[{"data":1,"prerenderedAt":372},["ShallowReactive",2],{"method-overlapnet2020":3},{"method":4,"reference":56,"equipment":79,"figures":101,"results":131},{"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":36,"association":37,"timeModel":38,"deskew":39,"loopClosure":40,"globalOptimization":41,"mapRepresentation":42,"prior":43,"outputGeometry":38,"compute":44,"codeUrl":45,"codeLicense":46,"relatedVersions":47},"overlapnet2020","Chen et al., 2020","OverlapNet","OverlapNet: Loop Closing for LiDAR-based SLAM",2020,"recent","C06","place_recognition_component","OverlapNet 以孿生網路（siamese network）比較兩次光達掃描，輸入由單次掃描產生的距離影像、法向量、強度與語意機率，輸出兩者的重疊率與相對偏航角。系統以位姿共變異數傳播決定迴圈搜尋範圍，取代 SuMa 原本只取最近幀的啟發式迴圈偵測，並可用預測的偏航角作為 ICP 初值。","OverlapNet learns scan-to-scan overlap and relative yaw from multi-cue range images and uses them, with covariance-bounded search, for loop closing in SuMa.","full_text_reviewed","peer_reviewed_published","background","not_reported（僅都市駕駛資料 KITTI 與 Ford Campus）",[20,21],"public_benchmark","cross_site",[23,24,25,26],"Generalized to the Ford Campus dataset while trained only on KITTI (Sec. IV; abstract)","Overlap gives a measure of loop-closure quality and yaw provides ICP initialization (Sec. IV-F, V)","With covariance-propagated search (CovNearestOfTop10) AUC and F1 reached 0.96 on KITTI 00, versus 0.87 and 0.88 without prior (Table III)","Yaw error mean 1.13 deg and std 3.34 deg without ICP, versus 12.67 and 15.23 deg for OREOS (Table IV)",[28,29,30,31],"(inference) Learned model trained on one car-mounted HDL-64E dataset (KITTI 03 to 10); only a different HDL-64E version on Ford Campus was tested, and no indoor or handheld data","Semantic cues (RangeNet++) increase computation time; the full cue set is only available on KITTI (Sec. IV-E, IV-G)","Without prior pose information, OREOS and LocNet++ reach slightly higher recall when more candidates are considered (Sec. IV-C)","On Ford Campus without the covariance prior, F1 was 0.84 versus 0.85 for M2DP; the authors attribute this to no training on US roads and geometry-only input (Sec. IV-A, Table II)",[33],"3D LiDAR",[35],"vehicle","not_applicable (integrated into SuMa surfel SLAM with pose graph)","learned siamese network on range, normal, intensity and semantic-probability images predicting scan overlap and relative yaw","not_applicable","not_reported","overlap-based candidate detection within a covariance-propagated search region, replacing SuMa's nearest-frame heuristic; yaw estimate can initialize ICP","via host SLAM (SuMa) pose graph","not_applicable (host uses surfels)","trained network (KITTI); odometry covariance for search region","Intel i7-8700 3.2 GHz with Nvidia GeForce GTX1080 Ti 11 GB; KITTI 00 with all cues: 75 ms preprocessing, 6 ms leg feature extraction and 27 ms head matching per frame (worst case 630 ms for all candidates); Ford Campus geometry only: 10 ms, 2 ms and 24 ms (worst case 550 ms); 17 ms per scan pair with depth and normal cues versus 1.2 s for exhaustive overlap evaluation","https:\u002F\u002Fgithub.com\u002FPRBonn\u002FOverlapNet","MIT (LICENSE.txt)",[48,52],{"relation":49,"title":50,"doi_or_url":51},"journal_extension","OverlapNet: a siamese network for computing LiDAR scan similarity with applications to loop closing and localization (Autonomous Robots 46(1):61-81, 2022; online 2021)","10.1007\u002Fs10514-021-09999-0",{"relation":53,"title":54,"doi_or_url":55},"preprint","arXiv posting of the RSS paper (2021-05-24, after