[{"data":1,"prerenderedAt":358},["ShallowReactive",2],{"method-ltmapper2022":3},{"method":4,"reference":57,"equipment":76,"figures":89,"results":130},{"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":22,"limitations":26,"sensors":32,"platform":35,"estimator":37,"association":38,"timeModel":39,"deskew":40,"loopClosure":41,"globalOptimization":42,"mapRepresentation":43,"prior":44,"outputGeometry":45,"compute":46,"codeUrl":47,"codeLicense":48,"relatedVersions":49},"ltmapper2022","Kim & Kim, 2022","LT-mapper","LT-mapper: A Modular Framework for LiDAR-based Lifelong Mapping",2022,"recent","C06","full_slam_with_global_correction","LT-mapper 把長期建圖拆成三個模組：LT-SLAM 以錨節點（anchor node）多時段位姿圖與 Scan Context 跨時段迴圈，對齊原點不同且各自漂移的時段；LT-removert 先移除高動態點，再以集合差分偵測低動態變化，分為新出現（PD）與消失（ND）的點；LT-map 維護最新狀態的即時地圖與持久結構的 meta map，並可串接差異地圖重建任一時間點的地圖。只需單一光達（IMU 可選）。","LT-mapper chains anchor-node multi-session SLAM with Scan Context loops, dynamic removal plus positive\u002Fnegative change detection, and a map manager that can reconstruct the map at any session time.","full_text_reviewed","peer_reviewed_published","main_body","未於工地測試（MulRan 都市與校園、停車場六時段）。作者的長期變化示例中，MulRan KAIST 01 與 04（約 1.5 年間隔）出現新設施工圍牆、DCC 01\u002F02 有施工圍牆遮蔽後移除（Fig. 1、Fig. 6），屬都市場景中的施工相關變化，而非工地內部監測。其跨時段對齊與正、負變化偵測流程，與工地重複掃描做進度與差異比對的需求結構相同（推論）；BIM-SLAM 直接沿用 LT-SLAM（bimslam2023 Sec. 4.3）。",[20,21],"public_benchmark","cross_site",[23,24,25],"No good initial alignment between sessions required (abstract; Sec. IV)","Inter-session loops reduced each session's internal drift, evaluated with the RPG trajectory tool (Sec. V-B, Fig. 8)","Tested with temporal gaps from days to a year (abstract)",[27,28,29,30,31],"Found in follow-up work: LT-SLAM failed on a repetitive indoor parking dataset because Scan Context could not recognize loops (yang2024lifelong Sec. IV-B)","(inference) Change detection inherits Removert's visibility assumptions and any residual inter-session misalignment","No point-wise ground truth for changes; change detection is assessed only implicitly by composing changes and comparing restored and real maps with patch-wise Chamfer distance (Sec. V-D, Table I)","The authors show that even RTK-GPS ground-truth maps of MulRan DCC 01 and 02 are not globally consistent across sessions (Fig. 2a)","Inter-session loop detection relies on Scan Context plus ICP fitness; false loops are left to a robust back end (Sec. IV-A)",[33,34],"3D LiDAR","IMU (optional, for initial odometry)",[36],"not described in the paper: MulRan KAIST and DCC sequences (a vehicle dataset according to its own paper) and the self-collected six-session LT-ParkingLot dataset (carrier not stated)","multi-session pose-graph optimization with anchor nodes (iSAM2 in GTSAM)","Scan Context inter-session loop candidates verified by ICP between keyframe submaps (loops accepted only with low ICP fitness score, which also sets an adaptive covariance) with a robust back end, followed by radius-search loops by pose proximity; high-dynamic points removed with Removert; low-dynamic changes found by a kd-tree test of whether a point has k target points within r m; weak negative differences reverted by a modified Removert (Sec. IV-A, IV-B)","discrete poses (keyframes)","not_reported (delegated to input odometry, e.g., LIO-SAM)","Scan Context intra- and inter-session loops","joint multi-session pose graph with per-session anchor nodes","keyframe point clouds; live map and meta map; delta maps of positive and negative changes","previous sessions (central map); no initial inter-session alignment required","aligned multi-session point clouds, static maps, positive\u002Fnegative change point sets, maps at any timestamp via change composition","Offline modular C++ commands (ltslam, ltremovert, ltmap); hardware