Birds Eye View - Performance on PC

Q: How was performance measured?

CPU and memory usage were measured with Linux top, GPU statistics were measured with nvidia-smi, and output frame rate was obtained from internal Bird's Eye View debug traces.

⇦ Performance

Home

Performance/Profiling ⇨

RidgeRun Bird's Eye View (BEV) was benchmarked on a PC with an Intel Core i7-8750H CPU and an NVIDIA GeForce GTX 1050 Ti Mobile GPU and four cameras. In the tested setup, BEV reached 27 FPS with 1920x1080 inputs at 30 FPS, 48 FPS with 1280x720 inputs at 60 FPS, and 58 FPS with 800x600 inputs at 60 FPS. These results show that the software can approach real-time performance on x86 hardware, but achievable output frame rate depends strongly on input resolution and the full pipeline configuration.

This page summarizes the tested hardware and software environment, the measurement method, and the observed CPU, RAM, GPU, and framerate results. Use it as a sizing reference for x86 evaluation, not as a universal guarantee for all PCs.

Benchmark environment

Hardware:

Intel(R) Core(TM) i7-8750H CPU @ 2.20GHz
GeForce GTX 1050 Ti Mobile
16 GB RAM
4 Fisheye USB Cameras

Software:

Ubuntu 18.04 LTS
GStreamer 1.14.5
OpenCV 4.3
Boost 1.65.1

Methodology

We used Linux's top to obtain the CPU/memory usage and nvidia-smi to obtain the GPU stats. The framerate information was obtained from internal Birds Eye View debug traces.

Performance Results

Table 1 presents the performance results for the CPU and GPU usage for three different input/output resolutions in our setup using four input cameras. Even with Full-HD inputs/output, Birds Eye View is able to keep up closely with the input framerate.

Table 1. Performance Results for Birds Eye View on x86: CPU, Memory and GPU usage

Input/Output Resolution	Input Camera Framerate	Output BEV Framerate	CPU Usage (12 Cores)	RAM Usage (16 GB)	GPU Usage	GPU Memory Usage
1920x1080	30 FPS	27 FPS	63.1%	4.5%	28%	70MiB
1280x720	60 FPS	48 FPS	56.3%	2.8%	26%	62MiB
800x600	60 FPS	58 FPS	50.1%	2.6%	24%	57MiB

FAQ

What performance does RidgeRun Bird's Eye View achieve on a PC?: In the benchmarked PC setup, RidgeRun Bird's Eye View reached 27 FPS with 1920x1080 inputs at 30 FPS, 48 FPS with 1280x720 inputs at 60 FPS, and 58 FPS with 800x600 inputs at 60 FPS. These results show that the software can approach real-time surround view performance on x86 hardware, although the exact result depends on resolution, camera setup, and the rest of the pipeline.
What hardware was used for the PC benchmark?: The benchmark used an Intel Core i7-8750H CPU, an NVIDIA GeForce GTX 1050 Ti Mobile GPU, 16 GB of RAM, and four fisheye USB cameras.
What software environment was used for the benchmark?: The measurements were taken on Ubuntu 18.04 LTS with GStreamer 1.14.5, OpenCV 4.3, and Boost 1.65.1.
How many cameras were used in the test?: The test used four input cameras. This is important because Bird's Eye View performance depends on the number of camera streams, their resolution, and the amount of processing required to stitch them into a surround view output.
Why does the output FPS drop at higher resolutions?: Higher input and output resolutions increase the amount of image data that must be dewarped, stitched, blended, and rendered for every frame. That added workload raises CPU and GPU demand, which reduces the maximum sustained output frame rate.
What CPU usage was observed during the benchmark?: CPU usage ranged from 50.1% to 63.1% across the tested resolutions. The highest CPU load was observed with 1920x1080 inputs and output, which is consistent with the larger processing workload.
What GPU usage and GPU memory usage were observed?: GPU usage ranged from 24% to 28%, and GPU memory usage ranged from 57 MiB to 70 MiB in the tested setup. This indicates that the pipeline used GPU acceleration while still depending significantly on CPU resources.
How was performance measured?: CPU and memory usage were measured with Linux top, GPU statistics were measured with nvidia-smi, and output frame rate was obtained from internal Bird's Eye View debug traces.
Which resolution offered the best balance between throughput and system load in this benchmark?: In this specific test, 1280x720 offered a strong balance between throughput and resource usage, reaching 48 FPS with 60 FPS camera inputs while keeping CPU, RAM, and GPU usage below the Full HD test case.