Using artificial neural networks for human body posture prediction

Engineers and ergonomists are seeking to exploit the potential of the three dimensional (3D) anthropometric technologies. These technologies make 3D measurements possible and provide us with a more detailed description of the human bodies in comparison with the traditional manual 1D or 2D data processing. In many industrial design cases, there is a need to take into consideration various postures of the human body when the product is designed. This paper presents an approach for transforming measured body data between various postures. In this research the measured human body is substituted by a proper set of landmarks, which is used as a basis of transforming the data, as they are needed to describe specific body postures. Artificial neural networks have been applied to the actual conversion of data. The input is a set of demographic data and the coordinates of the landmarks characterizing a given posture. The output is another set of landmarks characterizing the transformed posture. The artificial neural networks used are based on the principles of back-propagation feed-forward network. The results of transforming the whole body and transforming clustered landmarks are compared and evaluated. Our conclusion has been that the cluster-oriented transformation method is computationally more efficient than whole body-oriented posture transformation method in regenerating of human body postures and the clustering of landmarks lends itself to a reliable method. Relevance to industry: In many industrial design cases, there is a need to take into consideration various postures of the human body when the product is designed. The research presented in this paper will speed the posture transformation of the digital human modeling computationally for design purpose. (C) 2010 Elsevier B.V. All rights reserved.