Simultaneous Localization and Mapping Based on Semantic Information Optimization

Simultaneous localization and mapping (SLAM) have broad applications such as autonomous driving. However, in practical applications, the autonomous driving environment is often very complex, which often includes pedestrians and moving cars. It tends to lead to misregistration of the odometry. To solve such problems, this paper uses semantic information to fuse the original odometry method to extract feature points. Through this method, the registration accuracy of the odometry is improved and the error is reduced. This facilitates subsequent loop closure detection and map construction in the SLAM system. We compare it to alternative techniques and utilize the KITTI dataset to verify the algorithm's efficacy. The verification outcomes demonstrate that our strategy may significantly increase the system's accuracy.