[{"data":1,"prerenderedAt":374},["ShallowReactive",2],{"method-flashfusion2018":3},{"method":4,"reference":55,"equipment":75,"figures":91,"results":92},{"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":32,"platform":34,"estimator":38,"association":39,"timeModel":40,"deskew":41,"loopClosure":42,"globalOptimization":43,"mapRepresentation":44,"prior":45,"outputGeometry":46,"compute":47,"codeUrl":48,"codeLicense":49,"relatedVersions":50},"flashfusion2018","Han & Fang, 2018","FlashFusion","FlashFusion: Real-time Globally Consistent Dense 3D Reconstruction using CPU Computing",2018,"recent","C08","full_slam_with_global_correction","FlashFusion 是不使用 GPU 運算、可在可攜裝置上即時運作的全域一致稠密 RGB-D 重建系統。定位端以 ORB 特徵對應，把每個新關鍵影格與 MILD 迴圈偵測找出的前 5 個相似關鍵影格做全域配準，並以作者的 FastGO 預先累積對應點的二階統計量，使全域位姿最佳化能在 CPU 上即時求解；另以只修正相對位姿變化最大的 10 對影格來近似 Huber 穩健損失。重建端採空間雜湊 TSDF，以每個區塊 8 個角點的稀疏體素取樣快速挑出含表面的有效區塊（先以 20 mm、再以 5 mm 解析度檢查），且只在關鍵影格做篩選；網格擷取以自適應門檻、查表與鄰近區塊位址表加速。位姿更新後，每個關鍵影格重新整合至多 10 個先前關鍵影格，以 CPU 近似 BundleFusion 的重新整合。","CPU-only globally consistent RGB-D reconstruction: keyframe global pose optimization with pre-integrated ORB-correspondence statistics (FastGO) and MILD loop detection, a spatially hashed TSDF with sparse-corner valid-chunk selection, accelerated marching cubes, and bounded keyframe re-integration after pose updates; 5 mm voxels with TSDF fusion at 300 Hz and meshing at 25 Hz.","full_text_reviewed","peer_reviewed_published","background","論文未在施工現場測試，評估使用 ICL-NUIM 合成室內、TUM RGB-D 辦公室與 ElasticFusion 的 dyson lab（整間實驗室 6400 張影格）資料，並以 Asus Xtion 接 Surface Pro 平板做即時掃描示範。只用 CPU 即達 5 mm 體素與全域一致，適合沒有 GPU 的輕量手持室內掃描設備；但作者的 GitHub 儲存庫在本次查核時只有說明檔、尚未釋出程式碼，實際採用受限。",[20,21],"public_benchmark","simulation",[23,24,25,26],"At 5 mm voxels TSDF fusion takes 3.6 ms and mesh extraction 38.4 ms, versus 483 ms and 1518 ms for CHISEL (Table IV)","ICL-NUIM surface error 0.8 to 1.3 cm on a CPU, close to BundleFusion (0.5 to 0.8 cm), which uses two high-end GPUs (Table III)","Sparse voxel sampling keeps more than 98% of truly valid chunks at about 2% of the computation on fr3\u002Foffice (Sec. IV-B2, Fig. 4)","Globally consistent 5 mm reconstruction live on a Surface Pro tablet CPU (Sec. V)",[28,29,30,31],"Keyframe-based fusion leaves minor artifacts at keyframe borders because surfaces seen only by local frames outside the keyframe are skipped (Sec. VI, Fig. 9)","Localization and surface accuracy remain below BundleFusion (Tables I to III)","Re-integration is limited to 10 previous keyframes with at most 10 local frames each per keyframe, so older frames are not re-integrated in full (Sec. IV-D) (inference)","The efficiency comparison is with CPU systems only (CHISEL, FastFusion); GPU systems are compared on accuracy (Sec. V-B)",[33],"RGB-D camera (Asus Xtion for live scanning; TUM RGB-D real sequences; synthetic noisy ICL-NUIM)",[35,36,37],"portable tablet set-up: live scanning with an Asus Xtion connected to a Microsoft Surface Pro, carrying mode not described (Sec. V)","real RGB-D benchmark sequences (TUM RGB-D, ElasticFusion dyson lab)","simulation (ICL-NUIM)","Keyframe global pose optimization (authors' FastGO): Gauss-Newton on SE(3) minimizing distances between corresponding