[{"data":1,"prerenderedAt":298},["ShallowReactive",2],{"method-cblox2018":3},{"method":4,"reference":64,"equipment":89,"figures":115,"results":116},{"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":27,"sensors":33,"platform":37,"estimator":40,"association":41,"timeModel":42,"deskew":43,"loopClosure":44,"globalOptimization":45,"mapRepresentation":46,"prior":47,"outputGeometry":48,"compute":49,"codeUrl":50,"codeLicense":51,"relatedVersions":52},"cblox2018","Millane et al., 2018","C-blox","C-blox: A Scalable and Consistent TSDF-based Dense Mapping Approach",2018,"recent","C08","full_slam_with_global_correction","C-blox 把場景表示為一組相互重疊的 TSDF 子體積（subvolume），每個子體積固定附著在 ORB-SLAM2 的一個關鍵影格上。迴圈閉合後，只要以最佳化後的關鍵影格位姿更新子體積座標系，就能修正稠密地圖，不必重新整合深度影像。為避免子體積數量隨軌跡長度無限增加，作者先以地標共視圖找出可能重複觀測同一區域的子體積對，再從光束法平差資訊矩陣計算兩者相對定位的條件共變異，只有定位品質足夠時才把兩個子體積融合，藉此限制地圖成長。整個系統只用 CPU，並在 voxblox 中加入多執行緒快速整合器，可在無人機上的 Intel NUC 即時執行。","CPU TSDF submapping: overlapping voxblox TSDF subvolumes are anchored to ORB-SLAM2 keyframes so loop closures correct the dense map by moving subvolume frames, and map growth is limited by fusing covisible subvolume pairs whose relative-pose conditional covariance, recovered from the bundle-adjustment information matrix, is small.","full_text_reviewed","peer_reviewed_published","background","論文未在施工現場測試。實機實驗為無人機在室內工業區飛行兩次（可見管線結構），只有定性結果，沒有幾何參考量測（Fig. 1、Sec. V-C）。corpus 中 vizzo2021puma 在 Table I 以 C-blox 作為 TSDF 基準。子體積附著於稀疏 SLAM 關鍵影格、迴圈閉合後整塊移動的作法，可作為樓層尺度室內掃描事後修正的參考（推論）。",[20,21],"public_benchmark","simulation",[23,24,25,26],"Median surface RMSE 0.015 m on ICL-NUIM with 0.02 m voxels, close to voxblox with ground-truth poses (0.013 m) without using a GPU (Table I)","On two CARLA drives the median RMSE was 0.68 m (fusion off) and 0.72 m (fusion on) versus 2.09 m for voxblox with uncorrected ORB-SLAM2 poses (Table II)","Subvolume fusion reduced the map to 55% (l0) and 68% (l1) of the allocated blocks without fusion (Table II; Sec. V-B)","Runs in real time entirely on the MAV computer (Sec. V-C)",[28,29,30,31,32],"Slightly higher error than GPU-based ElasticFusion on ICL-NUIM kt0 to kt2 (Table I; Sec. V-A)","Compression depends on the path and on the fusion threshold q; earlier fusion gives smaller maps but can cost accuracy (Sec. V-B)","The CARLA 'ground truth' is itself a voxblox reconstruction with 0.25 m voxels from ground-truth poses (Sec. V-B)","Requires 3D observations (stereo or depth) so that ORB-SLAM2 has metric scale (Sec. IV-B)","The fast integrator terminates rays early and enforces a time budget, trading completeness of updates for speed (Sec. IV-C)",[34,35,36],"Stereo camera pair of a VI-sensor (global shutter, tightly synchronized) for ORB-SLAM2 tracking on the MAV","RGB-D camera Intel RealSense D415 (coloured pointclouds) for dense integration on the MAV","Synthetic RGB-D input (ICL-NUIM) and simulated RGB plus noiseless depth (CARLA) in the evaluations",[38,39],"UAV (hexacopter on a DJI F550 frame, two flights in an indoor industrial area)","simulation (CARLA car drives through two synthetic cities; ICL-NUIM synthetic living room)","Modified ORB-SLAM2 (keyframe bundle adjustment of feature re-projection errors with Huber cost) supplies camera poses; each subvolume is rigidly attached to a keyframe and moved when keyframe poses are re-optimized; no dense tracking