not reported. Change composition takes about 0.05 s per keyframe; for KAIST 04 versus 01, delta-map chaining took 9.8 s versus 87.0 s (without HD removal) or 160.0 s (with HD removal) when computing changes from scratch, and merged-map memory was 85.7 MB versus 213.6 MB for whole snapshots (Sec. V-D, Table II).","https:\u002F\u002Fgithub.com\u002Fgisbi-kim\u002Flt-mapper","MIT (LICENSE file, copyright year 2025)",[50,54],{"relation":51,"title":52,"doi_or_url":53},"preprint","arXiv 2107.07712","https:\u002F\u002Farxiv.org\u002Fabs\u002F2107.07712",{"relation":55,"title":56,"doi_or_url":47},"code_release","gisbi-kim\u002Flt-mapper",{"id":5,"kind":58,"shortName":7,"title":8,"authors":59,"year":9,"venue":62,"venueType":63,"publisher":64,"volumeIssuePages":65,"doi":66,"arxivId":67,"url":53,"firstPublicDate":68,"publicationStatus":16,"metadataStatus":69,"fulltextStatus":15,"era":10,"classicReason":70,"codeUrl":47,"cluster":11,"topics":71,"mdpi":72,"verification":73,"label":6,"fulltextRoute":74,"versionRead":75,"addedByCensus":72},"method",[60,61],"Giseop Kim","Ayoung Kim","2022 International Conference on Robotics and Automation (ICRA)","conference","IEEE","pp. 7995-8002","10.1109\u002Ficra46639.2022.9811916","2107.07712","2021-07-16","metadata_verified","not_applicable",[11],false,"corrected","arXiv","arXiv 2107.07712v1 (2021-07-16, CC BY 4.0); ICRA 2022 version of record not compared",[77,84],{"category":78,"model":79,"canonical":79,"role":80,"dataset":81,"specs":82,"locator":83},"lidar","single 3D LiDAR (model not named)","method input","MulRan KAIST and DCC; LT-ParkingLot (six sessions over three days)","not_reported","Sec. III; Sec. V-A2",{"category":85,"model":86,"canonical":86,"role":80,"dataset":87,"specs":82,"locator":88},"imu","IMU (optional, for initial odometry; model not named)",null,"Sec. III",[90,103,112,122],{"refId":5,"refLabel":6,"fig":91,"whatZh":92,"license":93,"licenseUrl":94,"sourceUrl":95,"src":96,"width":97,"height":98,"thumb":99,"thumbWidth":100,"thumbHeight":101,"modified":102},"Fig. 1","MulRan KAIST 01（2019 年 6 月）與 KAIST 04（2021 年 2 月）約 1.5 年間的永久結構變化，出現施工圍牆，停車格與樹木消失，藍點為偵測到的變化點","CC BY 4.0 (arXiv 2107.07712v1 licence); ICRA 2022 version of record © IEEE","http:\u002F\u002Fcreativecommons.org\u002Flicenses\u002Fby\u002F4.0\u002F","https:\u002F\u002Farxiv.org\u002Fhtml\u002F2107.07712v1\u002Fcover_small.png","\u002Ffigure-files\u002Fltmapper2022\u002Ffig-1.webp",889,463,"\u002Ffigure-files\u002Fltmapper2022\u002Ffig-1.thumb.webp",480,250,"converted to WebP",{"refId":5,"refLabel":6,"fig":104,"whatZh":105,"license":93,"licenseUrl":94,"sourceUrl":106,"src":107,"width":108,"height":109,"thumb":110,"thumbWidth":100,"thumbHeight":111,"modified":102},"Fig. 5","LT-removert 流程：接收對齊的中央與查詢地圖，移除高動態點，偵測新增與消失的低動態變化，並跨時段刪除殘留動態點","https:\u002F\u002Farxiv.org\u002Fhtml\u002F2107.07712v1\u002Fltremovert_visual_pipeline_small.png","\u002Ffigure-files\u002Fltmapper2022\u002Ffig-5.webp",886,346,"\u002Ffigure-files\u002Fltmapper2022\u002Ffig-5.thumb.webp",187,{"refId":5,"refLabel":6,"fig":113,"whatZh":114,"license":93,"licenseUrl":94,"sourceUrl":115,"src":116,"width":117,"height":118,"thumb":119,"thumbWidth":100,"thumbHeight":120,"modified":121},"Fig. 9","LT-mapper 的兩種地圖管理：保持最新狀態的即時地圖，與逐次移除非體積最大化點、保留持久結構的 meta map","https:\u002F\u002Farxiv.org\u002Fhtml\u002F2107.07712v1\u002Fmetamap.png","\u002Ffigure-files\u002Fltmapper2022\u002Ffig-9.webp",1400,906,"\u002Ffigure-files\u002Fltmapper2022\u002Ffig-9.thumb.webp",311,"resized to at most 1400 px wide and converted to WebP",{"refId":5,"refLabel":6,"fig":123,"whatZh":124,"license":93,"licenseUrl":94,"sourceUrl":125,"src":126,"width":117,"height":127,"thumb":128,"thumbWidth":100,"thumbHeight":129,"modified":121},"Fig. 10","以差異地圖串接進行變化合成：由 