ORB feature points of each keyframe and its top-5 most similar keyframes, with second-order statistics pre-integrated so each frame pair costs O(1); Huber norm approximated by an online correction of the 10 frame pairs whose relative poses change most; local frames fixed relative to their keyframe","About 1000 ORB features per frame; local registration of each frame to its keyframe and global registration of each new keyframe against the top-5 similar previous keyframes found by MILD appearance-based loop detection","discrete poses (keyframes optimized, local frames attached)","not_applicable (RGB-D input)","MILD multi-index hashing loop detection on ORB features (no training); matched keyframes added to the global optimization","Global keyframe pose optimization on every new keyframe; TSDF re-integration of up to 10 previous keyframes, each with at most 10 evenly selected local frames","spatially hashed TSDF with 8 x 8 x 8-voxel chunks and a second coarse hash of chunk cubes; valid chunks selected by sparse voxel sampling of the 8 chunk corners (first at 20 mm, then 5 mm) only on keyframes; colour stored as colour times weight","none","coloured triangle mesh with normals from accelerated marching cubes (adaptive per-chunk threshold, one-DoF vertex placement, neighbour-chunk look-up tables); 5 mm voxels","CPU only (GPU used only for visualization); experiments on Intel Core i7 7700 @3.6 GHz; live demo on a Microsoft Surface Pro tablet; tracking 30 Hz, TSDF fusion 300 Hz at 5 mm and 900 Hz at 1 cm, meshing 25 Hz",null,"not_applicable (no code released; the GitHub repository contains only a README stating 'Source code coming soon!')",[51],{"relation":52,"title":53,"doi_or_url":54},"project_page","lhanaf\u002FFlashFusion GitHub repository (README only, no code released at check time) and project website www.luvision.net\u002FFlashFusion cited in the paper","https:\u002F\u002Fgithub.com\u002Flhanaf\u002FFlashFusion",{"id":5,"kind":56,"shortName":7,"title":8,"authors":57,"year":9,"venue":60,"venueType":61,"publisher":62,"volumeIssuePages":63,"doi":64,"arxivId":48,"url":65,"firstPublicDate":66,"publicationStatus":16,"metadataStatus":67,"fulltextStatus":15,"era":10,"classicReason":68,"codeUrl":48,"cluster":11,"topics":69,"mdpi":70,"verification":71,"label":6,"fulltextRoute":72,"versionRead":73,"addedByCensus":74},"method",[58,59],"Lei Han","Lu Fang","Robotics: Science and Systems XIV (RSS 2018)","conference","Robotics: Science and Systems Foundation","not_reported (paper 6)","10.15607\u002Frss.2018.xiv.006","https:\u002F\u002Fwww.roboticsproceedings.org\u002Frss14\u002Fp06.pdf","2018-06-26","metadata_verified","not_applicable",[11],false,"corrected","publisher OA","RSS XIV online proceedings PDF (version of record, paper p06)",true,[76,82,88],{"category":77,"model":78,"canonical":78,"role":79,"dataset":48,"specs":80,"locator":81},"rgbd","Asus Xtion sensor","method input","live scanning at 5 mm voxel resolution","Sec. V",{"category":83,"model":84,"canonical":84,"role":85,"dataset":48,"specs":86,"locator":87},"compute","Microsoft Surface Pro tablet","compute for runtime","portable tablet; FlashFusion runs on its CPU for live