against the TSDF","Sparse ORB features for tracking; subvolume-fusion candidates from a landmark covisibility graph (edge weight = number of shared landmarks), accepted only if the relative localization quality q = 1\u002F||Sigma_i|j|| from the bundle-adjustment information matrix exceeds a threshold (Schur complement, constrained AMD reordering and Cholesky-based covariance recovery)","discrete keyframe poses","not_applicable (depth camera input)","ORB-SLAM2 loop detection and bundle adjustment; the dense map is corrected by updating subvolume base frames with the optimized keyframe poses","Sparse keyframe bundle adjustment in ORB-SLAM2; dense map corrected rigidly per subvolume, without re-integration of depth frames","collection of overlapping TSDF subvolumes (voxblox) with no spatial partitioning; a new subvolume starts after a maximum number of keyframes or a large sparse-map change; redundant well-localized subvolumes are fused by trilinear interpolation","none","TSDF subvolume collection and fused mesh; voxel size 0.02 m on ICL-NUIM and 0.5 m on CARLA","CPU only; on-board Intel NUC Core i7-7567U with tracking and integration at 10 Hz; the multi-threaded fast integrator cut depth integration from 93 ms to 25 ms per frame (Sec. V-C); desktop Intel Core i7-7820X for ICL-NUIM","https:\u002F\u002Fgithub.com\u002Fethz-asl\u002Fcblox","BSD-3-Clause (LICENSE file checked)",[53,57,60],{"relation":54,"title":55,"doi_or_url":56},"preprint","C-blox (arXiv v1 to v3)","https:\u002F\u002Farxiv.org\u002Fabs\u002F1710.07242",{"relation":58,"title":59,"doi_or_url":50},"code_release","ethz-asl\u002Fcblox",{"relation":61,"title":62,"doi_or_url":63},"follow_up_method","Voxgraph (RA-L 2020), which cites C-blox as the authors' previous work and reuses its indoor MAV dataset","10.1109\u002FLRA.2019.2953859",{"id":5,"kind":65,"shortName":7,"title":8,"authors":66,"year":9,"venue":73,"venueType":74,"publisher":75,"volumeIssuePages":76,"doi":77,"arxivId":78,"url":79,"firstPublicDate":80,"publicationStatus":16,"metadataStatus":81,"fulltextStatus":15,"era":10,"classicReason":82,"codeUrl":50,"cluster":11,"topics":83,"mdpi":84,"verification":85,"label":6,"fulltextRoute":86,"versionRead":87,"addedByCensus":88},"method",[67,68,69,70,71,72],"Alexander Millane","Zachary Taylor","Helen Oleynikova","Juan Nieto","Roland Siegwart","Cesar Cadena","2018 IEEE\u002FRSJ International Conference on Intelligent Robots and Systems (IROS)","conference","IEEE","pp. 995-1002","10.1109\u002Firos.2018.8593427","1710.07242","https:\u002F\u002Fapi.crossref.org\u002Fworks\u002F10.1109\u002FIROS.2018.8593427","2017-10-19","metadata_verified","not_applicable",[11],false,"corrected","arXiv","arXiv 1710.07242v3 (2018-09-25), camera-ready era version of the IROS 2018 paper; IEEE version of record not compared",true,[90,97,101,106,111],{"category":91,"model":92,"canonical":92,"role":93,"dataset":94,"specs":95,"locator":96},"platform","hexacopter based on a DJI F550 frame","method input",null,"px4 autopilot for low-level attitude stabilization; all high-level computation on board","Sec. V-C",{"category":98,"model":99,"canonical":99,"role":93,"dataset":94,"specs":100,"locator":96},"stereo_camera","Visual-Inertial (VI-)sensor [28]","tightly synchronized stereo images from a pair of global shutter cameras; used for camera