conference)","https:\u002F\u002Farxiv.org\u002Fabs\u002F2105.11344",{"id":5,"kind":57,"shortName":7,"title":8,"authors":58,"year":9,"venue":67,"venueType":68,"publisher":69,"volumeIssuePages":39,"doi":70,"arxivId":71,"url":55,"firstPublicDate":72,"publicationStatus":16,"metadataStatus":73,"fulltextStatus":15,"era":10,"classicReason":38,"codeUrl":45,"cluster":11,"topics":74,"mdpi":75,"verification":76,"label":6,"fulltextRoute":77,"versionRead":78,"addedByCensus":75},"component",[59,60,61,62,63,64,65,66],"Xieyuanli Chen","Thomas Läbe","Andres Milioto","Timo Röhling","Olga Vysotska","Alexandre Haag","Jens Behley","Cyrill Stachniss","Robotics: Science and Systems XVI","conference","RSS Foundation","10.15607\u002Frss.2020.xvi.009","2105.11344","2020-07-12","metadata_verified",[11],false,"confirmed","arXiv","arXiv 2105.11344v1 (24 May 2021), author posting of the RSS 2020 paper (comment 'Accepted by RSS 2020'); RSS proceedings version not compared line by line",[80,87,91,98],{"category":81,"model":82,"canonical":82,"role":83,"dataset":84,"specs":85,"locator":86},"lidar","Velodyne HDL-64E","dataset sensor","KITTI odometry","not_reported (the paper gives only processing settings: 64 x 900 range-image input and a 75 m cutoff for overlap ground truth)","Sec. IV; Sec. III-C",{"category":81,"model":88,"canonical":88,"role":83,"dataset":89,"specs":39,"locator":90},"a different version of the Velodyne HDL-64E","Ford Campus","Sec. IV",{"category":92,"model":93,"canonical":93,"role":94,"dataset":95,"specs":96,"locator":97},"compute","Intel i7-8700","compute for runtime",null,"3.2 GHz","Sec. IV-G",{"category":92,"model":99,"canonical":99,"role":94,"dataset":95,"specs":100,"locator":97},"Nvidia GeForce GTX1080 Ti","11 GB memory",[102,115,123],{"refId":5,"refLabel":6,"fig":103,"whatZh":104,"license":105,"licenseUrl":106,"sourceUrl":107,"src":108,"width":109,"height":110,"thumb":111,"thumbWidth":112,"thumbHeight":113,"modified":114},"Fig. 1","迴圈位置兩次掃描（藍、橙點）在不同相對轉換下的重疊示意。","CC BY 4.0","https:\u002F\u002Fcreativecommons.org\u002Flicenses\u002Fby\u002F4.0\u002F","https:\u002F\u002Farxiv.org\u002Fhtml\u002F2105.11344v1\u002Fpics\u002Fmotivation.png","\u002Ffigure-files\u002Foverlapnet2020\u002Ffig-1.webp",1400,1149,"\u002Ffigure-files\u002Foverlapnet2020\u002Ffig-1.thumb.webp",480,394,"resized to at most 1400 px wide and converted to WebP",{"refId":5,"refLabel":6,"fig":116,"whatZh":117,"license":105,"licenseUrl":106,"sourceUrl":118,"src":119,"width":109,"height":120,"thumb":121,"thumbWidth":112,"thumbHeight":122,"modified":114},"Fig. 3","OverlapNet 流程：由單次掃描產生距離、法向量、強度與語意影像，經共享權重的雙腿與兩個輸出頭預測重疊率與偏航角。","https:\u002F\u002Farxiv.org\u002Fhtml\u002F2105.11344v1\u002Farchitecture.png","\u002Ffigure-files\u002Foverlapnet2020\u002Ffig-3.webp",494,"\u002Ffigure-files\u002Foverlapnet2020\u002Ffig-3.thumb.webp",169,{"refId":5,"refLabel":6,"fig":124,"whatZh":125,"license":105,"licenseUrl":106,"sourceUrl":126,"src":127,"width":109,"height":128,"thumb":129,"thumbWidth":112,"thumbHeight":130,"modified":114},"Fig. 4","Delta 層以串接與轉置計算兩組特徵體所有像素差值的示意。","https:\u002F\u002Farxiv.org\u002Fhtml\u002F2105.11344v1\u002Fdelta_layer.png","\u002Ffigure-files\u002Foverlapnet2020\u002Ffig-4.webp",830,"\u002Ffigure-files\u002Foverlapnet2020\u002Ffig-4.thumb.webp",285,{"totalRows":132,"groupCount":133,"groups":134,"others":361},44,6,[135,207,260,314],{"slug":136,"group":137,"sourceId":5,"sourceLabel":6,"table":138,"selfRows":139,"metrics":140,"seqs":153,"entrants":157,"cells":167,"outcomes":201,"locators":202,"hardware":203,"wordings":204,"notes":205},"overlapnet2020-table-v","overlapnet2020:Table V","Table V",16,[141,144,146,150],{"label":142,"unit":143,"statistic":39,"alignment":38},"overlap AUC","ratio",{"label":145,"unit":143,"statistic":39,"alignment":38},"overlap F1",{"label":147,"unit":148,"statistic":149,"alignment":38},"yaw