scanning","Sec. V; footnote 1",{"category":83,"model":89,"canonical":89,"role":85,"dataset":48,"specs":90,"locator":81},"Intel Core i7 7700 @3.6GHz CPU","CPU used for all dataset experiments",[],{"totalRows":93,"groupCount":94,"groups":95,"others":348},30,9,[96,158,236,299],{"slug":97,"group":98,"sourceId":5,"sourceLabel":6,"table":99,"selfRows":100,"metrics":101,"seqs":108,"entrants":117,"cells":122,"outcomes":151,"locators":152,"hardware":153,"wordings":155,"notes":156},"flashfusion2018-table-iv","flashfusion2018:Table IV","Table IV",6,[102,106],{"label":103,"unit":104,"statistic":105,"alignment":105},"TSDF Fusion (ms)","ms","not_reported",{"label":107,"unit":104,"statistic":105,"alignment":105},"Mesh Extraction (ms)",[109,113,115],{"dataset":110,"sequence":111,"environment":112},"TUM RGB-D","fr3\u002Foffice (5mm voxels)","indoor office desk loop",{"dataset":110,"sequence":114,"environment":112},"fr3\u002Foffice (10mm voxels)",{"dataset":110,"sequence":116,"environment":112},"fr3\u002Foffice (20mm voxels)",[118,121],{"name":119,"methodId":120,"linkable":74,"proposed":70,"self":70},"CHISEL","chisel2015",{"name":7,"methodId":5,"linkable":74,"proposed":74,"self":74},[123,127,130,132,134,136,138,140,142,145,147,149],[124,124,124,125,126,124,124,126,124],0,483,-1,[124,128,124,129,126,124,124,126,124],1,1518,[128,124,124,131,126,124,124,126,124],3.6,[128,128,124,133,126,124,124,126,124],38.4,[124,124,128,135,126,124,124,126,124],86,[124,128,128,137,126,124,124,126,124],312,[128,124,128,139,126,124,124,126,124],1.1,[128,128,128,141,126,124,124,126,124],19.7,[124,124,143,144,126,124,124,126,124],2,15,[124,128,143,146,126,124,124,126,124],70,[128,124,143,148,126,124,124,126,124],0.7,[128,128,143,150,126,124,124,126,124],6.5,[],[99],[154],"Intel Core i7 7700 @3.6 GHz (CPU only)",[],[157],"Efficiency comparison between CPU-based CHISEL and FlashFusion on TUM fr3\u002Foffice at three voxel resolutions; time per operation in ms",{"slug":159,"group":160,"sourceId":161,"sourceLabel":162,"table":163,"selfRows":164,"metrics":165,"seqs":170,"entrants":181,"cells":195,"outcomes":230,"locators":231,"hardware":232,"wordings":233,"notes":234},"densesurfelmapping2019-table-i","densesurfelmapping2019:Table I","densesurfelmapping2019","Wang et al., 2019","Table I",4,[166],{"label":167,"unit":168,"statistic":169,"alignment":105},"reconstruction accuracy (cm)","cm","mean",[171,175,177,179],{"dataset":172,"sequence":173,"environment":174},"ICL-NUIM","kt0","synthetic indoor living room",{"dataset":172,"sequence":176,"environment":174},"kt1",{"dataset":172,"sequence":178,"environment":174},"kt2",{"dataset":172,"sequence":180,"environment":174},"kt3",[182,185,188,190,191,193],{"name":183,"methodId":184,"linkable":74,"proposed":70,"self":70},"BundleFusion","bundlefusion2017",{"name":186,"methodId":187,"linkable":74,"proposed":70,"self":70},"ElasticFusion","elasticfusion2015",{"name":189,"methodId":48,"linkable":70,"proposed":70,"self":70},"InfiniTAM [13] (Kaehler et al. ECCV 2016)",{"name":7,"methodId":5,"linkable":74,"proposed":70,"self":74},{"name":192,"methodId":161,"linkable":74,"proposed":74,"self":70},"Ours",{"name":194,"methodId":161,"linkable":74,"proposed":70,"self":70},"Ours