4000 fps camera with realtime image processing on GPU

Realtime long term video recording from high speed cameras is a problem because of camera data rate. Usually data rate is more than 1000 MPix/s and one have to use RAID or SSD to write video stream. File size for video could be really big and it's one more problem for end user. To tackle these problems we offer solution for realtime video processing on GPU. This technology gives us opportunity to solve all these problems altogether: to process video stream in realtime, to write it to conventional HDD and to make file size 10–20 times less.

We've utilized our high performance video processing SDK for NVIDIA GPUs. In this way we can process each frame and write compressed video stream to HDD in real time. Continuous high-speed video recording requires both high bandwidth and plenty of disk capacity due to huge data rates of high-speed cameras (1000 MPix/s or more). Addressing this issue requires real-time compression of video data stream.

Fastvideo has designed super fast video processing SDK that resolves both problems: by processing each frame of a video stream, we are able to significantly reduce the requirements to performance and capacity of the disk system.

Features of Fastvideo 4000 fps high speed camera

• Resolution 640×512, 8-bit or 12-bit, grayscale
• Frame rate 4000 fps (8 bits) for full resolution
• Realtime data transfer from camera to CPU memory via PCI-Express x4 Gen 2 interface
• C-mount lens adapter
• PCIE x4 cable length 3 m
• Realtime video processing on CUDA
• GUI and drivers for Windows and Linux

What we do at the preliminary stage

• Set camera parameters: image width and height, fps, bit depth (8/12-bit grayscale)
• Camera calibration: Dark frame (frame with black offsets), FFC data
• JPEG parameters: bit depth (8/12), compression quality, restart interval

GPU image processing workflow for monochrome high speed camera

• Raw data transfer from CPU to GPU
• Dark frame subtraction (FPN subtraction)
• Flat-Field Correction
• Denoising
• Curves and Levels with 1D LUT (8/12/16-bit)
• Crop, rotate, resize, sharp
• Historgam for each frame
• Gamma
• Smooth video output to monitor via OpenGL with minimum latency
• MJPEG video compression
• Compressed data transfer from GPU to CPU