Adaptive Sample Consensus for Efficient Random Optimization

Lixin Fan, Timo Pylvänäinen

Abstract: This paper approaches random optimization problem with adaptive sampling, which exploits knowledge about data structure obtained from historical samples. The proposal distribution is adaptive so that it invests more searching efforts on high likelihood regions. In this way, the probability of reaching the global optimum is improved. The method demonstrates improved performance as compared with standard RANSAC and related adaptive methods, for line/plane/ellipse fitting and pose estimation problems.