpenALPR configuration: /usr/share/openalpr/config/openalpr.defaults.conf
Downloading benchmark videos...
	Found local vga
	Found local 720p
	Found local 1080p
	Found local 4k
Testing with 3 stream(s)...
	Processing vga
	Processing 720p
	Processing 1080p
	Processing 4k
	Lowest average CPU usage 93.2%
Testing with 5 stream(s)...
	Processing vga
	Processing 720p
	Processing 1080p
	Processing 4k
	Lowest average CPU usage 95.3%
+---------------------------------------------------------+
|        OpenALPR Speed: 5 stream(s) on 12 threads        |
+------------+-----------+-----------+-----------+--------+
| Resolution | Total FPS | CPU (Avg) | CPU (Max) | Frames |
+------------+-----------+-----------+-----------+--------+
|    vga     |    66.5   |    95.3   |   100.0   |  479   |
|    720p    |    61.3   |    96.2   |   100.0   |  479   |
|   1080p    |    54.1   |    97.3   |   100.0   |  479   |
|     4k     |    29.5   |    99.2   |   100.0   |  479   |
+------------+-----------+-----------+-----------+--------+
Saving results to /home/user/git/speed_benchmark/speed-bench-20190618.csv