tracking at 10 Hz",{"category":102,"model":103,"canonical":104,"role":93,"dataset":94,"specs":105,"locator":96},"rgbd","Intel RealSense D415 Depth Camera","Intel RealSense D415","coloured pointclouds integrated at 10 Hz; extrinsics to the tracked camera calibrated offline with Kalibr",{"category":107,"model":108,"canonical":108,"role":109,"dataset":94,"specs":110,"locator":96},"compute","Intel NUC Core i7-7567U","compute for runtime","on board; runs the proposed approach entirely in real time",{"category":107,"model":112,"canonical":112,"role":109,"dataset":94,"specs":113,"locator":114},"desktop PC with Intel Core i7-7820X CPU at 3.60 GHz and Nvidia Titan Xp GPU","used for the ICL-NUIM reconstruction comparison; the GPU was used only by ElasticFusion","Sec. V-A",[],{"totalRows":117,"groupCount":118,"groups":119,"others":297},16,3,[120,204,266],{"slug":121,"group":122,"sourceId":5,"sourceLabel":6,"table":123,"selfRows":124,"metrics":125,"seqs":134,"entrants":143,"cells":153,"outcomes":197,"locators":198,"hardware":199,"wordings":200,"notes":201},"cblox2018-table-ii","cblox2018:Table II","Table II",10,[126,131],{"label":127,"unit":128,"statistic":129,"alignment":130},"RMSE (m)","m","RMSE","not_reported",{"label":132,"unit":133,"statistic":130,"alignment":130},"Size (blocks)","voxel blocks",[135,139,141],{"dataset":136,"sequence":137,"environment":138},"CARLA (simulated)","l0","simulated urban driving (CARLA)",{"dataset":136,"sequence":140,"environment":138},"l1",{"dataset":136,"sequence":142,"environment":138},"median of l0 and l1",[144,147,149,151],{"name":145,"methodId":146,"linkable":88,"proposed":84,"self":84},"Voxblox (GT Poses)","oleynikova2017voxblox",{"name":148,"methodId":146,"linkable":88,"proposed":84,"self":84},"Voxblox (ORB-SLAM Poses)",{"name":150,"methodId":5,"linkable":88,"proposed":84,"self":88},"Ours (subvolume fusion OFF, ablation)",{"name":152,"methodId":5,"linkable":88,"proposed":88,"self":88},"Ours (subvolume fusion ON)",[154,158,161,164,166,168,170,172,173,175,177,179,181,183,185,187,189,191,193,195],[155,155,155,156,157,155,157,157,155],0,0.52,-1,[159,155,155,160,157,155,157,157,155],1,2.12,[162,155,155,163,157,155,157,157,155],2,0.59,[118,155,155,165,157,155,157,157,155],0.66,[155,155,159,167,157,155,157,157,155],0.7,[159,155,159,169,157,155,157,157,155],2.06,[162,155,159,171,157,155,157,157,155],0.77,[118,155,159,171,157,155,157,157,155],[155,155,162,174,157,155,157,157,155],0.6,[159,155,162,176,157,155,157,157,155],2.09,[162,155,162,178,157,155,157,157,155],0.68,[118,155,162,180,157,155,157,157,155],0.72,[155,159,155,182,157,155,157,157,159],6315,[159,159,155,184,157,155,157,157,159],7205,[162,159,155,186,157,155,157,157,159],28908,[118,159,155,188,157,155,157,157,159],15873,[155,159,159,190,157,155,157,157,159],4561,[159,159,159,192,157,155,157,157,159],5240,[162,159,159,194,157,155,157,157,159],17856,[118,159,159,196,157,155,157,157,159],12220,[],[123],[],[],[202,203],"CARLA simulated drives l0 and l1 through two synthetic cities; reference geometry is a voxblox reconstruction with ground-truth poses and 0.25 m voxels; evaluated systems use 0.5 m voxels; tracking and integration at 10 Hz; 'Voxblox (ORB-SLAM Poses)' has no dense-map correction after loop closure","CARLA simulated drives l0 and l1 through two synthetic cities; reference