The International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences. Vol. XXXVII. Part B3b. Beijing 2008 
2. COLOUR 
Traditional vision metrology employed panchromatic digital 
cameras and ‘white’ retro-reflective targets on a black 
background, to provide the necessary high-contrast bright 
targets against an underexposed dark background (e.g. Ganci 
and Handley, 1998). When recording normally-exposed 
imagery with consumer-grade digital cameras, which are 
typically colour and do not provide a high level of control over 
illumination, white retro-reflective targets are often 
indistinguishable from background noise and ‘hot spots’. To 
overcome this issue, the authors have investigated alternative 
photogrammetric targeting materials, and have discovered one 
far better suited to the scanning of normally-exposed colour 
imagery and hybrid measurement. 
The optimum material found is red 3M™ 3272 Engineering 
Grade Reflective Sheeting. Red provides the best colour 
response because typical outdoor scenes are dominated by 
green vegetation and blue sky, with very little naturally 
occurring red (Cronk, 2007). 3M™ 3272 maintains the well- 
known benefits of retro-reflective material, with the added 
advantage of facilitating the exploitation of colour information 
within the recorded imagery. It is also highly durable, which is 
important for long life, and has a surface conducive to the 
screen-printing of black ink for masking out circular 
photogrammetric targets. This has long been an issue with 
normal ‘white’ retro-reflective material, where the black ink is 
printed directly on the tiny glass beads and is prone to 
scratching, which lowers coded target performance and 
longevity. 
Scanning a colour image for colour targets is a three stage 
process. Firstly, potential target ‘blobs’ are identified, which is 
fundamentally a line-by-line algorithm that considers the 
brightness and colour saturation of pixels (in this case, the 
presence of red over and above green and blue). A target 
‘on/off threshold value determines the start and end of a blob 
on each line, and then line sections are merged to form entire 
blobs. Secondly, each individual blob is refined. The initial 
identification of interest blobs reduces the effective search 
space dramatically, so blob refinement is locally intensive 
without compromising overall scan-time. The two primary 
functions are to split conjoined blobs, and to ‘grow’ tiny yet 
valid blobs so that as much information as possible is preserved. 
Finally, blobs are filtered in an effort to remove noise (i.e. non 
targets) based on a series of radiometric and geometric tests 
involving shape, size, brightness and circularity. Remaining 
‘blobs’ are considered valid targets, and their corresponding 
image measurements calculated via an intensity-weighted 
centroiding routine (e.g. Luhmann et al., 2006). 
The authors have devised a new coded target system using the 
3272 material. The design is based on a five point pattern 
forming a ‘T’ shape, with a subsequent three ‘bits’ occurring at 
various pre-defined locations on concentric circles around the 
bottom two points of the ‘T’. A total of 165 unique code 
combinations exist. The last 20 are reserved as ‘spinners’, 
which are codes that can be ‘spun’ during image recording, 
without compromising the integrity of photogrammetric 
measurement. The adoption of spinners means that fewer codes 
are required to measure a complex network, for example when 
images must be taken around comers, because the essential 
codes that tie the network together will be ‘seen’ in enough 
images for automatic network orientation to take place. The 
new coded target design is illustrated in Figure 1. 
Figure 1. Red coded target design showing 5-point ‘T’ pattern 
and other code ‘bit’ locations. 
3. AUTOMATIC NETWORK ORIENTATION & IMAGE 
POINT CORRESPONDENCE DETERMINATION 
The newly designed red retro-reflective coded target system 
facilitates automatic homologous point determination in multi 
image networks. The preliminary stage of this procedure is 
ideally based on automatic relative orientation (RO), or can also 
be achieved by the use of an exterior orientation (EO) device 
(Fraser & Cronk, 2007). The former approach is adopted by the 
iWitnessPRO system (Photometrix, 2008). 
Selection of the image pair most suitable for initial RO is based 
on three main criteria: the number of common coded-targets 
seen by the candidate image pair; the perspective dissimilarity 
between both images; and finally the area coverage by points 
within each image (Cronk, 2007). Once the initial image pair 
has been oriented, it is then simply a matter of spatially 
resecting the remaining images in the photogrammetric network 
based on the coded-target locations in each image. The initial 
parameters required for bundle adjustment - station EOs and 3D 
point coordinates - are then known. Camera self-calibration can 
also take place at this stage, however this topic will not be 
discussed in this paper; instead the reader is directed, for 
example, to Cronk et al. (2006). 
If the imagery only contains natural features of interest, then the 
sole purpose of this phase is to automatically orient each image 
in the photogrammetric network. Ongoing manual operations 
for hybrid measurement can then take place, because the EO for 
each station is known. If the imagery contains other targeted 
features, i.e. single red dots representing points of interest, then 
automatic detection and subsequent triangulation of these non 
code object feature points can take place via image point 
correspondence determination. This is completely optional for 
hybrid measurement, which is one of its attractive benefits 
compared with traditional automated photogrammetry. 
The image point correspondence determination procedure 
begins by pairing oriented images and ranking them based on 
geometry and the number of commonly ‘seen’ coded target