The paper introduces an active matching algorithm that dynamically guides feature matching to achieve global consensus with fewer image processing operations and lower computational cost. Unlike traditional methods that post-process candidate matches, this algorithm uses priors to dynamically guide the search for feature matches, reducing the number of image regions that need to be searched. The approach is based on a fully Bayesian algorithm that uses expected Shannon information gain to guide the search process. The algorithm is demonstrated in a sequential SLAM system for 3D camera tracking, showing robust and real-time performance even in cases of rapid and jerky camera motion. The key contributions include a dynamic mixture of Gaussians (MOG) representation to handle multiple hypotheses during active search and an information-theoretic approach to guide the search process. The results show that active matching can achieve the same accuracy as existing methods but with significantly reduced computational requirements, making it suitable for real-time applications.The paper introduces an active matching algorithm that dynamically guides feature matching to achieve global consensus with fewer image processing operations and lower computational cost. Unlike traditional methods that post-process candidate matches, this algorithm uses priors to dynamically guide the search for feature matches, reducing the number of image regions that need to be searched. The approach is based on a fully Bayesian algorithm that uses expected Shannon information gain to guide the search process. The algorithm is demonstrated in a sequential SLAM system for 3D camera tracking, showing robust and real-time performance even in cases of rapid and jerky camera motion. The key contributions include a dynamic mixture of Gaussians (MOG) representation to handle multiple hypotheses during active search and an information-theoretic approach to guide the search process. The results show that active matching can achieve the same accuracy as existing methods but with significantly reduced computational requirements, making it suitable for real-time applications.