geometry is a voxblox reconstruction with ground-truth poses and 0.25 m voxels; evaluated systems use 0.5 m voxels; tracking and integration at 10 Hz; 'Voxblox (ORB-SLAM Poses)' has no dense-map correction after loop closure; map size as number of allocated voxel blocks",{"slug":205,"group":206,"sourceId":5,"sourceLabel":6,"table":207,"selfRows":208,"metrics":209,"seqs":211,"entrants":224,"cells":231,"outcomes":260,"locators":261,"hardware":262,"wordings":263,"notes":264},"cblox2018-table-i","cblox2018:Table I","Table I",5,[210],{"label":127,"unit":128,"statistic":129,"alignment":130},[212,216,218,220,222],{"dataset":213,"sequence":214,"environment":215},"ICL-NUIM","kt0","synthetic indoor living room",{"dataset":213,"sequence":217,"environment":215},"kt1",{"dataset":213,"sequence":219,"environment":215},"kt2",{"dataset":213,"sequence":221,"environment":215},"kt3",{"dataset":213,"sequence":223,"environment":215},"median of kt0 to kt3",[225,228,229],{"name":226,"methodId":227,"linkable":88,"proposed":84,"self":84},"ElasticFusion","elasticfusion2015",{"name":145,"methodId":146,"linkable":88,"proposed":84,"self":84},{"name":230,"methodId":5,"linkable":88,"proposed":88,"self":88},"Ours (C-blox)",[232,234,236,238,240,242,244,245,247,249,251,252,254,257,258],[155,155,155,233,157,155,157,157,155],0.006,[159,155,155,235,157,155,157,157,155],0.01,[162,155,155,237,157,155,157,157,155],0.011,[155,155,159,239,157,155,157,157,155],0.009,[159,155,159,241,157,155,157,157,155],0.017,[162,155,159,243,157,155,157,157,155],0.024,[155,155,162,235,157,155,157,157,155],[159,155,162,246,157,155,157,157,155],0.014,[162,155,162,248,157,155,157,157,155],0.016,[155,155,118,250,157,155,157,157,155],0.048,[159,155,118,237,157,155,157,157,155],[162,155,118,253,157,155,157,157,155],0.013,[155,155,255,256,157,155,157,157,155],4,0.001,[159,155,255,253,157,155,157,157,155],[162,155,255,259,157,155,157,157,155],0.015,[],[207],[],[],[265],"ICL-NUIM living room (synthetic, noisy depth); RMSE between mesh vertices (or surfel centres) and the closest ground-truth surface point after alignment (method not stated); voxel size 0.02 m for voxblox-based systems; kt3 is the only sequence with a loop closure; the printed ElasticFusion median (0.001 m) is inconsistent with its per-sequence values",{"slug":267,"group":268,"sourceId":5,"sourceLabel":6,"table":269,"selfRows":159,"metrics":270,"seqs":275,"entrants":280,"cells":285,"outcomes":290,"locators":291,"hardware":292,"wordings":294,"notes":295},"cblox2018-text-sec-v-c","cblox2018:Text Sec.V-C","Text Sec.V-C",[271],{"label":272,"unit":273,"statistic":274,"alignment":47},"average time required to integrate depth data","ms","mean",[276],{"dataset":277,"sequence":278,"environment":279},"authors' MAV industrial flights","f0 and f1","indoor industrial area, MAV",[281,283],{"name":282,"methodId":146,"linkable":88,"proposed":84,"self":84},"voxblox original integrator",{"name":284,"methodId":5,"linkable":88,"proposed":88,"self":88},"C-blox fast integrator",[286,288],[155,155,155,287,157,155,155,157,155],93,[159,155,155,289,157,155,155,157,155],25,[],[96],[293],"Intel NUC Core i7-7567U (on board)",[],[296],"Average time to integrate depth data per frame in the industrial MAV environment, original voxblox integrator versus the proposed fast integrator",[],1790510662233]