angle Mean","deg","mean",{"label":151,"unit":148,"statistic":152,"alignment":38},"yaw angle Std","std",[154],{"dataset":155,"sequence":39,"environment":156},"not_reported (context: KITTI odometry)","urban driving",[158,161,163,165],{"name":159,"methodId":5,"linkable":160,"proposed":160,"self":160},"OverlapNet input: Depth",true,{"name":162,"methodId":5,"linkable":160,"proposed":160,"self":160},"OverlapNet input: Depth+Normals",{"name":164,"methodId":5,"linkable":160,"proposed":160,"self":160},"OverlapNet input: Depth+Normals+Intensity",{"name":166,"methodId":5,"linkable":160,"proposed":160,"self":160},"OverlapNet input: Depth+Normals+Intensity+Semantics",[168,172,175,178,181,182,184,186,188,189,190,192,194,195,197,199],[169,169,169,170,171,169,171,171,169],0,0.86,-1,[169,173,169,174,171,169,171,171,169],1,0.87,[169,176,169,177,171,169,171,171,169],2,11.67,[169,179,169,180,171,169,171,171,169],3,25.32,[173,169,169,170,171,169,171,171,169],[173,173,169,183,171,169,171,171,169],0.85,[173,176,169,185,171,169,171,171,169],2.97,[173,179,169,187,171,169,171,171,169],14.28,[176,169,169,174,171,169,171,171,169],[176,173,169,174,171,169,171,171,169],[176,176,169,191,171,169,171,171,169],2.53,[176,179,169,193,171,169,171,171,169],14.56,[179,169,169,174,171,169,171,171,169],[179,173,169,196,171,169,171,171,169],0.88,[179,176,169,198,171,169,171,171,169],1.13,[179,179,169,200,171,169,171,171,169],3.34,[],[138],[],[],[206],"Ablation on input modalities (overlap AUC and F1, yaw mean and std); dataset not named in the caption, context indicates KITTI because semantics are used",{"slug":208,"group":209,"sourceId":5,"sourceLabel":6,"table":210,"selfRows":211,"metrics":212,"seqs":217,"entrants":222,"cells":235,"outcomes":254,"locators":255,"hardware":256,"wordings":257,"notes":258},"overlapnet2020-table-iii","overlapnet2020:Table III","Table III",12,[213,215],{"label":214,"unit":143,"statistic":39,"alignment":38},"AUC (precision-recall)",{"label":216,"unit":143,"statistic":39,"alignment":38},"F1 score",[218,220],{"dataset":84,"sequence":219,"environment":156},"00",{"dataset":89,"sequence":219,"environment":221},"campus and downtown driving",[223,225,227,229,231,233],{"name":224,"methodId":5,"linkable":160,"proposed":160,"self":160},"MLPOnly",{"name":226,"methodId":5,"linkable":160,"proposed":160,"self":160},"DeltaOnly",{"name":228,"methodId":5,"linkable":160,"proposed":160,"self":160},"CovNearestOfTop10",{"name":230,"methodId":5,"linkable":160,"proposed":160,"self":160},"Ours (AllChannel, TwoHeads)",{"name":232,"methodId":5,"linkable":160,"proposed":160,"self":160},"Ours (GeoOnly)",{"name":234,"methodId":5,"linkable":160,"proposed":160,"self":160},"GeoCovNearestOfTop10",[236,238,240,241,242,244,245,246,247,249,251,253],[169,169,169,237,171,169,171,171,169],0.58,[169,173,169,239,171,169,171,171,169],0.65,[173,169,169,183,171,169,171,171,169],[173,173,169,196,171,169,171,171,169],[176,169,169,243,171,169,171,171,169],0.96,[176,173,169,243,171,169,171,171,169],[179,169,169,174,171,169,171,171,169],[179,173,169,196,171,169,171,171,169],[248,169,173,183,171,169,171,171,169],4,[248,173,173,250,171,169,171,171,169],0.84,[252,169,173,183,171,169,171,171,169],5,[252,173,173,196,171,169,171,171,169],[],[210],[],[],[259],"Comparison with OverlapNet variants; CovNearestOfTop10 uses covariance-propagated Mahalanobis search space (prior pose information)",{"slug":261,"group":262,"sourceId":5,"sourceLabel":6,"table":263,"selfRows":264,"metrics":265,"seqs":284,"entrants":289,"cells":291,"outcomes":307,"locators":308,"hardware":309,"wordings":311,"notes":312},"overlapnet2020-text-sec-iv-g","overlapnet2020:Text Sec.IV-G","Text