X hits on this document

PDF document

A Prototype Optical Tracking System Investigation and Development - page 129 / 170

433 views

0 shares

0 downloads

0 comments

129 / 170

8.3 Pose algorithm testing (3D testing)

115

1

2

3

+

+

+

282

281.8

281.6

265

264.5

264.4

264.3

264.9

264.8

264.7

264.6

Figure 8.11 Two groups of centroids are shown corre- sponding to one marker. Two groups can be produced if there is sufficient background noise.

Figure 8.12 A ROI that moves back and forth by one pixel can be broken into three parts. When the ROI is to the left it consists of the left part and the centre part. When the ROI is in the right it consists of the right part and centre part.

of centroids to occur. The image sensor has a black level subtraction mode [78] that allows the average dark pixel value to be subtracted from each pixel. With this enabled, the effects of background noise were reduced substantially [96].

8.3

Pose algorithm testing (3D testing)

This section covers testing the pose algorithm in isolation. In Figure 8.1 this corresponds to testing the system between points 1 and 2. The intention is to determine how well it performs under ideal conditions, using synthetic data. The synthetic data represents the test configuration used in Section 8.1 and does not have any noise overlaid onto it.

Figures 8.13a,b show two examples of the the algorithm’s behaviour over iterations. Figure 8.13a shows an example of the algorithm converging well.

a number of s the number

of iterations increases the point cloud variance (F (q), defined in Chapter 7) decreases.

s

the variance decreases so is approximately 0.005 m.

does the position error. The position error after The algorithm has reached a local minimum for

the final step F (q) as there

is a flat part of the converge further.

graph

where

despite

the

reduction

in

step

size,

the

algorithm

doe

not

Figure 8.13b shows an example of poor performance from the algorithm. In this case the

algorithm stops converging early.

lthough the variance decreases monotonically the po-

sition

error

reaches

a

minimum

and

increases

slightly

again.

In

this

example

the

mini-

mum position error is 0.026 m.

Clearly this is much worse than the first example.

Further

Document info
Document views433
Page views433
Page last viewedSat Dec 10 06:39:52 UTC 2016
Pages170
Paragraphs6307
Words54996

Comments