[{"data":1,"prerenderedAt":404},["ShallowReactive",2],{"method-ptam2007":3},{"method":4,"reference":47,"equipment":67,"figures":80,"results":81},{"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":24,"sensors":27,"platform":29,"estimator":31,"association":32,"timeModel":33,"deskew":34,"loopClosure":35,"globalOptimization":36,"mapRepresentation":37,"prior":38,"outputGeometry":39,"compute":40,"codeUrl":41,"codeLicense":42,"relatedVersions":43},"ptam2007","Klein & Murray, 2007","PTAM","Parallel Tracking and Mapping for Small AR Workspaces",2007,"classic","C08","odometry_with_local_mapping","PTAM 將相機追蹤（tracking）與建圖（mapping）拆成兩個平行執行緒：追蹤執行緒以地圖點重投影估計每張影像的位姿，建圖執行緒則對關鍵影格（keyframe）執行計算量較大的光束法平差（bundle adjustment, BA）。此設計讓即時系統可以使用原本多用於離線 SfM 的批次最佳化。作者將其定位為小型擴增實境工作區，並未支援大範圍探索。","PTAM separates frame-rate tracking from keyframe-based bundle-adjusted mapping in parallel threads, enabling batch optimisation in a real-time monocular system for small AR workspaces.","full_text_reviewed","peer_reviewed_published","background","論文未報告營建或建築量測測試；設計目標為桌面或房間角落等小型 AR 工作區。對本文僅作為關鍵影格 BA 架構的技術源頭。",[20,21],"controlled_experiment","simulation",[23],"add: on a synthetic 600-frame sequence (camera path 18.2 m) the trajectory standard deviation from ground truth after 7-DOF alignment was 6 mm versus 135 mm for an EKF-SLAM implementation, with near-constant 20 ms tracking (Sec. 7.3)",[25,26],"add: the map is only a point cloud","after extracting the dominant plane the system makes little effort to extract further geometric understanding, and patch normals are initialised parallel to the image plane and not optimised (Sec. 6.5, 8.2)",[28],"monocular camera",[30],"handheld","tracking thread: pose update by ten iterations of reweighted least squares on a Tukey-biweight reprojection objective; mapping thread: Levenberg-Marquardt bundle adjustment with a Tukey M-estimator, run globally or locally over the newest keyframe and its four nearest keyframes (Sec. 5.4, 6.3)","FAST-10 corners on a four-level image pyramid; affine-warped 8x8 patches searched by zero-mean SSD at FAST corners around reprojected map points, 50 coarse-level points then up to 1000 points per frame; new map points triangulated by epipolar search against the nearest keyframe (Sec. 5.1, 5.3, 5.5, 6.2)","discrete poses (keyframes)","not_applicable","no dedicated loop-detection module; authors state the system is not designed to close large loops, although the BA mapping can absorb loops when the camera is placed near the map boundary (Sec. 8)","bundle adjustment over keyframes and map points in the mapping thread","sparse point-feature map with keyframes","no prior map; user-assisted five-point stereo initialisation in which metric scale is set by assuming a 10 cm camera translation between the first two keyframes and the dominant plane found by RANSAC is placed at z = 0 (Sec. 6.1)","keyframe poses and sparse 3D point features","Intel Core 2 Duo 2.66 GHz desktop PC running Linux, C++ with libCVD and TooN (Sec. 6.5); about 19.2 ms to track one frame with a 4000-point map (Table 1); mean global bundle adjustment 380 ms, 1.7 s and 6.9 s for maps of 2-49, 50-99 and 100-149 keyframes versus 170, 270 and 440 ms for local bundle adjustment (Table 2)","https:\u002F\u002Fgithub.com\u002FOxford-PTAM\u002FPTAM-GPL","GPLv3 (LICENSE file of Oxford-PTAM\u002FPTAM-GPL; the PTAM project page links this repository)",[44],{"relation":45,"title":46,"doi_or_url":41},"code_release","PTAM-GPL",{"id":5,"kind":48,"shortName":7,"title":8,"authors":49,"year":9,"venue":52,"venueType":53,"publisher":54,"volumeIssuePages":55,"doi":56,"arxivId":57,"url":58,"firstPublicDate":59,"publicationStatus":16,"metadataStatus":60,"fulltextStatus":15,"era":10,"classicReason":61,"codeUrl":41,"cluster":11,"topics":62,"mdpi":63,"verification":64,"label":6,"fulltextRoute":65,"versionRead":66,"addedByCensus":63},"method",[50,51],"Georg Klein","David