An experimental system for three-dimensional image analysis from twodimensional input has been designed jmplemented and evaluated. The present system is given the acronym EXDIMASIS: Experimental system for three-Dlmensional iMAge analysis from two-dimensional input. The system first task is to acquire the scene frames that represents different instances of the object motion in space. Then a motion-based segmentation algorithm, which was proposed in this work, is applied to extract the object of interest and obtain a binary image. From this binary image clean silhouettes and/or boundaries are extracted. Using these silhouettes and boundaries, certain features „ depending on the moment's calculation, are then computed. Finally on the bases of these features, recognition algorithms are designed to identify a threedimensional object and to estimate its position and orientation in space. The moment normalisation algorithm that used for invariant feature extraction is evaluated by comparing it with other normalisation algorithms in terms of computational cost and performance. Computation cost is evaluated by the computation time and search time. Performance is completely described by the probability of misclassification or the classification accuracy. The immunity of the system to noise is also considered.