KAIST 04 回溯還原 KAIST 01，未對應的牆面消失、原有牆面復原","https:\u002F\u002Farxiv.org\u002Fhtml\u002F2107.07712v1\u002Fcompose.png","\u002Ffigure-files\u002Fltmapper2022\u002Ffig-10.webp",268,"\u002Ffigure-files\u002Fltmapper2022\u002Ffig-10.thumb.webp",92,{"totalRows":131,"groupCount":132,"groups":133,"others":357},15,4,[134,210,292,333],{"slug":135,"group":136,"sourceId":5,"sourceLabel":6,"table":137,"selfRows":138,"metrics":139,"seqs":158,"entrants":163,"cells":169,"outcomes":204,"locators":205,"hardware":206,"wordings":207,"notes":208},"ltmapper2022-table-i","ltmapper2022:Table I","Table I",7,[140,144,147,149,152,154,156],{"label":141,"unit":82,"statistic":142,"alignment":143},"Chamfer Distance Max","max","none",{"label":145,"unit":82,"statistic":146,"alignment":143},"Chamfer Distance Avg","mean",{"label":148,"unit":82,"statistic":82,"alignment":143},"Chamfer Distance Var",{"label":150,"unit":151,"statistic":82,"alignment":143},"NP with CD > tau, tau = 1","patches",{"label":153,"unit":151,"statistic":82,"alignment":143},"NP with CD > tau, tau = 2",{"label":155,"unit":151,"statistic":82,"alignment":143},"NP with CD > tau, tau = 3",{"label":157,"unit":151,"statistic":82,"alignment":143},"NPvalid",[159],{"dataset":160,"sequence":161,"environment":162},"MulRan","KAIST 01 and 04 (about 1.5-year gap)","urban campus with construction wall change",[164,167],{"name":165,"methodId":5,"linkable":166,"proposed":166,"self":166},"Pos. Pair (01 vs Restored 01)",true,{"name":168,"methodId":87,"linkable":72,"proposed":72,"self":72},"Neg. Pair (01 vs 04)",[170,174,177,180,183,185,187,190,192,194,196,198,200,202],[171,171,171,172,173,171,173,173,171],0,6.41,-1,[171,175,171,176,173,171,173,173,171],1,0.29,[171,178,171,179,173,171,173,173,171],2,0.31,[171,181,171,182,173,171,173,173,171],3,38,[171,132,171,184,173,171,173,173,171],9,[171,186,171,175,173,171,173,173,171],5,[171,188,171,189,173,171,173,173,171],6,1424,[175,171,171,191,173,171,173,173,171],29.22,[175,175,171,193,173,171,173,173,171],0.51,[175,178,171,195,173,171,173,173,171],1.67,[175,181,171,197,173,171,173,173,171],100,[175,132,171,199,173,171,173,173,171],20,[175,186,171,201,173,171,173,173,171],8,[175,188,171,203,173,171,173,173,171],1386,[],[137],[],[],[209],"Change composition check on MulRan KAIST: KAIST 01 restored from KAIST 04 by chaining delta maps (01 to 02 and 02 to 04) compared with the real KAIST 01 (positive pair) and KAIST 04 compared with 01 (negative pair); aligned maps split into 5 m cubic patches with at least 25 points; for 100 KF01s and near 200 KF04s",{"slug":211,"group":212,"sourceId":213,"sourceLabel":214,"table":215,"selfRows":186,"metrics":216,"seqs":220,"entrants":238,"cells":247,"outcomes":285,"locators":287,"hardware":288,"wordings":289,"notes":290},"yang2024lifelong-table-iii","yang2024lifelong:Table III","yang2024lifelong","Yang et al., 2024","Table III",[217],{"label":218,"unit":219,"statistic":146,"alignment":143},"Chamfer distance after multi-session map alignment","not_reported (mean of squared nearest-neighbour distances)",[221,225,229,233,235],{"dataset":222,"sequence":223,"environment":224},"XGrid-Outdoor","6 session maps","outdoor, commercial hand-held LiDAR",{"dataset":226,"sequence":227,"environment":228},"XGrid-Parking","5 session maps","indoor car park with repetitive structure, commercial hand-held LiDAR",{"dataset":230,"sequence":231,"environment":232},"LT-ParkingLot","6 session maps, SC-LIO-SAM poses (row LT-ParkingLot-SC-LIO-SAM)","parking lot",{"dataset":230,"sequence":234,"environment":232},"6 session maps, SC-A-LOAM poses (row LT-ParkingLot-SC-A-LOAM)",{"dataset":160,"sequence":236,"environment":237},"DCC, 3 session maps (row