A comparison of global optimizers applied to point cloud registration

Abstract

The registration of point cloud data is essential in various applications, such as computer vision and robotics. The Iterative Closest Point (ICP) algorithm offered a solution to this problem, with several subsequent methods addressing problems including occlusions and variable point data overlap. To also account for detection errors, the Particle Swarm Optimization - Cardinalized Optimal Linear Assignment (PSO-COLA) point data registration algorithm was introduced. This algorithm offers robust registration solutions in the presence of data miss-detections and false alarms, but being based on a Particle Swarm Optimization (PSO) concept is susceptible to local minima problems. To address this problem, we propose the use of two additional meta-heuristic algorithms, namely Artificial Rabbit Optimisation (ARO) and Artificial Bee Colony (ABC), in combination with the Cardinalized Optimal Linear Assignment (COLA) metric. Our experiments show that the resulting ARO-COLA algorithm reduces the execution time compared with the former PSO-COLA algorithm while maintaining high registration accuracy, especially in scenarios with cardinality and spatial errors. The results indicate that the ARO-COLA algorithm is a promising alternative for efficient and accurate point cloud registration.