[{"data":1,"prerenderedAt":211},["ShallowReactive",2],{"method-bouaziz2013sparseicp":3},{"method":4,"reference":40,"equipment":61,"figures":62,"results":63},{"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":28,"estimator":29,"association":30,"timeModel":31,"deskew":31,"loopClosure":32,"globalOptimization":32,"mapRepresentation":33,"prior":34,"outputGeometry":35,"compute":36,"codeUrl":37,"codeLicense":38,"relatedVersions":39},"bouaziz2013sparseicp","Bouaziz et al., 2013","Sparse ICP","Sparse Iterative Closest Point",2013,"classic","C02","registration_component","作者指出一般 ICP 依賴修剪或重新加權對應點的經驗法則來處理離群值與部分重疊，這些法則不穩定且難以調整。論文將配準目標改為對每組對應點的殘差向量施加 p 介於 0 與 1 之間的 ℓp 範數（群組稀疏），保留最近點搜尋步驟，並以交替方向乘子法（ADMM）與收縮運算子求解剛體轉換，點對點與線性化點對平面兩種版本皆可使用，實驗多採 p = 0.4。在虛擬掃描的貓頭鷹模型上，其配準 RMSE 為 4.8e-4，優於 ℓ1-ICP 的 1.6e-2 與使用距離門檻剔除的傳統 ICP。作者也指出 p 越小收斂越慢，而目標幾何含大量離群值或初始位置相距過遠時，最近點步驟仍會使結果落入錯誤的局部極小值。","Replaces heuristic correspondence pruning with sparsity-inducing norms in ICP to handle outliers and missing data.","full_text_reviewed","peer_reviewed_published","supplementary","not_reported（實驗對象為雕像與物件的虛擬掃描、合成雜訊掃描及消費級深度相機掃描，未涉及建物或工地）",[20,21],"simulation","controlled_experiment",[23],"[\"heuristic-free outlier handling with a single free parameter p (Sec. 8)\", \"on the virtually scanned owl model, RMSE 4.8e-4 versus 1.6e-2 for l1-ICP and 2.9e-2 to 4.1e-1 for thresholded least-squares ICP (Fig. 4)\", \"ADMM is stable and faster than iterative reweighting for point-to-plane, where reweighting is ill-conditioned (Sec. 7, Fig. 10)\", \"selects inliers automatically, unlike trimmed ICP which needs the inlier ratio (Sec. 7, Fig. 12)\", \"handles whole-in-part registration where Tukey and fair weights fail (Fig. 7)\"]",[25,26],"[\"the closest-point step is still affected by outliers, so performance degrades with many outliers in the target geometry (Sec. 7 Limitations, Fig. 9)\", \"local method: converges to a wrong minimum when source and target start far apart (Fig. 9)\", \"convergence slows as p decreases","p = 0.4 chosen empirically as a trade-off (Sec. 7, Fig. 11)\", \"evaluation is a limited set of illustrative experiments rather than an exhaustive comparison (Sec. 7)\"]",[],[],"l_p norm (p in [0, 1], p = 0.4 by default) of per-correspondence residual vectors (group sparsity) minimized with ADMM: a shrinkage step on auxiliary variables, a classical least-squares rigid fit, and a multiplier update; point-to-point and linearized point-to-plane versions (Sec. 4 to 7)","closest point on the target via an l2 kd-tree; unchanged by the l_p metric because |r|^p is monotone (Sec. 5.1)","not_applicable","none","3D scans \u002F geometric data sets","initial pose required","rigid transformation","not_reported (C++ implementation released; no runtime or hardware reported)","https:\u002F\u002Fgithub.com\u002Fopengp\u002Fsparseicp","MPL-2.0 (stated in README; no LICENSE file)",[],{"id":5,"kind":41,"shortName":7,"title":8,"authors":42,"year":9,"venue":46,"venueType":47,"publisher":48,"volumeIssuePages":49,"doi":50,"arxivId":51,"url":52,"firstPublicDate":53,"publicationStatus":16,"metadataStatus":54,"fulltextStatus":15,"era":10,"classicReason":55,"codeUrl":37,"cluster":11,"topics":56,"mdpi":57,"verification":58,"label":6,"fulltextRoute":59,"versionRead":60,"addedByCensus":57},"method",[43,44,45],"Sofien Bouaziz","Andrea Tagliasacchi","Mark Pauly","Computer Graphics Forum","journal","Wiley","32(5):113-123","10.1111\u002Fcgf.12178",null,"https:\u002F\u002Fapi.crossref.org\u002Fworks\u002F10.1111\u002Fcgf.12178","2013-08-19","metadata_verified","principle reused: sparsity-inducing norms as an alternative to heuristic outlier pruning in ICP.",[11],false,"corrected","author copy","publisher-layout PDF of Computer Graphics Forum 32(5) (SGP 2013) posted on co-author A. Tagliasacchi's lab website; carries the Eurographics and Blackwell copyright line; Wiley HTML not opened (HTTP 403 to curl)",[],[],{"totalRows":64,"groupCount":65,"groups":66,"others":210},13,2,[67,183],{"slug":68,"group":69,"sourceId":70,"sourceLabel":71,"table":72,"selfRows":73,"metrics":74,"seqs":81,"entrants":96,"cells":110,"outcomes":176,"locators":177,"hardware":178,"wordings":180,"notes":181},"zhou2016fgr-table-3","zhou2016fgr:Table 