7. Conclusions
We presented a novel generative
approach to reconstruct
3D human pose from a single
monocular image as well as
from monocular image sequences. The main contribution is
to optimize human pose in learnt
latent human motion
space. Pose analysis in the latent
space learnt using ISOMAP
happens to be more ef f icient and accurate. In the search
strategy, we apply the immune genetic algorithm for pose
estimation. A sequential IGA framework is proposed for pose
tracking by incorporating the temporal continuity informa-
tion into the traditional IGA. Compared with GA and PSO,
IGA has the ability to use the prior knowledge of human
motion. Experiment results on different motion types and
image sequences demonstrated that our IGA-based method
for pose estimation is effective to deal
with occlusion, left-
right
ambiguity, and the viewpoint problem. The sequential
IGA method can achieve stable and accurate pose tracking.
Quantitative experiments compared with other state-of-art
methods show that our methods achieve better results.