w\u002Fo loop (ablation: ORB-SLAM2 loop closure disabled)",[196,198,199,201,204,205,207,209,210,211,212,214,215,216,217,219,220,221,222,223,225,226,227,228],[124,124,124,197,126,124,126,126,124],0.5,[128,124,124,148,126,124,126,126,124],[143,124,124,200,126,124,126,126,124],1.3,[202,124,124,203,126,124,126,126,124],3,0.8,[164,124,124,148,126,124,126,126,124],[206,124,124,148,126,124,126,126,124],5,[124,124,128,208,126,124,126,126,124],0.6,[128,124,128,148,126,124,126,126,124],[143,124,128,139,126,124,126,126,124],[202,124,128,203,126,124,126,126,124],[164,124,128,213,126,124,126,126,124],0.9,[206,124,128,213,126,124,126,126,124],[124,124,143,148,126,124,126,126,124],[128,124,143,203,126,124,126,126,124],[143,124,143,218,126,124,126,126,124],0.1,[202,124,143,128,126,124,126,126,124],[164,124,143,139,126,124,126,126,124],[206,124,143,139,126,124,126,126,124],[124,124,202,203,126,124,126,126,124],[128,124,202,224,126,124,126,126,124],2.8,[143,124,202,224,126,124,126,126,124],[202,124,202,200,126,124,126,126,124],[164,124,202,203,126,124,126,126,124],[206,124,202,229,126,124,126,126,124],1.7,[],[163],[],[],[235],"ICL-NUIM living room with simulated noise; reconstruction accuracy = mean difference between reconstructed model and ground-truth model; ORB-SLAM2 RGB-D mode for tracking, G_delta = 20; only FlashFusion and Ours run without GPU; 'Ours w\u002Fo loop' disables ORB-SLAM2 loop closure; comparator values coincide with FlashFusion Table III",{"slug":237,"group":238,"sourceId":5,"sourceLabel":6,"table":163,"selfRows":164,"metrics":239,"seqs":243,"entrants":253,"cells":262,"outcomes":293,"locators":294,"hardware":295,"wordings":296,"notes":297},"flashfusion2018-table-i","flashfusion2018:Table I",[240],{"label":241,"unit":168,"statistic":242,"alignment":105},"ATE rmse (cm)","RMSE",[244,247,249,251],{"dataset":110,"sequence":245,"environment":246},"fr1\u002Fdesk","real indoor office scenes, RGB-D camera (carrying mode not stated in the paper)",{"dataset":110,"sequence":248,"environment":246},"fr2\u002Fxyz",{"dataset":110,"sequence":250,"environment":246},"fr3\u002Foffice",{"dataset":110,"sequence":252,"environment":246},"fr3\u002Fnst",[254,256,257,259,261],{"name":255,"methodId":48,"linkable":70,"proposed":70,"self":70},"RGBD SLAM [3] (Endres et al.)",{"name":186,"methodId":187,"linkable":74,"proposed":70,"self":70},{"name":258,"methodId":184,"linkable":74,"proposed":70,"self":70},"BundleFusion (on-line)",{"name":260,"methodId":184,"linkable":74,"proposed":70,"self":70},"BundleFusion (off-line)",{"name":7,"methodId":5,"linkable":74,"proposed":74,"self":74},[263,265,266,267,269,271,272,273,275,276,277,279,280,282,284,286,287,288,289,291],[124,124,124,264,126,124,126,126,124],2.3,[128,124,124,143,126,124,126,126,124],[143,124,124,229,126,124,126,126,124],[202,124,124,268,126,124,126,126,124],1.6,[164,124,124,270,126,124,126,126,124],1.9,[124,124,128,203,126,124,126,126,124],[128,124,128,139,126,124,126,126,124],[143,124,128,274,126,124,126,126,124],1.4,[202,124,128,139,126,124,126,126,124],[164,124,128,200,126,124,126,126,124],[124,124,143,278,126,124,126,126,124],3.2,[128,124,143,229,126,124,126,126,124],[143,124,143,281,126,124,126,126,124],2.9,[202,124,143,283,126,124,126,126,124],2.2,[164,124,143,285,126,124,126,126,124],2.5,[124,124,202,229,126,124,126,126,124],[128,124,202,268,126,124,126,126,124],[143,124,202,268,126,124,126,126,124],[202,124,202,290,126,124,126,126,124],1.2,[164,124,202,292,126,124,126,126,124],1.8,[],[163],[],[],[298],"Localization