Murray","2007 6th IEEE and ACM International Symposium on Mixed and Augmented Reality (ISMAR)","conference","IEEE","pp. 1-10 as registered in Crossref (citing works often give pp. 225-234; not verified on IEEE Xplore)","10.1109\u002Fismar.2007.4538852",null,"https:\u002F\u002Fwww.robots.ox.ac.uk\u002F~gk\u002Fpublications\u002FKleinMurray2007ISMAR.pdf","2007-11","metadata_verified","principle reused: splitting tracking and keyframe-based bundle-adjusted mapping into parallel threads, which ORB-SLAM explicitly states it extends.",[11],false,"corrected","author copy","author-hosted copy of the ISMAR 2007 paper (10 pages) read in full; version of record HTML on IEEE Xplore (NTU access) checked: same nine sections and matching key values (Core 2 Duo 2.66 GHz, 6 mm versus 135 mm, 11000 points and 280 keyframes, 10 cm initial baseline); VoR table images not compared; IEEE page shows no page range",[68,74],{"category":69,"model":70,"canonical":70,"role":71,"dataset":57,"specs":72,"locator":73},"camera","Unibrain Fire-i","method input","video camera with 2.1 mm wide-angle lens; 640x480 YUV411 frames at 30 Hz, converted to 8-bit greyscale","Sec. 5.1",{"category":75,"model":76,"canonical":76,"role":77,"dataset":57,"specs":78,"locator":79},"compute","Intel Core 2 Duo 2.66 GHz desktop PC","compute for runtime","dual-core processor, Linux, C++ with libCVD and TooN","Sec. 6.5",[],{"totalRows":82,"groupCount":83,"groups":84,"others":380},39,8,[85,249,293,333],{"slug":86,"group":87,"sourceId":88,"sourceLabel":89,"table":90,"selfRows":91,"metrics":92,"seqs":98,"entrants":133,"cells":141,"outcomes":240,"locators":244,"hardware":245,"wordings":246,"notes":247},"orbslam2015-table-iii","orbslam2015:Table III","orbslam2015","Mur-Artal et al., 2015","Table III",16,[93],{"label":94,"unit":95,"statistic":96,"alignment":97},"Absolute KeyFrame Trajectory RMSE","cm","RMSE","Sim3",[99,103,105,107,109,111,113,115,117,119,121,123,125,127,129,131],{"dataset":100,"sequence":101,"environment":102},"TUM RGB-D","fr1_xyz","indoor, hand-held",{"dataset":100,"sequence":104,"environment":102},"fr2_xyz",{"dataset":100,"sequence":106,"environment":102},"fr1_floor",{"dataset":100,"sequence":108,"environment":102},"fr1_desk",{"dataset":100,"sequence":110,"environment":102},"fr2_360_kidnap",{"dataset":100,"sequence":112,"environment":102},"fr2_desk",{"dataset":100,"sequence":114,"environment":102},"fr3_long_office",{"dataset":100,"sequence":116,"environment":102},"fr3_nstr_tex_far",{"dataset":100,"sequence":118,"environment":102},"fr3_nstr_tex_near",{"dataset":100,"sequence":120,"environment":102},"fr3_str_tex_far",{"dataset":100,"sequence":122,"environment":102},"fr3_str_tex_near",{"dataset":100,"sequence":124,"environment":102},"fr2_desk_person",{"dataset":100,"sequence":126,"environment":102},"fr3_sit_xyz",{"dataset":100,"sequence":128,"environment":102},"fr3_sit_halfsph",{"dataset":100,"sequence":130,"environment":102},"fr3_walk_xyz",{"dataset":100,"sequence":132,"environment":102},"fr3_walk_halfsph",[134,137,138],{"name":135,"methodId":88,"linkable":136,"proposed":136,"self":63},"ORB-SLAM",true,{"name":7,"methodId":5,"linkable":136,"proposed":63,"self":136},{"name":139,"methodId":140,"linkable":136,"proposed":63,"self":63},"LSD-SLAM","lsdslam2014",[142,146,149,152,154,156,158,160,161,163,166,167,169,172,174,175,178,179,181,184,185,187,189,190,192,194,196,198,200,202,204,207,209,210,213,214,216,219,221,223,226,227,229,232,233,235,238,239],[143,143,143,144,145,143,145,145,143],0,0.9,-1,[147,143,143,148,145,143,145,145,143],1,1.15,[150,143,143,151,145,143,145,145,143],2,9,[143,143,147,153,145,143,145,145,143],0.3,[147,143,147,155,145,143,145,145,143],0.2,[150,143,147,157,145,143,145,145,143],2.15,[143,143,150,159,145,143,145,145,143],2.99,[147,143,150,57,143,143,145,145,143],[150,143,150,162,145,143,145,145,143],38.07,[143,143,164,165,145,143,145,145,143],3,1.69,[147,143,164