In a multiagent simultaneous localization and mapping (SLAM) system, multiple agents actively exchange information with the centralized system, which creates a 3D map by merging the information taken from the agents. If the number of agents increases, handling the agents' data will become challenging for the system. A swift multiagent SLAM system builds active connections between the agents at the first instance of loop closure. Furthermore, it requires minimal information exchange to handle different types of agents. We firstly propose an efficient and novel method, PCR-Pro, for different scales and sparse 3D point clouds registration that cannot be handled by the current popular ICP approaches. The good estimation of transformation and scale helps in the calculation of the covariance...[ Read more ]

In a multiagent simultaneous localization and mapping (SLAM) system, multiple agents actively exchange information with the centralized system, which creates a 3D map by merging the information taken from the agents. If the number of agents increases, handling the agents' data will become challenging for the system. A swift multiagent SLAM system builds active connections between the agents at the first instance of loop closure. Furthermore, it requires minimal information exchange to handle different types of agents. We firstly propose an efficient and novel method, PCR-Pro, for different scales and sparse 3D point clouds registration that cannot be handled by the current popular ICP approaches. The good estimation of transformation and scale helps in the calculation of the covariance matrix and information matrix for pose graph optimization. We further develop a generic method, Loop-box, for the keychallenging scenarios in multiagent 3D mapping based on different camera systems. Based on the initial matching, our system can calculate the optimal scale difference between multiple 3D maps and then estimate an accurate relative pose transformation for large-scale global mapping.

Deep learning-based image retrieval techniques for the loop closure detection demonstrate satisfactory performance on pre-trained datasets. However, it is still challenging to achieve high-level performance based on previously trained models on a different dataset. The general baseline approach uses additional information, such as GPS, sequential keyframes tracking, and re-training the whole environment, to enhance the recall rate. To avoid this, we present an intelligent method, MAQBOOL, to magnify the power of pre-trained models for better image recall. We use spatial information to improve the recall rate in image retrieval probabilistically on pre-trained models. Moreover, we achieve comparable image retrieval results at a low descriptor dimension (512-D), compared to the high descriptor dimension (4096-D) of state-of-the-art methods.