Top > Research > Human Centered Design > 3D Measurement of Human Foot Shape based on Statistical Model


3D Measurement of Human Foot Shape based on Statistical Model

[collaborate with Keio Univ. and I-Ware Labolatory Co.]


We human does not have arbitrary 3D shapes. The 3D shape of human body or any particular part of human is obviously limited in some range. In our lab, the databases of human shape (body, head, hand, foot, etc) is being constructed. By analysis of such a database, not only average and variance values but also the statistical possibility of 3D shape is acquired.

In other words, it is possible to make a human shape model from the database based on statistics (e.g. PCA). By utilizing the statistical model, measurement of human shape becomes easier. This idea is applied to human foot shape measurement in this research. Thanks to the statistical foot model, we constructed a very cheap system to measure human foot shape.

System Environment

Two ordinary projectors are set in front and back of the measurement area. They projects randomized color pattern onto the measurement target, so that the surface of the target object can have rich texture. Ten web cameras are distributed around the measurement area. All cameras are calibrated in the common coordinate system in advance.

Theoretically, any 3D shape may be reconstructed by general multi-view stereo method in this environment. However, the accuracy of the measurement and the stability of the process are not enough in actual cases of multi-view stereo methods. In this research, the shape measurement problem is replaced with the parameter optimization problem. The parameter space is defined by PCA of real measurement data, which are measured by other high-cost measurement systems. In the optimization process, the parameter is evaluated by the multi-view stereo principle like this: "If this parameter is correct, then the foot shape denoted by this parameter must exist in the target area. Then, the texture on the surface must be consistent in the all capturing cameras." In this way, the parameter optimization is equivalent to efficient multi-view stereo method. It reconstructs the foot shape using a process in low degree-of-freedom.


Example of the measurement

These experimental results below are showing measurement of real human foot. In our experiments using plaster feet, the average error was approx. 1.25mm.

img1.png   img2.png   img3.png   img4.png   img5.png   img6.png


  1. "PCA-based 3D shape reconstruction of human foot using multiple viewpoint cameras", E. Amstutz, T. Teshima, M. Kimura, H. Saito, M. Mochimaru, International Journal of Automation and Computing, Vol.5, Issue 3, pp.217-225, Jul. 2008
  2. "Human Foot Reconstruction from Multiple Camera Images with Foot Shape Database", J. Wang, H. Saito, M. Kimura, M. Mochimaru, T. Kanade, IEICE Transactions on Information and Systems, Vol. E89-D, No.5, pp.1732-1742, May. 2006

Top ∆