Information Technology Reference
In-Depth Information
1.4
1.4
1.3
1.3
1.2
1.2
1.1
1.1
1
1
0.9
0.9
0.8
0.8
0.7
0.7
0
0.2
0.4
0.6
0.8
1
1.2
1.4
1.6
1.8
0
0.2
0.4
0.6
0.8
1
1.2
1.4
1.6
1.8
Time (s)
Time (s)
(a)
(b)
1.4
1.4
1.3
1.3
1.2
1.2
1.1
1.1
1
1
0.9
0.9
0.8
0.8
0.7
0.7
0
0.2
0.4
0.6
0.8
1
1.2
1.4
1.6
1.8
0
0.2
0.4
0.6
0.8
1
1.2
1.4
1.6
1.8
Time (s)
Time (s)
(c)
(d)
Fig. 3
Fitting a group of speed values using different methods: (a) the Nelder-Mead method;
(b) the Levenberg-Marquardt method (3-order); (c) the M-estimator; and (d) the Gauss-
Newton method. Dots: real velocity data; solid lines: fitted curves.
longer period of time and so improve the efficiency of gait based ego-motion track-
ing, which is the principal contribution of this paper. Our use of a periodic motion
model to represent gait is similar to Molton
et al.
[19], but we derive the gait model
directly from the images, thus relieving the pedestrian of additional instrumentation
(compass and inclinometer). We have also extended the representation to a trun-
cated series of sinusoidal terms for greater accuracy. As defined, the methodology
is not general but gait-specific because of the form of Eq. (7), the principle of using
a longer term motion model could be applied to other examples of periodic motion.
Here, the gait model is constrained by the fact that the translation and rotation of
the camera between two neighboring frames are less than 0.1 m and 10 degrees,
respectively.
4
Phase 2: Recovering Ego-Motion and Scene Structure Using a
Dynamic Gait Model
The continuous tracking phase has two main stages: prediction and correction. If the
prediction yielded by the gait model is accurate, then this leads either to a smaller
Search WWH ::
Custom Search