accuracy on TUM RGB-D as ATE RMSE (Sturm et al.) in cm; alignment not stated",{"slug":300,"group":301,"sourceId":5,"sourceLabel":6,"table":302,"selfRows":164,"metrics":303,"seqs":305,"entrants":311,"cells":317,"outcomes":342,"locators":343,"hardware":344,"wordings":345,"notes":346},"flashfusion2018-table-ii","flashfusion2018:Table II","Table II",[304],{"label":241,"unit":168,"statistic":242,"alignment":105},[306,308,309,310],{"dataset":172,"sequence":173,"environment":307},"synthetic living room",{"dataset":172,"sequence":176,"environment":307},{"dataset":172,"sequence":178,"environment":307},{"dataset":172,"sequence":180,"environment":307},[312,313,314,315,316],{"name":255,"methodId":48,"linkable":70,"proposed":70,"self":70},{"name":186,"methodId":187,"linkable":74,"proposed":70,"self":70},{"name":258,"methodId":184,"linkable":74,"proposed":70,"self":70},{"name":260,"methodId":184,"linkable":74,"proposed":70,"self":70},{"name":7,"methodId":5,"linkable":74,"proposed":74,"self":74},[318,320,321,322,323,324,325,326,327,329,330,331,332,333,334,335,337,339,340,341],[124,124,124,319,126,124,126,126,124],2.6,[128,124,124,213,126,124,126,126,124],[143,124,124,203,126,124,126,126,124],[202,124,124,208,126,124,126,126,124],[164,124,124,148,126,124,126,126,124],[124,124,128,203,126,124,126,126,124],[128,124,128,213,126,124,126,126,124],[143,124,128,197,126,124,126,126,124],[202,124,128,328,126,124,126,126,124],0.4,[164,124,128,203,126,124,126,126,124],[124,124,143,292,126,124,126,126,124],[128,124,143,274,126,124,126,126,124],[143,124,143,139,126,124,126,126,124],[202,124,143,328,126,124,126,126,124],[164,124,143,139,126,124,126,126,124],[124,124,202,336,126,124,126,126,124],43.3,[128,124,202,338,126,124,126,126,124],10.6,[143,124,202,290,126,124,126,126,124],[202,124,202,139,126,124,126,126,124],[164,124,202,274,126,124,126,126,124],[],[302],[],[],[347],"Localization accuracy on ICL-NUIM (with noise) as ATE RMSE in cm; alignment not stated",[349,354,359,364,369],{"group":350,"slug":351,"sourceLabel":6,"table":352,"selfRows":164,"datasets":353},"flashfusion2018:Table III","flashfusion2018-table-iii","Table III",[172],{"group":355,"slug":356,"sourceLabel":6,"table":357,"selfRows":143,"datasets":358},"flashfusion2018:Text Sec.IV-B2","flashfusion2018-text-sec-iv-b2","Text Sec.IV-B2",[110],{"group":360,"slug":361,"sourceLabel":6,"table":362,"selfRows":143,"datasets":363},"flashfusion2018:Text Sec.IV-B3","flashfusion2018-text-sec-iv-b3","Text Sec.IV-B3",[105],{"group":365,"slug":366,"sourceLabel":6,"table":367,"selfRows":143,"datasets":368},"flashfusion2018:Text Sec.V-B","flashfusion2018-text-sec-v-b","Text Sec.V-B",[110,105],{"group":370,"slug":371,"sourceLabel":6,"table":372,"selfRows":143,"datasets":373},"flashfusion2018:Text Sec.VI","flashfusion2018-text-sec-vi","Text Sec.VI",[105],1790510658126]