[{"data":1,"prerenderedAt":485},["ShallowReactive",2],{"method-yan2026_underground3dgsslam":3},{"method":4,"reference":54,"equipment":77,"figures":113,"results":114},{"id":5,"label":6,"shortName":7,"title":8,"year":9,"era":10,"cluster":11,"scope":12,"keyIdeaZh":13,"keyIdeaEn":14,"fulltextStatus":15,"publicationStatus":16,"recommendation":17,"constructionRelevance":18,"validationEnvironment":19,"strengths":23,"limitations":29,"sensors":37,"platform":39,"estimator":41,"association":42,"timeModel":43,"deskew":44,"loopClosure":45,"globalOptimization":46,"mapRepresentation":47,"prior":48,"outputGeometry":49,"compute":50,"codeUrl":51,"codeLicense":52,"relatedVersions":53},"yan2026_underground3dgsslam","Yan et al., 2026b","Underground RGB-D 3DGS SLAM","RGB-D Perception-Enhanced 3D Gaussian Splatting SLAM: A Robust Framework for Mapping Underground Spaces",2026,"recent","C09","full_slam_with_global_correction","本研究提出只用低成本 RGB-D 相機的地下空間 3DGS SLAM。前處理以 MSRCR、側窗濾波與 HIS 色彩空間正規化 gamma 校正增強低照度影像，並以預訓練深度補全網路（非局部傳播架構，未在地下資料上微調）填補深度空洞；追蹤以渲染色彩與深度殘差最佳化位姿，並以混合歐氏距離（0.3 m 或 15 影格）與重疊度兩階段選取關鍵影格；建圖以不透明度門檻（τ0 = 0.15）與加權觀測次數低於 5 剔除高斯；偵測迴圈後以 Levenberg-Marquardt 求解位姿圖，並剛性更新高斯。資料由搭載 Kinect2 的 Autolabor-Pro1 四輪機器人在煤礦巷道、車庫與地下室共 9 個場景（38.4 至 101.3 m²）蒐集，另以 TUM RGB-D 三序列驗證。軌跡真值來自 LiDAR、IMU 與相機融合的 SLAM 流程，並非獨立測量；地圖只以新視角 PSNR、SSIM、LPIPS 評估，作者明言沒有稠密三維幾何真值。9 個場景中本方法 ATE 為 4.8 至 9.6 cm，有 6 個場景在稠密方法中最低或並列最低，但 ORB-SLAM2 在 4 個場景更低、2 個場景持平。","RGB-D 3DGS SLAM for underground spaces with fixed-parameter low-light enhancement, a pretrained depth-completion network, two-stage keyframe selection, opacity and observation-count Gaussian pruning, and LM pose-graph loop closure. Evaluated on nine self-collected coal-mine, garage and basement sequences (Kinect2 on an Autolabor-Pro1 robot) with ATE against a LiDAR-IMU-camera SLAM reference, plus TUM RGB-D; maps are assessed only by novel-view PSNR, SSIM and LPIPS.","full_text_reviewed","peer_reviewed_published","main_body","場域為煤礦巷道（回採巷道、運輸巷道、行人巷道、受限通道）、廢棄車庫與電動機車車庫，以及泵房、配電室與簡報室等地下室空間，皆為營運中或既有設施，非施工中工地，對隧道與地下設施數位孿生有參考價值。軌跡以多感測器 SLAM 結果為參考，地圖只評估渲染品質而缺少三維幾何精度；深度補全區域屬學習推定幾何，論文未另行標記（推論）。",[20,21,22],"underground_or_tunnel","completed_building","public_benchmark",[24,25,26,27,28],"13.8% higher PSNR than the best baseline in the underground tests (abstract; Table IV)","Lowest or tied-lowest ATE among the eight tested NeRF and 3DGS SLAM methods in 6 of 9 field sequences, e.g., 4.8 cm on the pedestrian walkway (Table III)","About 8.6 FPS with 10.2 GiB peak GPU memory; authors state this is comparable to SplaTAM and MonoGS but slower than GS-ICP SLAM and RTG-SLAM (Sec. IV-C; Table VII)","Removing enhancement, completion and loop closure raises ATE from 8.2 to 14.9 cm and lowers PSNR from 22.4 to 18.2 dB (Table VIII)","Field dataset released at huggingface.co\u002Fdatasets\u002FMISPM\u002FUnderground_RGB-D (Sec. IV-A)",[30,31,32,33,34,35,36],"No 3D geometric accuracy is evaluated; authors state dense geometric ground truth is unavailable and assess maps only by novel-view rendering (Sec. IV-A; Sec. IV-B2)","ATE reference comes from a multi-sensor (LiDAR, IMU, camera) SLAM pipeline whose sensor models and accuracy are not quantified; the ATE alignment method is not stated (Sec. IV-A)","ORB-SLAM2 attains lower ATE than the proposed method in 4 of 9 field sequences and ties in 2; on TUM RGB-D ORB-SLAM2 is