MulRan-SC-LIO-LOAM)","urban",[239,241,243,245],{"name":240,"methodId":213,"linkable":166,"proposed":166,"self":72},"Ours",{"name":242,"methodId":5,"linkable":166,"proposed":72,"self":166},"LT-SLAM",{"name":244,"methodId":87,"linkable":72,"proposed":72,"self":72},"ICP",{"name":246,"methodId":87,"linkable":72,"proposed":72,"self":72},"NDT",[248,250,252,254,256,258,259,260,262,264,266,268,270,272,274,276,278,280,282,283],[171,171,171,249,173,171,173,173,171],0.069,[175,171,171,251,173,171,173,173,171],0.0862,[178,171,171,253,173,171,173,173,171],0.1389,[181,171,171,255,173,171,173,173,171],0.1342,[171,171,175,257,173,171,173,173,171],0.0812,[175,171,175,87,171,171,173,173,171],[178,171,175,87,171,171,173,173,171],[181,171,175,261,173,171,173,173,171],0.1235,[171,171,178,263,173,171,173,173,171],0.0329,[175,171,178,265,173,171,173,173,171],0.0592,[178,171,178,267,173,171,173,173,171],0.1211,[181,171,178,269,173,171,173,173,171],0.1378,[171,171,181,271,173,171,173,173,171],0.0173,[175,171,181,273,173,171,173,173,171],0.0526,[178,171,181,275,173,171,173,173,171],0.1228,[181,171,181,277,173,171,173,173,171],0.1282,[171,171,132,279,173,171,173,173,171],0.1004,[175,171,132,281,173,171,173,173,171],0.2059,[178,171,132,87,171,171,173,173,171],[181,171,132,284,173,171,173,173,171],0.1231,[286],"failed",[215],[],[],[291],"Average Chamfer distance after aligning several session maps into one frame; poses from XGrids proprietary software (XGrid datasets) or SC-LIO-SAM and SC-A-LOAM (LT-ParkingLot, MulRan); outlier threshold tau 0.5; the proposed method also uses Chamfer distance to pick its alignment",{"slug":293,"group":294,"sourceId":5,"sourceLabel":6,"table":295,"selfRows":178,"metrics":296,"seqs":303,"entrants":307,"cells":316,"outcomes":327,"locators":328,"hardware":329,"wordings":330,"notes":331},"ltmapper2022-table-ii","ltmapper2022:Table II","Table II",[297,300],{"label":298,"unit":299,"statistic":82,"alignment":143},"Memory Usage (merged map)","MB",{"label":301,"unit":302,"statistic":82,"alignment":143},"Computation Time (between 04 and 01)","s",[304],{"dataset":160,"sequence":305,"environment":306},"KAIST 01 and 04","urban campus",[308,310,312,314],{"name":309,"methodId":87,"linkable":72,"proposed":72,"self":72},"Baseline (saving whole snapshot)",{"name":311,"methodId":5,"linkable":166,"proposed":166,"self":166},"Ours (LT-map, delta map chaining)",{"name":313,"methodId":87,"linkable":72,"proposed":72,"self":72},"Baseline, w\u002Fo HD removal",{"name":315,"methodId":87,"linkable":72,"proposed":72,"self":72},"Baseline, w\u002F HD removal",[317,319,321,323,325],[171,171,171,318,173,171,173,173,171],213.6,[175,171,171,320,173,171,173,173,171],85.7,[178,175,171,322,173,171,173,173,171],87,[181,175,171,324,173,171,173,173,171],160,[175,175,171,326,173,171,173,173,171],9.8,[],[295],[],[],[332],"Efficiency of LT-map delta-map chaining versus saving whole snapshots for the Fig. 10 scene (KAIST 04 vs 01); hardware not reported",{"slug":334,"group":335,"sourceId":213,"sourceLabel":214,"table":336,"selfRows":175,"metrics":337,"seqs":341,"entrants":344,"cells":346,"outcomes":349,"locators":351,"hardware":353,"wordings":354,"notes":355},"yang2024lifelong-text-sec-iv-b","yang2024lifelong:Text Sec. IV-B","Text Sec. IV-B",[338],{"label":339,"unit":340,"statistic":82,"alignment":143},"total alignment time (over 10 min)","min",[342],{"dataset":222,"sequence":343,"environment":224},"about 10M points",[345],{"name":242,"methodId":5,"linkable":166,"proposed":72,"self":166},[347],[171,171,171,348,171,171,173,173,171],10,[350],"lower bound: reported as over 10 min",[352],"Sec. IV-B",[],[],[356],"Time to align XGrid-Outdoor session maps of about 10M points including the full grid search, compared with LT-SLAM keyframe descriptor traversal and graph optimization",[],1790510665235]