Rectification
Posted by Lenny Rudin June 1, 2017
A surveillance camera is usually positioned at a fixed angle to the stationary or dynamic scene that is being recorded. Body-worn and vehicle-mounted cameras also exhibit a fixed viewing angle for each observed video frame, which can then dynamically vary in time. If that angle is not at 90 degrees, an oblique optical projection distorts the geometrical structure of the recorded objects and subjects. For example, a vehicle and its license plate can be stationary or dynamically foreshortened in ways that makes it difficult to identify and read the license plate. Similarly, a person may be observed at an unnatural angle(s), thus distorting proper 3-dimensional visual perception. Other examples indicating a forensic need for the correction of geometrical camera distortions are oblique crime scene photography of shoe-prints, tire-marks, and tool-marks.
Projective optical camera distortion is described by a simple 'analog' mathematical formula and a perfect inversion formula. However, due to the digital nature of the recorded images, more resolution information is lost in the most oblique direction. A direct application of the analog-based, un-distorted transformation will yield digital interpolation artifacts, such as blurring and/or pixilation.
Cognitech has developed algorithms that exploit multi-directional information correlations that are naturally present. The algorithms also exploit sophisticated, non-oscillatory and artifact-reducing image interpolation to recover the best quality digital images that rectify the observed scene back to the 90 degree viewing angle. Moreover, if the object of interest, e.g. a license plate or a person is non-stationary (moving) and therefore changes its viewing transformation parameters between frames, Cognitech's Rectify plug-in interface is designed to accurately track these changes and transform the changing video sequence to a common rectified viewing angle and scale. This also works if the camera itself is mobile, such as body-worn or vehicle-mounted camera(s).
Cognitech’s Rectify plug-in is synergetic with the Distortion Correction plug-in that corrects curvature-based optical distortions and restores natural straight lines. A future software release will also address the rectification of markings present on 3-dimensional curved surfaces, such as images of body tattoos, writing (baseball cap mottos) and/or patterns (artwork) on clothing.
Here is a link to our Rectification interactive tutorial! Check it out!
https://tutorials.cognitech.com/InteractiveTutorials/Rectify/story.html
Lenny Rudin
Lenny Rudin