better on all three sequences (Tables III and VI)","About 8.6 FPS does not meet higher frame-rate robotic needs; preprocessing (72.5 ms per frame) is the bottleneck (Sec. IV-C; Table V; Sec. V)","Ablation text and Table VIII disagree (11.0 cm ATE paired with 22.2 dB in the text, but with 20.6 dB in the table), and the ablation scene is not identified (Sec. IV-E; Table VIII)","Local artifacts in top views under severe illumination imbalance and repetitive textures (Sec. IV-B3)","(inference) Depth-completed regions are network predictions and are not flagged in the map; no code link is given",[38],"RGB-D camera (Kinect2, 512x424 pixels, 10 Hz, 70 x 60 deg field of view)",[40],"wheeled mobile robot (Autolabor-Pro1, four-wheel drive) carrying a Kinect2 RGB-D camera","Per-frame gradient-based pose optimization against colour and depth rendered from the Gaussian map (pixels with silhouette S(p) > 0.99 and valid depth; weighted colour and depth residuals), then sliding-window Gaussian optimization with poses fixed","Direct photometric and geometric rendering residuals against enhanced RGB and completed depth; keyframes by hybrid Euclidean pose distance (0.3 m or 15 frames) then overlap ratio; loop candidates by cosine similarity and geometric overlap following GLC-SLAM","discrete per-frame poses; initial pose from a motion model (type not specified)","not_applicable","Loop frames detected by cosine similarity and geometric overlap ratio (strategy similar to GLC-SLAM [58]); relative loop poses added as pose-graph edges","Keyframe pose graph with odometry and loop edges minimized by Levenberg-Marquardt on the Lie group; Gaussian means and covariances rigidly updated with each keyframe correction","3D Gaussian ellipsoids (following GSORB-SLAM [56]) with isotropic scale regularization; pruning by an opacity threshold (tau0 = 0.15; the text words it as transparency below zero or above 0.15) and by a weighted observation count below 5","Depth-completion network with non-local spatial propagation, pretrained on unnamed public RGB-D datasets and not fine-tuned on underground data; fixed-parameter image enhancement (MSRCR, side window filtering, normalized gamma correction in HIS space)","3D Gaussian map; no 3D geometric accuracy evaluated (authors state dense geometric ground truth is unavailable); map quality reported only by novel-view PSNR, SSIM and LPIPS on held-out non-keyframes","Processing on a desktop (Intel i7-14700K, RTX 4090D, 64 GB DDR5, Ubuntu 18.04, Python and PyTorch); about 8.6 FPS; per-frame preprocessing 72.5 ms, tracking 36.2 ms, mapping 41.8 ms; peak GPU 10.2 GiB; robot carries an Autolabor-PC control console (Ryzen 3 3200G, 8 GB)",null,"not_verified",[],{"id":5,"kind":55,"shortName":7,"title":8,"authors":56,"year":9,"venue":64,"venueType":65,"publisher":66,"volumeIssuePages":67,"doi":68,"arxivId":51,"url":69,"firstPublicDate":70,"publicationStatus":16,"metadataStatus":71,"fulltextStatus":15,"era":10,"classicReason":44,"codeUrl":51,"cluster":11,"topics":72,"mdpi":73,"verification":74,"label":6,"fulltextRoute":75,"versionRead":76,"addedByCensus":73},"method",[57,58,59,60,61,62,63],"Tao Yan","Xiaohu Lin","Wanqiang Yao","Bolin Ma","Qianjin Cheng","Yinan Gao","Zhiyue Jiang","IEEE Transactions on Visualization and Computer Graphics","journal","IEEE","32(7), 