With the benefit of EPSRC funding under an Initiative in Crime Detection and Prevention, we have developed a prototype system capable of delivering surveillance images with much greater detail than is normally possible when image resolution is limited by atmospheric turbulence. A high-speed camera is used to freeze the atmospheric turbulence and the output image is assembled from the sharpest elements of the captured images. We have demonstrated improvements in resolution of up to 4 times, depending on the criteria we chose for selecting the sharpest image details. Moving targets may be tracked and those images displayed when they are at their sharpest. This will help with the identification of individuals or other targets and their intentions. The provision of substantially better quality images will allow automatic object recognition systems to perform much more easily and much more accurately.
In virtually any situation where the distance between camera and target limits the image quality because of atmospheric turbulence we are able to provide images that are typically those that one would obtain with the target at one third to one quarter of its true distance. The system uses largely commercially available hardware components and therefore should be relatively economic to deploy in significant numbers. The hardware is robust and therefore can be ruggedised for use in the field.
We see here selected images using different selection rates. The top image is the sequence average (what you would get with a long exposure image). Others show selections of 10%, and 1%, and 1% with blind deconvolution. The resolution improvement is dramatic. We typically are able to reduce the effective distance of a target by a factor of 3-4 with Lucky alone, and a factor of 8-10 with deconvolution.
These are typical images taken with our present system that showed the "before" and "after" images. These test charts shown were taken at a range of 400 m and the text on the targets has a line spacing of 33 mm. The difference between these images shown here is quite typical of our experience over a wide range of weather conditions including hot summer days in the UK.
The basic principle of the system proposed is that it uses a noiseless, solid-state, high-speed, ultra-low light level camera to record images much faster than conventional TV frame rates so that it freezes atmospheric turbulence. Each image is broken into sub arrays and the sharpest sub images are used to re-synthesise a much higher resolution output image. Within a few seconds it automatically evaluates the conditions relevant to that day, time and location, and sets the optical configuration (magnification, telescope aperture diameter) and image recording parameters (frame rate, isoplanatic cell size, etc.) accordingly. Data are processed in real time allowing either full-frame image enhancement or even faster tracking of a target with the sharpest images recorder being presented to the operator. In essence therefore what we have is a special-purpose TV type camera that produces a much more detailed, higher resolution output image than would otherwise be possible.
A draft of a research paper describing many of our results has recently been produced. To see a copy of the paper in draft form yet giving most of the points of what we have done (in Microsoft Word format), click here.