GeneA-SLAM: Enhancing SLAM with Genetic Algorithm-Based Feature Points Re-sampling

The uneven distribution of feature points significantly affects the performance of SLAM systems based on feature point matching. In this paper, a novel feature point resampling module based on genetic algorithm is proposed to alleviate the problem of feature point clustering. By integrating this module into the visual front end with the highly scalable ORB- SLAM2 as the global SLAM solution, GeneA-SLAM is developed. Experimental results on the TUM dataset reveal that, compared to ORB-SLAM2, GeneA-SLAM attains a maximum optimization of 19.73% in APE and 18.43% in RPE when taking translational components into account. Considering the rotational component, the maximum optimization rate of RPE reaches 50.10%. Moreover, GeneA-SLAM has a relatively lower runtime than ORB-SLAM2. Furthermore, GeneA-SLAM successfully accomplishes localization and constructs the environmental map on the sculpture dataset taken in a real environment. Overall, experiments confirm that GeneA-SLAM exhibits high accuracy and robustness. The GeneA-SLAM is available at https://github.com/qingshufan/GeneA-SLAM.