6695-6711","10.1109\u002Ftvcg.2026.3686359","https:\u002F\u002Fapi.crossref.org\u002Fworks\u002F10.1109\u002Ftvcg.2026.3686359","2026-04-22","metadata_verified",[11],false,"corrected","NTU institutional (curl)","Version of record, IEEE Xplore PDF (IEEE TVCG 32(7), pp. 6695-6711, July 2026; 17 pages incl. biographies)",[78,86,91,96,101,107,110],{"category":79,"model":80,"canonical":81,"role":82,"dataset":83,"specs":84,"locator":85},"rgbd","Kinect2","Microsoft Kinect v2","method input","Underground_RGB-D (authors' field test dataset)","Sampling rate 10 Hz; resolution 512x424 pixels; horizontal 70 deg, vertical 60 deg","Table I; Sec. IV-A; Fig. 4",{"category":87,"model":88,"canonical":88,"role":82,"dataset":83,"specs":89,"locator":90},"platform","Autolabor-Pro1","Mobile robot, four-wheel drive; displacement speed 0.5 to 1.5 m\u002Fs; angular velocity 0.56 rad\u002Fs","Table I; Fig. 4",{"category":92,"model":93,"canonical":93,"role":82,"dataset":83,"specs":94,"locator":95},"compute","Autolabor-PC (control console)","CPU AMD Ryzen3 3200G, DDR4 8GB; listed in Table I as the control console of the data-collection platform, while all SLAM experiments ran on the desktop computer (Sec. IV-A)","Table I",{"category":92,"model":97,"canonical":97,"role":98,"dataset":51,"specs":99,"locator":100},"Custom desktop computer (Intel i7-14700K, NVIDIA GeForce RTX 4090D)","compute for runtime","CPU Intel i7-14700K; GPU NVIDIA GeForce RTX 4090D; DDR5 64GB; Ubuntu 18.04","Table I; Sec. IV-A; Sec. IV-C",{"category":102,"model":103,"canonical":103,"role":104,"dataset":83,"specs":105,"locator":106},"lidar","not_reported (LiDAR in ground-truth fusion pipeline)","reference or ground truth","not_reported","Sec. IV-A",{"category":108,"model":109,"canonical":109,"role":104,"dataset":83,"specs":105,"locator":106},"imu","not_reported (IMU in ground-truth fusion pipeline)",{"category":111,"model":112,"canonical":112,"role":104,"dataset":83,"specs":105,"locator":106},"camera","not_reported (camera in ground-truth fusion pipeline)",[],{"totalRows":115,"groupCount":116,"groups":117,"others":484},16,4,[118,346,380,463],{"slug":119,"group":120,"sourceId":5,"sourceLabel":6,"table":121,"selfRows":122,"metrics":123,"seqs":128,"entrants":150,"cells":179,"outcomes":340,"locators":341,"hardware":342,"wordings":343,"notes":344},"yan2026-underground3dgsslam-table-iii","yan2026_underground3dgsslam:Table III","Table III",9,[124],{"label":125,"unit":126,"statistic":127,"alignment":105},"ATE [cm]","cm","RMSE",[129,132,134,136,138,141,143,146,148],{"dataset":83,"sequence":130,"environment":131},"Extraction roadway (E-r)","coal mine scene",{"dataset":83,"sequence":133,"environment":131},"Haulage roadway (H-r)",{"dataset":83,"sequence":135,"environment":131},"Pedestrian walkway (P-w)",{"dataset":83,"sequence":137,"environment":131},"Confined passageway (C-p)",{"dataset":83,"sequence":139,"environment":140},"Abandoned garage (A-g)","garage scene",{"dataset":83,"sequence":142,"environment":140},"Electric scooter garage (E-s-g)",{"dataset":83,"sequence":144,"environment":145},"Pump house (P-h)","basement scene",{"dataset":83,"sequence":147,"environment":145},"Power distribution room (P-d-r)",{"dataset":83,"sequence":149,"environment":145},"Briefing room (B-r)",[151,155,158,161,164,167,169,171,173,176],{"name":152,"methodId":153,"linkable":154,"proposed":73,"self":73},"NICE-SLAM","niceslam2022",true,{"name":156,"methodId":157,"linkable":154,"proposed":73,"self":73},"Co-SLAM","coslam2023",{"name":159,"methodId":160,"linkable":154,"proposed":73,"self":73},"ESLAM","eslam2023",{"name":162,"methodId":163,"linkable":154,"proposed":73,"self":73},"SplaTAM","splatam2024",{"name":165,"methodId":166,"linkable":154,"proposed":73,"self":73},"MonoGS","monogs2024",{"name":168,"methodId":51,"linkable":73,"proposed":73,"self":73},"GS-ICP SLAM",{"name":170,"methodId":51,"linkable":73,"proposed":73,"self":73},"RTG-SLAM",{"name":172,"methodId":5,"linkable":154,"proposed":154,"self":154},"Ours",{"name":174,"methodId":175,"linkable":154,"proposed":73,"self":73},"ElasticFusion","elasticfusion2015",{"name":177,"methodId":178,"linkable":154,"proposed":73,"self":73},"ORB-SLAM2","orbslam2_2017",[180,184,187,190,193,195,198,201,204,207,209,211,213,215,217,219,221,223,225,227,228,230,231,233,235,236,238,240,241,243,245,247,249,251,253,255,256,258,260,262,264,265,267,269,271,273,275,276,278,280,281,282,284,286,288,289,291,292,294,295,296,297,299,301,302,303,305,306,307,309,310,312,313,315,316,318,319,320,322,323,324,326,327,329,331,332,334,335,336,338],[181,181,181,182,183,181,183,183,181],0,18.9,-1,[181,181,185,186,183,181,183,183,181],1,16.4,[181,181,188,189,183,181,183,183,181],2,14.8,[181,181,191,192,183,181,183,183,181],3,29.7,[181,181,116,194,183,181,183,183,181],12.2,[181,181,196,197,183,181,183,183,181],5,26.1,[181,181,199,200,183,181,183,183,181],6,9.4,[181,181,202,203,183,181,183,183,181],7,7.8,[181,181,205,206,183,181,183,183,181],8,10.8,[185,181,181,208,183,181,183,183,181],11,[185,181,185,210,183,181,183,183,181],18.3,[185,181,188,212,183,181,183,183,181],13.3,[185,181,191,214,183,181,183,183,181],12.6,[185,181,116,216,183,181,183,183,181],17.4,[185,181,196,218,183,181,183,183,181],10.5,[185,181,199,220,183,181,183,183,181],8.9,[185,181,202,222,183,181,183,183,181],16.9,[185,181,205,224,183,181,183,183,181],10.3,[188,181,181,226,183,181,183,183,181],9.9,[188,181,185,218,183,181,183,183,181],[188,181,188,229,183,181,183,183,181],7.4,[188,181,191,218,183,181,183,183,181],[188,181,116,232,183,181,183,183,181],8.3,[188,181,196,234,183,181,183,183,181],7.9,[188,181,199,234,183,181,183,183,181],[188,181,202,237,183,181,183,183,181],5.1,[188,181,205,239,183,181,183,183,181],9.7,[191,181,181,214,183,181,183,183,181],[191,181,185,242,183,181,183,183,181],22.7,[191,181,188,244,183,181,183,183,181],18.2,[191,181,191,246,183,181,183,183,181],24.4,[191,181,116,248,183,181,183,183,181],16.1,[191,181,196,250,183,181,183,183,181],20.2,[191,181,199,252,183,181,183,183,181],11.9,[191,181,202,254,183,181,183,183,181],12.4,[191,181,205,208,183,181,183,183,181],[116,181,181,257,183,181,183,183,181],15.2,[116,181,185,259,183,181,183,183,181],21.4,[116,181,188,261,183,181,183,183,181],15.7,[116,181,191,263,183,181,183,183,181],14.4,[116,181,116,189,183,181,183,183,181],[116,181,196,266,183,181,183,183,181],13.6,[116,181,199,268,183,181,183,183,181],9.8,[116,181,202,270,183,181,183,183,181],11.8,[116,181,205,272,183,181,183,183,181],19.5,[196,181,181,274,183,181,183,183,181],9.5,[196,181,185,208,183,181,183,183,181],[196,181,188,277,183,181,183,183,181],7.1,[196,181,191,279,183,181,183,183,181],8.2,[196,181,116,252,183,181,183,183,181],[196,181,196,206,183,181,183,183,181],[196,181,199,283,183,181,183,183,181],6.9,[196