In this paper we present an automatic hybrid matching system mixing images coming from central catadioptric systems and conventional cameras. We analyze three models of hybrid fundamental matrices. We perform experiments with synthetic and real data to test the behavior of these three approaches. The sensitivy to noise of lifted coordinates induces to select the simplest model to build an automatic matching system between these kind of images. Scale invariant features with a simple unwarping tool are considered to help initial putative matching. Then a robust estimation gives an estimation of the hybrid fundamental matrix and allows to detect wrong matches. Experimental results show the feasibility of this system.