,57,143,143,145,145,143],[150,143,164,168,145,143,145,145,143],10.65,[143,143,170,171,145,143,145,145,143],4,3.81,[147,143,170,173,145,143,145,145,143],2.63,[150,143,170,57,143,143,145,145,143],[143,143,176,177,145,143,145,145,143],5,0.88,[147,143,176,57,143,143,145,145,143],[150,143,176,180,145,143,145,145,143],4.57,[143,143,182,183,145,143,145,145,143],6,3.45,[147,143,182,57,143,143,145,145,143],[150,143,182,186,145,143,145,145,143],38.53,[143,143,188,57,147,143,145,145,143],7,[147,143,188,57,150,143,145,145,143],[150,143,188,191,145,143,145,145,143],18.31,[143,143,83,193,145,143,145,145,143],1.39,[147,143,83,195,145,143,145,145,143],2.74,[150,143,83,197,145,143,145,145,143],7.54,[143,143,151,199,145,143,145,145,143],0.77,[147,143,151,201,145,143,145,145,143],0.93,[150,143,151,203,145,143,145,145,143],7.95,[143,143,205,206,145,143,145,145,143],10,1.58,[147,143,205,208,145,143,145,145,143],1.04,[150,143,205,57,143,143,145,145,143],[143,143,211,212,145,143,145,145,143],11,0.63,[147,143,211,57,143,143,145,145,143],[150,143,211,215,145,143,145,145,143],31.73,[143,143,217,218,145,143,145,145,143],12,0.79,[147,143,217,220,145,143,145,145,143],0.83,[150,143,217,222,145,143,145,145,143],7.73,[143,143,224,225,145,143,145,145,143],13,1.34,[147,143,224,57,143,143,145,145,143],[150,143,224,228,145,143,145,145,143],5.87,[143,143,230,231,145,143,145,145,143],14,1.24,[147,143,230,57,143,143,145,145,143],[150,143,230,234,145,143,145,145,143],12.44,[143,143,236,237,145,143,145,145,143],15,1.74,[147,143,236,57,143,143,145,145,143],[150,143,236,57,143,143,145,145,143],[241,242,243],"failed (tracking lost, significant portion not processed)","not_run (initialisation refused: twofold planar ambiguity detected)","two values reported, 4.92 \u002F 34.74 cm, from runs with the true and the corrupted planar initialisation (Sec. VIII-B)",[90],[],[],[248],"TUM RGB-D keyframe ATE RMSE (cm), median over 5 executions; ORB-SLAM, PTAM (two manually chosen initial frames) and LSD-SLAM (first 10 keyframes discarded) aligned with Sim(3); RGBD-SLAM trajectories from the benchmark website aligned with SE(3), Sim(3) value in brackets; X = tracking lost",{"slug":250,"group":251,"sourceId":5,"sourceLabel":6,"table":252,"selfRows":182,"metrics":253,"seqs":261,"entrants":270,"cells":273,"outcomes":286,"locators":287,"hardware":288,"wordings":290,"notes":291},"ptam2007-table-2","ptam2007:Table 2","Table 2",[254,259],{"label":255,"unit":256,"statistic":257,"alignment":258},"local bundle adjustment time","ms","mean","none",{"label":260,"unit":256,"statistic":257,"alignment":258},"global bundle adjustment time (reported as 380 ms, 1.7 s, 6.9 s)",[262,266,268],{"dataset":263,"sequence":264,"environment":265},"typical timings from live operation (no specific sequence named; Sec. 7.2)","map with 2-49 keyframes","not specified; the authors state timings vary with map size and scene structure and are hard to reproduce from disk sequences (Sec. 7.2)",{"dataset":263,"sequence":267,"environment":265},"map with 50-99 keyframes",{"dataset":263,"sequence":269,"environment":265},"map with 100-149 keyframes",[271],{"name":272,"methodId":5,"linkable":136,"proposed":136,"self":136},"PTAM (proposed system)",[274,276,278,280,282,284],[143,143,143,275,145,143,143,145,143],170,[143,147,143,277,145,143,143,145,143],380,[143,143,147,279,145,143,143,145,143],270,[143,147,147,281,145,143,143,145,143],1700,[143,143,150,283,145,143,143,145,143],440,[143,147,150,285,145,143,143,145,143],6900,[],[252],[289],"Intel Core 2 Duo 2.66 GHz desktop PC, Linux (Sec. 6.5)",[],[292],"Mean bundle adjustment time by map size (keyframes); timings vary with map size and scene structure",{"slug":294,"group":295,"sourceId":5,"sourceLabel":6,"table":296,"selfRows":176,"metrics":297,"seqs":309,"entrants":314,"cells":316,"outcomes":327,"locators":328,"hardware":329,"wordings":330,"notes":331},"ptam2007-table-1","ptam2007:Table 