MRTech solutions
MRTech presents a series of ready-made hardware/software solutions for machine vision systems.
All the cases are proven in real projects and open for modifications to meet customer’s requirements.
Image delivery for AI
This solution is designed to capture images from multiple machine vision cameras, pre-process the images with required processing pipeline, and then export the images to a user’s Machine Learning or AI application.
Image delivery for AI works on a wide range of CPU / GPU systems.
Standard or custom image processing pipeline:
- Image acquisition from cameras
- Color pre-processing on GPU
- Automatic control of exposure time and white balance
- Image format transformation, gamma correction, denoise, and other
- Image export to the client code
x86 platform, top-end NVIDIA GPUs:
- High resolution and multiple cameras
- Linux, Windows OS
- Unprecedented processing performance:
- 2000+ FPS @ Full HD
- 500+ FPS @ 4K
- 160+ FPS @ 8K
NVIDIA Jetson:
- Machine vision cameras
- MIPI camera modules
- Low latency image processing
- High performance (e.g. for AGX Orin):
- 240+ FPS @ Full HD
- 120+ FPS @ 4K
Qualcomm ecosystem:
- Machine vision / MIPI cameras
- Low power AI systems
- IFF SDK inside with support of:
- Linux OS
- Vulkan image processing
IFF SDK free test drive
Jetson platform
AI with Qualcomm
Ultra-low latency streaming
This ready-to-use solution allows you to stream images from a remote computer with Glass-to-Glass latency of less than 50 ms for 4K and 25 ms for 1080p images.
Standard or custom image processing pipeline:
- Acquisition from machine vision or MIPI cameras
- Color pre-processing on GPU
- Auto- exposure and white balance
- Image transformation, gamma correction, denoise when needed
- H.264 / HEVC (H.265) encoding and decoding
- Ultra-low latency 8/10-bit RTSP streaming
System with NVIDIA GPU:
- High-resolution and multiple cameras
- Linux, Windows OS
- Top-end streaming performance:
- 4K240 (H.265)
- 8K60 (H.265) and more
NVIDIA Jetson:
- Machine vision / MIPI cameras
- 2x 4K60 H.265 encoder performance
- Glass-to-Glass latency for AGX Orin:
- 25 ms for 1080p images
- less than 50 ms for 4K images
WebRTC streaming:
- Streaming using NVIDIA GPU
or NVIDIA Jetson - Receive and view in any browser
- Remote control via HTTP API
- Less than 60 ms G2G latency
High-speed image recording and post-processing
The solution allows recording raw image streams from one or multiple cameras with a bandwidth of more than 10 GB/s using a carefully chosen hardware setup.
The main features of this solution are recording without frame loss, live preview, raw histogram, camera control, camera color calibration and other.
Additional application also applies cinematic quality offline processing to previously recorded high-resolution raw images using state-of-the-art demosaicing, color correction, and denoising technologies.
By using MRTech IFF SDK this solution is compatible with various professional standards and applications, such as CinemaDNG raw footage format, TIFF dailies, ACES workflow (including pre-processed EXR file format), DaVinci Resolve video editing software.
Configuration example:
- 8 advanced machine vision cameras with fast interface such as PCIe Gen3 or 10GigE
- High-performance computer with top Intel CPU (AMD Ryzen) and SSD NVMe M.2 storage array
- Recording capabilities:
- RAW / DNG / TIFF image format
- 8x 4K (3840 x 2160) @ 10-bit @ 85 FPS
In another experiment we measured 4.7 GB/s bandwidth while recording a 300 FPS 10-bit footage from 12 MPix machine vision camera produced by XIMEA (watch Youtube video about this below). Second camera can be easily added, bringing total bandwidth close to the NVME RAID write speed limit of 10 GB/s.
Even higher speeds are possible with carefully chosen hardware configuration. Our solution guarantees no frame loss and includes live preview, color calibration workflow, high quality offline post-processing with output to DNG, EXR or TIFF format.
Application areas
Teleoperation
Aerospace industry