,181,202,285,183,181,183,183,181],10.6,[196,181,205,287,183,181,183,183,181],11.7,[199,181,181,268,183,181,183,183,181],[199,181,185,290,183,181,183,183,181],8.7,[199,181,188,283,183,181,183,183,181],[199,181,191,293,183,181,183,183,181],9.1,[199,181,116,220,183,181,183,183,181],[199,181,196,220,183,181,183,183,181],[199,181,199,200,183,181,183,183,181],[199,181,202,298,183,181,183,183,181],7.6,[199,181,205,300,183,181,183,183,181],9.6,[202,181,181,293,183,181,183,183,181],[202,181,185,298,183,181,183,183,181],[202,181,188,304,183,181,183,183,181],4.8,[202,181,191,279,183,181,183,183,181],[202,181,116,300,183,181,183,183,181],[202,181,196,308,183,181,183,183,181],7.7,[202,181,199,220,183,181,183,183,181],[202,181,202,311,183,181,183,183,181],5.4,[202,181,205,300,183,181,183,183,181],[205,181,181,314,183,181,183,183,181],13.2,[205,181,185,250,183,181,183,183,181],[205,181,188,317,183,181,183,183,181],16.7,[205,181,191,182,183,181,183,183,181],[205,181,116,257,183,181,183,183,181],[205,181,196,321,183,181,183,183,181],20.4,[205,181,199,224,183,181,183,183,181],[205,181,202,252,183,181,183,183,181],[205,181,205,325,183,181,183,183,181],11.2,[122,181,181,220,183,181,183,183,181],[122,181,185,328,183,181,183,183,181],8.6,[122,181,188,330,183,181,183,183,181],6.6,[122,181,191,122,183,181,183,183,181],[122,181,116,333,183,181,183,183,181],7.3,[122,181,196,308,183,181,183,183,181],[122,181,199,220,183,181,183,183,181],[122,181,202,337,183,181,183,183,181],4.5,[122,181,205,339,183,181,183,183,181],8.4,[],[121],[],[],[345],"ATE RMSE on the authors' nine underground RGB-D field sequences (Kinect2 on a mobile robot); trajectory reference from a multi-sensor (LiDAR, IMU, camera) fusion SLAM pipeline, not an independent survey; all methods rerun with official code on the same PC (Sec. IV-A); values transcribed from the table image",{"slug":347,"group":348,"sourceId":5,"sourceLabel":6,"table":349,"selfRows":191,"metrics":350,"seqs":360,"entrants":364,"cells":366,"outcomes":373,"locators":374,"hardware":375,"wordings":377,"notes":378},"yan2026-underground3dgsslam-table-v","yan2026_underground3dgsslam:Table V","Table V",[351,356,358],{"label":352,"unit":353,"statistic":354,"alignment":355},"Average duration, Preprocessing (image enhancement and depth completion)","ms","mean","none",{"label":357,"unit":353,"statistic":354,"alignment":355},"Average duration, Tracking (keyframe selection and pose optimization)",{"label":359,"unit":353,"statistic":354,"alignment":355},"Average duration, Mapping (Gaussian optimization and management)",[361],{"dataset":83,"sequence":362,"environment":363},"average over sequences","underground spaces",[365],{"name":172,"methodId":5,"linkable":154,"proposed":154,"self":154},[367,369,371],[181,181,181,368,183,181,181,183,181],72.5,[181,185,181,370,183,181,181,183,181],36.2,[181,188,181,372,183,181,181,183,181],41.8,[],[349],[376],"Intel i7-14700K, NVIDIA GeForce RTX 4090D, 64 GB DDR5, Ubuntu 18.04 (Python, PyTorch)",[],[379],"Average time per frame for each stage, averaged over multiple underground sequences",{"slug":381,"group":382,"sourceId":5,"sourceLabel":6,"table":383,"selfRows":191,"metrics":384,"seqs":386,"entrants":395,"cells":407,"outcomes":457,"locators":458,"hardware":459,"wordings":460,"notes":461},"yan2026-underground3dgsslam-table-vi","yan2026_underground3dgsslam:Table