Abstract: A SURF based three-dimensional boundary recognition algorithm has been devel-oped for identifying and reconstructing three-dimensional geometry of arbitrary solid boundaries. This method is used in conjunction with MLOS-SMART three-dimensional particle image recon-struction and three-dimensional cross-correlation algorithms to simultaneously measure the veloc-ity field as well as the boundary geometry.The boundary recognition algorithm was firstly veri-fied by using a set of cylinder images where the exact curvatures were known.The three-dimen-sional velocity field of a cylinder wake was then calculated by CFD and used for generating syn-thetic particle image sets.Finally,the validity and accuracy of the algorithms were verified by processing the synthetic images and comparing the calculated velocity filed and boundary geome-try with CFD data and the exact cylinder dimensions.