3.1.3. Camera Calibration When using photographic ...

7 downloads 5524 Views 712KB Size Report
It is possible to eliminate curvilinear distortion using software such as Photoshop or other open source applications. However, as the correction procedure.
3.1.3. Camera Calibration When using photographic equipment as a means for gathering data, the possible short comings of the equipment needs to be taken into account and eliminated in order for the results to be valid. Radial distortions (also referred to as curvilinear distortion) of photographs arise from the symmetry of a photographic lens and are usually classified into two main types; Barrel distortion and Pincushion distortion. Zoom lenses, such as the one fitted on the Sony R1, are particularly notorious for exhibiting pronounced distortion characteristics Barrel distortion occurs when the image magnification decreases with distance from the optical axis resulting in an image that appears to have been mapped around a sphere or barrel (Figure 31, middle)(K.T. Gribbon, 2003). A

B

C

Figure 1 Distortion of a rectangular grid. (A) Undistorted grid. (B) Barrel distortion. (C) Pincushion distortion.

Pincushion distortion is the opposite of barrel distortion and appears when the image magnification increases with the distance from the optical axis. The visible effect is that lines that do not go through the centre of the image are bowed inwards, towards the centre of the image as shown on right hand of Figure 31. Although an optical lens can exhibit both of the distortion types, wide angle lenses (focal length of 35mm or less) are especially prone to ‘barrelling’ whilst telephoto lenses (focal length of 50mm or greater) show pincushion distortion characteristics. The simplest method of assessing the distortion characteristics of a zoom lens is by taking photographs of a rectangular grid at different focal lengths and checking for warping of the horizontal and vertical lines. Since the camera that has been used has a permanently fixed lens with a focal length ranging from 24mm to 120mm, photographs of the grid were taken at 24, 28, 35, 50, 70 and 120mm. These focal lengths were chosen as they were clearly marked on the zoom ring of the lens.

Looking at Figure 32 it can be seem that the lens shows an abnormal level of barrel distortion at 24mm. The barrelling effect is most apparent at the edges of the image where the horizontal and vertical lines bend away from the red reference lines that are perfectly straight. As the lines get closer to the centre of the image, the distortion tends to diminish.

24mm

Figure 2 Lens distortion test at 24mm focal length. (dpreview.com)

At a focal length of 120mm, there is no measureable distortion. The lines of the grid are perfectly parallel to the red reference lines. This is interesting since a degree of pincushion distortion was expected at the telephoto end of the lens.

120mm

Figure 3 Lens distortion test at 120mm focal length. (dpreview.com)

It is possible to eliminate curvilinear distortion using software such as Photoshop or other open source applications. However, as the correction procedure

reinterprets/interpolates the pixel data, degradation of the image quality with regards to sharpness and contrast is inevitable. And since image analysis applications depend primarily on the pixel contrast for object detection, any form of software based distortion correction is undesirable. It is therefore good practice to remove distortion at the image capture stage by choosing a focal length with minimal distortion – 120mm in this case. It is also important to ensure that the object being photographed is captured at the correct angle of view so that only the plane of interest is visible to the camera. Figure 34 is an illustration of 3 objects being photographed where only the top edge is of interest. But as the subject moves away from the optical axis, the vertical edge becomes visible to the image sensor.

Angle of view Optical Axis

Subject

Figure 4 illustration of angle of view

This effect can be better seen in the Figure 35 below. Nine AA batteries were photographed with the cameras image sensor at a distance of 1.0m from the top of the

Figure 5 Photograph of nine cylindrical AA batteries taken at 120mm focal length with a 1m distance between the batteries and image sensor

batteries. Five of these were placed very close to the optical axis and remaining four, further out. It is clear from the photo that the projection from the four batteries furthest away from optical axis is not parallel. Instead, planes that are orthogonal to the image sensor are visible. On the other hand, the batteries that are in and around the optical axis (occupying approximately 10% of image centre) only project the plane that is parallel to the image sensor. As a result only the centre 10% of the frame was used for the characterisation process.