We formulate human motion analysis as a high dimensional optimization problem.A generative method is proposed in the framework of evolutionary computation.Firstly

we learn the nonlinear latent space of motion state and use a manifold reconstruction method to establish the inverse mapping.In the search strategy

we apply clonal selection algorithm (CSA) for human pose optimization in the learnt latent space

which proves to be more efficient and accurate.Then

we introduce the temporal information into the traditional CSA and propose a sequential CSA (S-CSA) method for motion tracking.Experimental results show that our method achieves better results than state-of-art methods.