Active matching is a method for efficiently finding global consensus in feature matching tasks by dynamically using priors to guide the search process. Unlike traditional methods that first generate candidate matches and then resolve them, active matching integrates image processing into the search for global consensus. This approach reduces the number of image processing operations and computational cost by focusing on regions where true positive matches are most probable. The method uses a dynamic mixture of Gaussians (MOG) to handle image ambiguity and is guided by expected Shannon information gain to select the most informative search steps. The algorithm is demonstrated in a sequential SLAM system for 3D camera tracking, achieving robust, real-time matching even in cases of jerky, rapid motion where traditional methods struggle. The algorithm works by iteratively searching for feature matches in regions that are most likely to contain true positives, updating the search regions based on the results of previous searches. This dynamic updating allows the algorithm to avoid examining low-probability regions, significantly reducing the number of image processing operations. The method uses mutual information (MI) to guide the search process, selecting the most informative features and regions to search next. MI is used to evaluate the information gain from each candidate measurement, and the algorithm terminates when the expected information gain falls below a threshold. The algorithm is implemented using a mixture of Gaussians to represent the probability distribution over feature locations, allowing for dynamic updates as new information is obtained. This approach is more efficient than traditional methods like RANSAC and JCBB, as it reduces the number of image processing operations while maintaining accuracy. The algorithm is tested on real-world data, demonstrating its effectiveness in handling ambiguous and rapidly changing scenes. The results show that active matching can achieve global consensus matching with significantly fewer operations, making it suitable for real-time applications. The method is also computationally efficient, with the number of image processing operations reduced by a large factor, making it suitable for applications with high feature density and dynamic tracking requirements.Active matching is a method for efficiently finding global consensus in feature matching tasks by dynamically using priors to guide the search process. Unlike traditional methods that first generate candidate matches and then resolve them, active matching integrates image processing into the search for global consensus. This approach reduces the number of image processing operations and computational cost by focusing on regions where true positive matches are most probable. The method uses a dynamic mixture of Gaussians (MOG) to handle image ambiguity and is guided by expected Shannon information gain to select the most informative search steps. The algorithm is demonstrated in a sequential SLAM system for 3D camera tracking, achieving robust, real-time matching even in cases of jerky, rapid motion where traditional methods struggle. The algorithm works by iteratively searching for feature matches in regions that are most likely to contain true positives, updating the search regions based on the results of previous searches. This dynamic updating allows the algorithm to avoid examining low-probability regions, significantly reducing the number of image processing operations. The method uses mutual information (MI) to guide the search process, selecting the most informative features and regions to search next. MI is used to evaluate the information gain from each candidate measurement, and the algorithm terminates when the expected information gain falls below a threshold. The algorithm is implemented using a mixture of Gaussians to represent the probability distribution over feature locations, allowing for dynamic updates as new information is obtained. This approach is more efficient than traditional methods like RANSAC and JCBB, as it reduces the number of image processing operations while maintaining accuracy. The algorithm is tested on real-world data, demonstrating its effectiveness in handling ambiguous and rapidly changing scenes. The results show that active matching can achieve global consensus matching with significantly fewer operations, making it suitable for real-time applications. The method is also computationally efficient, with the number of image processing operations reduced by a large factor, making it suitable for applications with high feature density and dynamic tracking requirements.