3","zhou2016fgr","Zhou et al., 2016","Table 3",12,[75,79],{"label":76,"unit":77,"statistic":78,"alignment":32},"running time per pairwise registration (s)","s","not_reported",{"label":76,"unit":77,"statistic":80,"alignment":32},"mean",[82,86,88,90,92,94],{"dataset":83,"sequence":84,"environment":85},"Synthetic range images","Bimba (9,416 points avg)","synthetic",{"dataset":83,"sequence":87,"environment":85},"Children (11,148 points avg)",{"dataset":83,"sequence":89,"environment":85},"Dragon (11,232 points avg)",{"dataset":83,"sequence":91,"environment":85},"Angel (12,072 points avg)",{"dataset":83,"sequence":93,"environment":85},"Bunny (13,357 points avg)",{"dataset":83,"sequence":95,"environment":85},"Average (11,445 points avg)",[97,101,104,106,108],{"name":98,"methodId":99,"linkable":100,"proposed":57,"self":57},"PCL ICP point-to-point","besl1992icp",true,{"name":102,"methodId":103,"linkable":100,"proposed":57,"self":57},"PCL ICP point-to-plane","chen1992pointtoplane",{"name":105,"methodId":5,"linkable":100,"proposed":57,"self":100},"Sparse ICP point-to-point [5]",{"name":107,"methodId":5,"linkable":100,"proposed":57,"self":100},"Sparse ICP point-to-plane [5]",{"name":109,"methodId":70,"linkable":100,"proposed":100,"self":57},"Our approach (FGR)",[111,115,118,120,123,126,128,130,132,134,136,138,140,142,144,146,148,150,152,154,156,158,160,162,164,166,169,171,172,174],[112,112,112,113,114,112,112,114,112],0,0.73,-1,[116,112,112,117,114,112,112,114,112],1,0.31,[65,112,112,119,114,112,112,114,112],3.1,[121,112,112,122,114,112,112,114,112],3,11.8,[124,112,112,125,114,112,112,114,112],4,0.13,[112,112,116,127,114,112,112,114,112],0.75,[116,112,116,129,114,112,112,114,112],0.46,[65,112,116,131,114,112,112,114,112],3.9,[121,112,116,133,114,112,112,114,112],15,[124,112,116,135,114,112,112,114,112],0.2,[112,112,65,137,114,112,112,114,112],0.99,[116,112,65,139,114,112,112,114,112],0.47,[65,112,65,141,114,112,112,114,112],3.6,[121,112,65,143,114,112,112,114,112],13.8,[124,112,65,145,114,112,112,114,112],0.23,[112,112,121,147,114,112,112,114,112],0.81,[116,112,121,149,114,112,112,114,112],1.01,[65,112,121,151,114,112,112,114,112],4.9,[121,112,121,153,114,112,112,114,112],18.5,[124,112,121,155,114,112,112,114,112],0.26,[112,112,124,157,114,112,112,114,112],2.1,[116,112,124,159,114,112,112,114,112],1.7,[65,112,124,161,114,112,112,114,112],9.2,[121,112,124,163,114,112,112,114,112],10.3,[124,112,124,165,114,112,112,114,112],0.28,[112,116,167,168,114,112,112,114,112],5,1.08,[116,116,167,170,114,112,112,114,112],0.79,[65,116,167,151,114,112,112,114,112],[121,116,167,173,114,112,112,114,112],13.9,[124,116,167,175,114,112,112,114,112],0.22,[],[72],[179],"Intel Core i7-5960X 3.00 GHz, single thread",[],[182],"Timing of local refinement methods and FGR on the same five synthetic models as Table 2 (same point counts; FGR column identical to Table 2); the caption does not state how per-model times were aggregated; single thread",{"slug":184,"group":185,"sourceId":5,"sourceLabel":6,"table":186,"selfRows":116,"metrics":187,"seqs":192,"entrants":197,"cells":200,"outcomes":203,"locators":204,"hardware":206,"wordings":207,"notes":208},"bouaziz2013sparseicp-fig-4","bouaziz2013sparseicp:Fig. 4","Fig. 4",[188],{"label":189,"unit":190,"statistic":191,"alignment":32},"e, RMSE w.r.t. ground-truth alignment, panel (f)","model units (not stated)","RMSE",[193],{"dataset":194,"sequence":195,"environment":196},"virtually scanned 'owl' statue model (lgg.epfl.ch\u002Fstatues)","owl","simulation (virtual scan)",[198],{"name":199,"methodId":5,"linkable":100,"proposed":100,"self":100},"lp-ICP, p = 0.4 (proposed)",[201],[112,112,112,202,114,112,114,114,112],0.00048,[],[205],"Fig. 4 and caption; Sec. 7",[],[],[209],"Virtually scanned 'owl' model registered to its ground truth; e = RMSE of registered point locations with respect to the ground-truth alignment; values are the numbers printed under each panel of Fig. 4 (not read from an axis); distance thresholds are percentages of the bounding-box diagonal",[],1790510665834]