VI","Table VI",[385],{"label":125,"unit":126,"statistic":127,"alignment":105},[387,391,393],{"dataset":388,"sequence":389,"environment":390},"TUM RGB-D","fr1\u002Fdesk","diverse indoor scenes (TUM RGB-D)",{"dataset":388,"sequence":392,"environment":390},"fr2\u002Fxyz",{"dataset":388,"sequence":394,"environment":390},"fr3\u002Foffice",[396,397,398,399,400,401,403,404,405,406],{"name":152,"methodId":153,"linkable":154,"proposed":73,"self":73},{"name":156,"methodId":157,"linkable":154,"proposed":73,"self":73},{"name":159,"methodId":160,"linkable":154,"proposed":73,"self":73},{"name":162,"methodId":163,"linkable":154,"proposed":73,"self":73},{"name":165,"methodId":166,"linkable":154,"proposed":73,"self":73},{"name":402,"methodId":51,"linkable":73,"proposed":73,"self":73},"GS-ICP-SLAM",{"name":170,"methodId":51,"linkable":73,"proposed":73,"self":73},{"name":172,"methodId":5,"linkable":154,"proposed":154,"self":154},{"name":174,"methodId":175,"linkable":154,"proposed":73,"self":73},{"name":177,"methodId":178,"linkable":154,"proposed":73,"self":73},[408,410,411,413,415,417,418,420,422,424,426,428,430,431,432,434,435,436,437,438,440,442,443,444,446,448,450,452,454,456],[181,181,181,409,183,181,183,183,181],3.2,[181,181,185,188,183,181,183,183,181],[181,181,188,412,183,181,183,183,181],4.7,[185,181,181,414,183,181,183,183,181],2.7,[185,181,185,416,183,181,183,183,181],1.8,[185,181,188,191,183,181,183,183,181],[188,181,181,419,183,181,183,183,181],2.8,[188,181,185,421,183,181,183,183,181],1.2,[188,181,188,423,183,181,183,183,181],2.5,[191,181,181,425,183,181,183,183,181],3.3,[191,181,185,427,183,181,183,183,181],1.4,[191,181,188,429,183,181,183,183,181],5.2,[116,181,181,409,183,181,183,183,181],[116,181,185,427,183,181,183,183,181],[116,181,188,433,183,181,183,183,181],2.1,[196,181,181,414,183,181,183,183,181],[196,181,185,416,183,181,183,183,181],[196,181,188,414,183,181,183,183,181],[199,181,181,433,183,181,183,183,181],[199,181,185,439,183,181,183,183,181],0.9,[199,181,188,441,183,181,183,183,181],1.9,[202,181,181,433,183,181,183,183,181],[202,181,185,427,183,181,183,183,181],[202,181,188,445,183,181,183,183,181],1.7,[205,181,181,447,183,181,183,183,181],2.53,[205,181,185,449,183,181,183,183,181],1.17,[205,181,188,451,183,181,183,183,181],2.52,[122,181,181,453,183,181,183,183,181],1.6,[122,181,185,455,183,181,183,183,181],0.4,[122,181,188,185,183,181,183,183,181],[],[383],[],[],[462],"ATE RMSE on three TUM RGB-D sequences; ElasticFusion and ORB-SLAM2 values are identical to those printed in SplaTAM Table 1 (from Point-SLAM), suggesting reuse of published values although Sec. IV-A states all comparisons were reproduced (inference)",{"slug":464,"group":465,"sourceId":5,"sourceLabel":6,"table":466,"selfRows":185,"metrics":467,"seqs":471,"entrants":473,"cells":475,"outcomes":477,"locators":478,"hardware":480,"wordings":481,"notes":482},"yan2026-underground3dgsslam-text-sec-iv-c","yan2026_underground3dgsslam:Text Sec. IV-C","Text Sec. IV-C",[468],{"label":469,"unit":470,"statistic":354,"alignment":355},"average frame rate","FPS",[472],{"dataset":83,"sequence":362,"environment":363},[474],{"name":172,"methodId":5,"linkable":154,"proposed":154,"self":154},[476],[181,181,181,328,183,181,181,183,181],[],[479],"Sec. IV-C",[376],[],[483],"Average frame rate over multiple underground sequences",[],1790510662041]