Sec.IV-G",8,[266,269,271,273,276,278,280,282],{"label":267,"unit":268,"statistic":149,"alignment":38},"input preprocessing per frame","ms",{"label":270,"unit":268,"statistic":149,"alignment":38},"legs feature extraction per frame",{"label":272,"unit":268,"statistic":149,"alignment":38},"head matching per frame",{"label":274,"unit":268,"statistic":275,"alignment":38},"head matching worst case, all candidates in search space","max",{"label":277,"unit":268,"statistic":149,"alignment":38},"input generation per frame",{"label":279,"unit":268,"statistic":149,"alignment":38},"feature extraction per frame",{"label":281,"unit":268,"statistic":149,"alignment":38},"matching per frame",{"label":283,"unit":268,"statistic":275,"alignment":38},"matching worst case",[285,287],{"dataset":84,"sequence":286,"environment":156},"00 (all cues incl. RangeNet++ semantics)",{"dataset":89,"sequence":288,"environment":221},"00 (geometric cues only)",[290],{"name":7,"methodId":5,"linkable":160,"proposed":160,"self":160},[292,294,295,297,299,301,302,304],[169,169,169,293,171,169,169,171,169],75,[169,173,169,133,171,169,169,171,169],[169,176,169,296,171,169,169,171,169],27,[169,179,169,298,171,169,169,171,169],630,[169,248,173,300,171,169,169,171,169],10,[169,252,173,176,171,169,169,171,169],[169,133,173,303,171,169,169,171,169],24,[169,305,173,306,171,169,169,171,169],7,550,[],[97],[310],"Intel i7-8700 3.2 GHz + Nvidia GeForce GTX1080 Ti 11 GB",[],[313],"Runtime breakdown of OverlapNet",{"slug":315,"group":316,"sourceId":5,"sourceLabel":6,"table":317,"selfRows":248,"metrics":318,"seqs":321,"entrants":324,"cells":335,"outcomes":354,"locators":356,"hardware":357,"wordings":358,"notes":359},"overlapnet2020-table-ii","overlapnet2020:Table II","Table II",[319,320],{"label":214,"unit":143,"statistic":39,"alignment":38},{"label":216,"unit":143,"statistic":39,"alignment":38},[322,323],{"dataset":84,"sequence":219,"environment":156},{"dataset":89,"sequence":219,"environment":221},[325,327,330,333,334],{"name":326,"methodId":95,"linkable":75,"proposed":75,"self":75},"Histogram",{"name":328,"methodId":329,"linkable":160,"proposed":75,"self":75},"M2DP","m2dp2016",{"name":331,"methodId":332,"linkable":160,"proposed":75,"self":75},"SuMa","suma2018",{"name":230,"methodId":5,"linkable":160,"proposed":160,"self":160},{"name":232,"methodId":5,"linkable":160,"proposed":160,"self":160},[336,338,339,340,341,342,343,344,345,346,347,348,349,350,352,353],[169,169,169,337,171,169,171,171,169],0.83,[169,173,169,337,171,169,171,171,169],[173,169,169,337,171,169,171,171,169],[173,173,169,174,171,169,171,171,169],[176,169,169,95,169,169,171,171,169],[176,173,169,183,171,169,171,171,169],[179,169,169,174,171,169,171,171,169],[179,173,169,196,171,169,171,171,169],[169,169,173,250,171,169,171,171,169],[169,173,173,337,171,169,171,171,169],[173,169,173,250,171,169,171,171,169],[173,173,173,183,171,169,171,171,169],[176,169,173,95,169,169,171,171,169],[176,173,173,351,171,169,171,171,169],0.33,[248,169,173,183,171,169,171,171,169],[248,173,173,250,171,169,171,171,169],[355],"not_applicable (SuMa yields a single precision-recall point)",[317],[],[],[360],"Loop closure detection vs state of the art; best candidate per query, 100 latest scans excluded, overlap threshold 30%; KITTI uses all cues, Ford uses geometry only",[362,367],{"group":363,"slug":364,"sourceLabel":6,"table":365,"selfRows":179,"datasets":366},"overlapnet2020:Table IV","overlapnet2020-table-iv","Table IV",[84],{"group":368,"slug":369,"sourceLabel":6,"table":370,"selfRows":173,"datasets":371},"overlapnet2020:Text Sec.III-B","overlapnet2020-text-sec-iii-b","Text Sec.III-B",[39],1790510654847]