RFEL optimises its distortion correction solution
Above: Original picture atop with ULL Distortion, followed by corrected image underneath.
By significantly streamlining the input video stream management, the corrected output pixel data can be generated as soon as the original distorted feature pixels are presented to the algorithm - making the overall frame latency predominantly a function of the time it takes to get the relevant data from the sensor into the processing core. The Ultra-Low Latency Distortion Correction IP core will be launched at the AUVSI Unmanned System 2015 show on the 5-7 May 2015.
Functionally, the new IP core corrects either barrel or pin-cushion radial lens distortion to produce undistorted images from wide angle lenses, which is particularly useful as a pre-conditioner for image stitching applications. It can also be used to pre-distort an image so that it appears undistorted when projected onto curved and cylindrical surfaces, which is ideal for human-in-the-loop simulation or virtualisation systems. Distortion Correction is easily implemented into existing or new systems for surveillance, reconnaissance, unmanned and piloted air systems, driver and pilot aids and underwater systems.
Radial distortion of up to 25% can be corrected as standard, although the solution can readily scale to higher correction capability on request. The function implements the OpenCV 2.4 radial and tangential 'undistort' function, making it easy to access. It supports all twelve 'undistort' parameters and includes a digital zoom feature to help the user specify the perfect output region of interest. The input format can be up to 16-bit mono, 16-bit YCbCr422, or 24-bit formats. AXI-4 streaming and memory interfaces are supported as standard, while the output is typically the same as the input, with conversion available as an option.
Dr Alex Kuhrt, RFEL's CEO, added, "Distortion Correction solves the problem of the distortion that occurs when using a wide angle lens to capture as much of a scene as possible. Distortion is tiring for humans monitoring the scene and can result in errors. Similarly a distorted scene is hard for computer recognition systems to monitor. By paring latency back to the theoretical minimum for a rastered solution, our Ultra-Low Latency Distortion Correction IP is the latest addition to our world-class range of real time, HD video enhancement solutions that set new standards in performance, with low power consumption and the lowest real-time delays."
The Ultra-Low Latency Distortion Correction is available now as IP that runs on Xilinx Zynq 7000, Artix 7 or Kintex 7 FPGAs, with Altera support available soon. A reference design kit is available to enable Distortion Correction to be rapidly integrated into designs and the core is available as a bit-true MATLAB model to verify anticipated performance giving a rapid path to market.
