Chinese Journal of Liquid Crystals and Displays, Volume. 40, Issue 5, 727(2025)
Visual SLAM algorithm based on dynamic feature elimination and dense mapping
Fig. 8. Comparative results of ATE between two systems on TUM dataset
Table 0. [in Chinese]
View tableView in ArticleTable 0. [in Chinese]
算法1 动态特征点过滤算法输入：阈值
；先前帧 ；当前帧 ； 输出：静态特征点合集
； 1.
=Extractor_Keypoints（ ）//先前帧特征点提取 2.
=Extractor_Keypoints（ ）//当前帧特征点提取 3.
=Calculate_Optical_FlowLK（ ， ， ）//计算两帧的稀疏光流 4.
=Estimate_Fundamental_Matrix（ ， ）//估计基础矩阵 5.
for in range（len（ ）） if not Semantic_Prior（
［ ］）then//特征点坐标不与动态物体检测框重合 append
［ ］to else if Semantic_Prior（
［ ］）then if Dynamic_Value（
［ ］< ）and judge（ ［ ］， ）is ture then//对极几何约束与光流法先后判定 append
［ ］to end if
Table 1. Ablation experiment
View tableView in ArticleTable 1. Ablation experiment
实验序号 FasterNet SimAM 参数量 GFLOPs 精确度/% mAP@0.5/% Baseline 315 718 8.9 72.3 75.2 1 √ 193 441 6.2 69.4 73.4 2 √ 318 431 8.9 73.4 76.3 3 √ √ 194 531 6.3 71.4 74.8
Table 2. Performance comparison of different BOW
View tableView in ArticleTable 2. Performance comparison of different BOW
词袋空间大小/MB 平均加载时间/ms 原有词袋 145.3 8 213.5 二进制词袋 44.4 772.4
Table 3. Comparison of ATE experimental results
View tableView in ArticleTable 3. Comparison of ATE experimental results
序列 ORB-SLAM3 本文算法 提升率/% RMSE Mean Median S.D. RMSE Mean Median S.D. RMSE Mean Median S.D. Walking_xyz 0.534 2 0.487 2 0.458 6 0.218 0 0.015 1 0.013 3 0.011 9 0.007 1 97.17 97.27 97.40 94.54 Walking_rpy 1.133 4 0.956 6 1.098 5 0.607 9 0.024 8 0.031 7 0.023 6 0.011 7 96.04 97.81 97.85 98.07 Walking_half 0.360 3 0.284 1 0.243 0 0.221 6 0.028 3 0.025 1 0.023 1 0.013 1 92.14 91.16 90.49 94.08 Sitting_static 0.008 9 0.007 8 0.007 0 0.004 2 0.006 1 0.006 2 0.005 5 0.003 3 31.46 20.51 21.42 21.42
Table 4. Comparison of experimental results of RPE translation part
View tableView in ArticleTable 4. Comparison of experimental results of RPE translation part
序列 ORB-SLAM3 本文算法 提升率/% RMSE Mean Median S.D. RMSE Mean Median S.D. RMSE Mean Median S.D. Walking_xyz 0.816 5 0.637 0 0.514 2 0.510 7 0.022 1 0.019 6 0.018 0 0.010 2 97.29 96.92 96.49 98.01 Walking_rpy 1.770 2 1.403 0 1.077 9 1.079 5 0.063 1 0.047 6 0.037 3 0.041 3 96.43 96.60 96.53 96.17 Walking_half 0.506 5 0.367 3 0.358 7 0.348 8 0.040 0 0.035 9 0.034 4 0.017 6 92.10 90.22 90.40 94.95 Sitting_static 0.014 0 0.012 1 0.010 4 0.007 0 0.010 0 0.008 9 0.008 2 0.004 6 28.57 26.44 21.15 34.28
Table 5. Comparison of experimental results of RPE rotation part
View tableView in ArticleTable 5. Comparison of experimental results of RPE rotation part
序列 ORB-SLAM3 本文算法 提升率/% RMSE Mean Median S.D. RMSE Mean Median S.D. RMSE Mean Median S.D. Walking_xyz 16.079 12.475 9.868 9 10.144 0.612 7 0.487 1 0.417 8 0.371 6 96.18 96.09 95.76 96.33 Walking_rpy 36.607 28.987 21.160 22.356 1.298 8 0.971 8 0.749 5 0.861 8 96.45 96.64 96.45 96.14 Walking_half 10.314 7.553 9 7.438 3 7.023 4 0.895 9 0.800 8 0.745 1 0.401 4 91.31 89.39 89.98 94.28 Sitting_static 0.375 2 0.338 4 0.323 8 0.162 1 0.339 9 0.307 7 0.291 9 0.144 4 9.40 9.07 9.85 10.91
Table 6. Comparison of the error results between proposed algorithm and the same type of algorithms
View tableView in ArticleTable 6. Comparison of the error results between proposed algorithm and the same type of algorithms
Sequence DS-SLAM DynaSLAM SG-SLAM RDS-SLAM Ours RMSE S.D. RMSE S.D. RMSE S.D. RMSE S.D. RMSE S.D. Walking_xyz 0.027 4 0.015 8 0.019 - 0.016 3 0.007 3 0.057 3 0.023 1 0.015 1 0.007 1 Walking_rpy 0.435 7 0.241 1 0.037 - 0.034 5 0.018 4 0.165 4 0.086 3 0.024 8 0.011 7 Walking_half 0.030 3 0.014 9 0.024 - 0.027 1 0.014 5 0.081 1 0.045 8 0.028 3 0.013 1 Sitting_static 0.006 5 0.003 3 0.006 - 0.006 8 0.005 9 0.008 4 0.004 3 0.006 1 0.003 3
Table 7. Real-time comparison of proposed algorithm with similar algorithms
View tableView in ArticleTable 7. Real-time comparison of proposed algorithm with similar algorithms
SLAM系统 平均每帧处理时间/ms ORB-SLAM3 31.42 RDS-SLAM 59.35 DS-SLAM 61.47 SG-SLAM 40.82 DynaSLAM 698.34 Ours 49.73
Get Citation
Copy Citation Text
Heng ZHANG, Lei WANG, Pengchang ZHANG, Jian CHANG, Xing HE. Visual SLAM algorithm based on dynamic feature elimination and dense mapping[J]. Chinese Journal of Liquid Crystals and Displays, 2025, 40(5): 727
Category:
Received: Nov. 13, 2024
Accepted: --
Published Online: Jun. 18, 2025
The Author Email: Lei WANG (leiwang@xaut.edu.cn)
