A new particle tracking algorithm is derived based on consolidated methods with the aim to overcome the current limits encountered with high particle density flows. The proposed method consists in an integration of the relaxation algorithm based on match probabilities into vision-based features association concepts. Hybridization with PIV pre-analysis is suggested to help with parameters estimation. A dual calculation strategy is also developped in order to reduce the amount of spurious vectors. Simulation tests using synthetically generated images are carried out to evaluate the sensitivity of the proposed method to the particle image density, the background noise and the nature of the flow. Three flow configurations with a growing degree of complexity are successively considered: a 2-D flow over a moving-wall, a steady 2-D Lamb-Oseen vortex ring, and a 3-D unsteady homogeneous isotropic turbulence. The ability of the new tracking algorithm to provide better matching performances with high reliability than conventional techniques out of a dense particle image field is demonstrated.