Gait Analysis and Human Motion Tracking - Intelligent Video Event Analysis and Understanding - page 52

(a) 1st frame (b) 50th frame

Fig. 4 Synthetic images with the feature points superimposed

2 2

1 1

0 0

−1 −1

−2 −2

0 10 20 30 40 50 0 10 20 30 40 50

Frame number Frame number

(a) Without additive noise (b) With noise

Fig. 5 Combination of translation and rotation (the arbitrary units are relative to unknown
parameters and hereafter). Estimating t x in the sequence with random noise added using the
STK-based approach: (a) no additive noise, (b) (0.0,12.0) noise. “-" - ground truth, and “o" -
STK-based estimates.

is to compare results for tracking in phase 2 using the gait model against the bench-
mark STK algorithm from frame 51 onwards. The estimates of the translational ( t x
and t z ) and rotational (

z ) components of motion with added noise of variance 12.0
are illustrated in Fig. 6. Table 1 provides error statistics for the accuracy of feature
location. The mean and standard deviation of the errors in all three estimated pa-
rameters are reduced in most cases when using the gait model, as is also apparent
from Fig. 6 since the gait data, marked with “

θ

" symbols, is closer to the ground
truth marked by the solid line. Hence, the data demonstrates quantitatively that the
gait-based technique recovers the motion parameters and locates the features more
accurately.

×