1","Table 1",[298,301,303,305,307],{"label":299,"unit":256,"statistic":300,"alignment":258},"tracking timing: Keyframe preparation","not_reported",{"label":302,"unit":256,"statistic":300,"alignment":258},"tracking timing: Feature projection",{"label":304,"unit":256,"statistic":300,"alignment":258},"tracking timing: Patch search",{"label":306,"unit":256,"statistic":300,"alignment":258},"tracking timing: Iterative pose update",{"label":308,"unit":256,"statistic":300,"alignment":258},"tracking timing: Total",[310],{"dataset":311,"sequence":312,"environment":313},"own live video (Sec. 7.1)","typical frame, map of M=4000 points (Table 1 is not tied to a named sequence)","live hand-held operation; Sec. 7.1 describes a cluttered desk sequence, but Table 1 is only described as a typical frame",[315],{"name":272,"methodId":5,"linkable":136,"proposed":136,"self":136},[317,319,321,323,325],[143,143,143,318,145,143,143,145,143],2.2,[143,147,143,320,145,143,143,145,143],3.5,[143,150,143,322,145,143,143,145,143],9.8,[143,164,143,324,145,143,143,145,143],3.7,[143,170,143,326,145,143,143,145,143],19.2,[],[296],[289],[],[332],"Tracking time for a typical frame broken down by step, map size M=4000 (Sec. 7.1, Table 1)",{"slug":334,"group":335,"sourceId":140,"sourceLabel":336,"table":337,"selfRows":170,"metrics":338,"seqs":341,"entrants":353,"cells":358,"outcomes":372,"locators":375,"hardware":376,"wordings":377,"notes":378},"lsdslam2014-fig-9","lsdslam2014:Fig. 9","Engel et al., 2014","Fig. 9",[339],{"label":340,"unit":95,"statistic":96,"alignment":300},"absolute trajectory RMSE (cm)",[342,345,347,351],{"dataset":100,"sequence":343,"environment":344},"fr2\u002Fdesk","TUM RGB-D real sequences (fast rotational movement, strong motion blur and rolling-shutter artefacts, Sec. 4.2)",{"dataset":100,"sequence":346,"environment":344},"fr2\u002Fxyz",{"dataset":348,"sequence":349,"environment":350},"synthetic sequences (Handa et al. 2012)","sim\u002Fdesk","simulated sequence from Handa et al. [12]; scene and motion not described in this paper",{"dataset":348,"sequence":352,"environment":350},"sim\u002Fslowmo",[354,356],{"name":355,"methodId":57,"linkable":63,"proposed":63,"self":63},"semi-dense mono-VO [9]",{"name":357,"methodId":5,"linkable":136,"proposed":63,"self":136},"keypoint-based mono-SLAM [15] (PTAM)",[359,361,362,364,366,368,369,371],[143,143,143,360,145,143,145,145,143],13.5,[147,143,143,57,143,143,145,145,143],[143,143,147,363,145,143,145,145,143],3.79,[147,143,147,365,145,143,145,145,143],24.28,[143,143,150,367,145,143,145,145,143],1.53,[147,143,150,57,147,143,145,145,143],[143,143,164,370,145,143,145,145,143],2.21,[147,143,164,57,147,143,145,145,143],[373,374],"failed (tracking failure)","not_available (no data)",[337],[],[],[379],"Result table printed as Fig. 9: absolute trajectory RMSE (cm) on TUM RGB-D and two simulated sequences from Handa et al.; LSD-SLAM also lists keyframes created; x = tracking failure, '-' = no data; [14] and [7] use sensor depth; alignment not stated. The ORB-SLAM authors report being unable to reproduce the PTAM fr2\u002Fxyz value (orbslam2015, Sec. VIII-B)",[381,386,392,398],{"group":382,"slug":383,"sourceLabel":89,"table":384,"selfRows":170,"datasets":385},"orbslam2015:Table IV","orbslam2015-table-iv","Table IV",[100],{"group":387,"slug":388,"sourceLabel":6,"table":389,"selfRows":150,"datasets":390},"ptam2007:Text Sec. 7.3","ptam2007-text-sec-7-3","Text Sec. 7.3",[391],"synthetic sequence (own)",{"group":393,"slug":394,"sourceLabel":395,"table":252,"selfRows":147,"datasets":396},"ghadimzadeh2025slamnde:Table 2","ghadimzadeh2025slamnde-table-2","Ghadimzadeh Alamdari et al., 2025",[397],"Luleå SubT tunnel dataset (Koval et al. 2022)",{"group":399,"slug":400,"sourceLabel":6,"table":401,"selfRows":147,"datasets":402},"ptam2007:Text Sec. 7.2","ptam2007-text-sec-7-2","Text Sec. 7.2",[403],"own live video",1790510659485]