6 IMAGE BASED INSPECTION SYSTEM REQUIREMENTS AND CAPABILITIES
6.2 Practical requirements for data collection
6.3.1 What are the constraints placed on the system imposed by the image collection process?
Ideally the collection of images would be a simple process using either a vehicle mounted camera which automatically collected images as it passed through or around the bridge, or a handheld camera which the inspector could use to snap images with little or no constraint on the image collection process. However, if the images are to be of acceptable standard, show adequate detail, and represent the size and shape of features accurately and consistently then the image collection process is necessarily more complicated.
For images to provide faithful representations of scenes certain criteria must be met. The following summary is necessary in understanding the decisions leading to a proposed specification for an IBIS.
6.3.1.1 Lighting
The scene being imaged must be adequately illuminated. Too much light will overexpose the image (Figure 33, left), while insufficient light will result in underexposed images (Figure 33, right). Both over- and under-exposed images will fail to show details which may be of interest to the inspector.
Figure 33: Example of over- (left) and under- (right) exposed images. The need for the system to operate with no traffic management or disruption precludes the use of flash illumination and makes it difficult to use any form of illumination routinely. To simplify the development and demonstrate the potential for an IBI approach the decision was made to operate the IBIS with no artificial illumination of any kind, and make use of natural light.
6.3.1.2 Field Of View
The area which will be captured in an image is known as the ‘Field Of View’ (FOV). If the IBIS is to collect images of all visible parts of the structure then it follows that the FOV of the camera must either include the entire structure, or that a number of images must be taken. The larger the FOV of the camera, the fewer images will be needed to cover the whole bridge.
The physical dimensions and the number of individual pixels in the camera sensor are constant for any particular camera model. The angular FOV of a given lens of
fixed focal length does not change. If the camera and lens are used to take an image of a plane surface from a distance of 10m away, and another image of the same plane surface from a distance of 5m away, then the parts of the surface which are included in the images will be different. Figure 34 illustrates this.
Assuming that the camera FOV (
) remains constant, if the surface being imaged (represented by the rainbow coloured rectangles) is located at distance A, then only the two central bands will be included in the image. If the surface being imaged is at distance B then the four central bands will be imaged. Because the same number of pixels is available to represent the imaged scene, each pixel will represent a smaller area when the image is collected at point A, resulting in a more detailed image of a smaller area.Figure 34: Demonstration of effect of imaging distance on area imaged. Camera A B
It is therefore important to make sure that the images of the bridge will be collected at a distance which will produce sufficiently detailed images. In order to increase the level of detail in the images there are two things which can be done: the camera can be moved closer to the bridge, or the camera could use a lens with a narrower FOV. Moving the camera closer to the bridge is not always possible due to traffic, obstructions or the size of the bridge.
6.3.1.3 Angle of view
The angle at which the image is taken can have a great effect on the usefulness of the image. If the image is taken at a particularly oblique angle then the area of the scene represented by each image pixel will vary. Pixels representing parts of the scene which were closer to the camera will represent a smaller area than those from further away.
The images in Figure 35 show two views of the same part of a bridge. The image on the left has been taken from close (approximately 2m) to the bridge at a steep viewing angle. The image on the right has been taken from approximately 12m away, with the camera pointing more perpendicularly to the imaged surface. Despite being taken from farther away the image on the right gives a more easily interpreted impression of the condition of the wingwall.
Figure 35: Different views of the same part of a bridge showing the effect of imaging angle.
6.3.1.4 Resolution
The resolution of an image is related to the level of detail which can be seen within the image. A higher resolution image will show more detail than a lower resolution one. There are a number of factors which influence the resolution of the image: the camera sensor; the lens; the camera-object distance.
For a given camera sensor, moving closer to the scene, or choosing a lens with a longer focal length, makes the area of the scene within the camera FOV smaller, and means that each sensor element represents a smaller area of scene, producing a more detailed image. Changing to a lens with a longer focal length consequently has the effect of either producing more detail in the image obtained, or enabling an equally detailed image to be obtained from further away.
The camera sensor can be thought of as an array of individual sensor elements. Light from the scene gets focused by the lens and falls on this array. Each sensor element records the amount of light which it detects, which is then converted into an image. If this array was 1000 x 1000 then there would be a total of 1 million sensor elements, and the resultant image would have 1 million pixels. In practical
terms the camera sensor cannot be changed, other than by selecting a different camera.
If the hypothetical 1 million pixel sensor above was used to image a scene measuring 1m x 1m then each pixel would represent the light detected from an area of 1mm x 1mm. If the camera was used to image a scene of 0.5m x 0.5m then each pixel would represent light from an area of 0.5mm x 0.5mm. The image requirement is to show features as small as 0.4mm (Chapter 2). This does not mean that there must be 1 pixel for every 0.4mm of scene area. Dark, sub-pixel size features will affect the light detected by each pixel, and reduce the reported pixel value compared to its’ neighbours, meaning that the feature will be visible in the output image. Further detail on image resolution and visual acuity is discussed in Section 6.3.2.
6.3.1.5 Focus
Perhaps most crucially, the images must be properly focused. It does not matter how carefully the scene is lit, or what the sensor size of the camera is, if the images are not properly focused then the light from each part of the scene will be spread over a number of sensor elements, resulting in blurry images in which fine details cannot be resolved or detected.
Ensuring all images are in focus is difficult without manually focusing the camera for each image. The Depth Of Field (DOF) of a camera is the distance between the nearest and farthest parts of a scene which are displayed sharply. Only a single distance can be precisely focused at a time, but the decrease in sharpness at either side of this focused distance can be steep or shallow, meaning that more or less of the scene can be tolerably sharp.
Even when viewing planar objects such as bridge abutments, if the camera is not moved to be perpendicular to the plane being imaged, and held at a constant
camera-object distance for all images, then some of the scene will be closer to the camera than others. If the difference in this camera-object distance exceeds the DOF, then parts of the image will appear unfocussed.
The DOF can be increased by decreasing the aperture size of the camera. Decreasing the aperture size lets less light in to the sensor, making the image darker, or requiring additional light sources or a longer exposure. If a longer exposure is used then the image will be more susceptible to blurring due to movement or small vibrations.