[{"data":1,"prerenderedAt":300},["ShallowReactive",2],{"method-keller2013pointfusion":3},{"method":4,"reference":51,"equipment":74,"figures":107,"results":108},{"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":23,"limitations":27,"sensors":32,"platform":34,"estimator":39,"association":40,"timeModel":41,"deskew":42,"loopClosure":43,"globalOptimization":44,"mapRepresentation":45,"prior":44,"outputGeometry":46,"compute":47,"codeUrl":48,"codeLicense":49,"relatedVersions":50},"keller2013pointfusion","Keller et al., 2013","Point-based fusion","Real-Time 3D Reconstruction in Dynamic Scenes Using Point-Based Fusion",2013,"classic","C12","odometry_with_local_mapping","此系統全程只用一個扁平的點（surfel）清單表示場景，每點存位置、法向量、半徑、信心計數與時間戳，不建立體素或其他空間資料結構。每個影格先以三層階層式稠密 ICP 將深度圖對齊到由模型點渲染出的深度圖來估計 6DoF 位姿，再把模型點渲染成超取樣索引圖做投影式資料關聯；對應點依距影像中心遠近的高斯信心加權平均融合，信心累積到門檻 10 才由不穩定轉為穩定。系統另移除違反自由空間的點，並從 ICP 找不到對應的像素出發做區域成長，把移動物體整體標為動態並排除於位姿估計之外。","A flat, unstructured list of surfels (position, normal, radius, confidence, timestamp) is updated in the graphics pipeline: frame-to-model hierarchical ICP gives the pose, a supersampled index map provides projective association, confidence-weighted averaging fuses points, free-space violations are removed, and region growing from ICP outliers segments dynamic objects.","full_text_reviewed","peer_reviewed_published","background","原研究為室內物件與辦公室尺度 RGB-D、ToF 場景（Large Office 約 10 m x 6 m x 2.5 m），未涉及營建。",[20,21,22],"simulation","controlled_experiment","independent_reference",[24,25,26],"[\"Speed and memory efficiency from the point representation","robustness to dynamic scenes (abstract).\", \"Sim scene: mean model-point error 0.019 mm with ground-truth poses and 0.20 cm with ICP poses","ICP camera position error 0.87 cm and viewing-direction error 0.1 degrees on average (Sec. 7, Fig. 3).\", \"Large Office (about 10 m x 6 m x 2.5 m): 4.6 million model points stored in 110 MB of GPU memory, whereas a predefined 512 MB voxel grid would force voxels larger than 1 cm (Sec. 7).\", \"Tracking on Flowerpot is similar to KinectFusion against Vicon ground truth, with the largest difference about 1 cm (Fig. 5 caption).\", \"Dynamic segmentation keeps tracking in the Moving Person scene where earlier approaches fail, and model size converges after one turntable revolution (Sec. 7, Figs. 6 and 8).\"]",[28,29,30,31],"[\"Sensor drift is not tackled","drift in larger environments remains future work (Sec. 8).\", \"Opaque splats are rendered without blending or prefiltering, trading local surface quality for speed (Sec. 5).\", \"Only the last RGB sample is stored per point (Sec. 7).\", \"Only the synthetic Sim scene has geometric ground truth","the comparison with KinectFusion on Flowerpot and Teapot is visual apart from Vicon-based tracking (Sec. 7, Figs. 4-5).\", \"Hashing authors later state that point-based fusion quality is not on par with true volumetric methods (Niessner et al. 2013, Sec. 2)","this is a competing-method claim, not an independent evaluation.\"]",[33],"[\"RGB-D camera (Microsoft Kinect, near mode, 640x480 depth)\", \"time-of-flight camera (PMD CamBoard, 200x200, per-pixel amplitude used for confidence)\"]",[35,36,37,38],"[\"stationary Kinect with objects on a turntable (Flowerpot, Teapot","sequences recorded by Nguyen et al.)\", \"Kinect (near mode stated for most scenes) in the indoor Large Office (two rooms and connecting corridors, about 10 m x 6 m x 2.5 m), Moving Person and Ballgame scenes","camera carrier and motion are not described per scene\", \"PMD CamBoard ToF camera (PMD scene","carrier not stated)\", \"synthetic depth maps from a virtual camera rotating around the Sim scene\"]","frame-to-model dense hierarchical (pyramid) ICP against the rendered model point map, estimating a 6-DoF pose before fusion","projective data association by rendering the global point model as an index map","discrete poses","not_applicable (each depth map is registered with a single 6DoF camera pose; the paper does not discuss shutter type or intra-frame motion)","none (authors state sensor drift is not tackled; loop closure is future work, Sec. 8)","none","flat list of points\u002Fsurfels with position, normal, radius and confidence counter; unstable-to-stable status; no spatial data structure","fused point (surfel) model with position, normal, radius, confidence and timestamp, optionally the last RGB sample per point; visualised by opaque surface splatting; no mesh extraction","GPU (Intel i7 8-core CPU, NVIDIA GTX 680); 640x480 input (200x200 for the PMD scene); per-frame ICP about 11-22 ms, dynamic segmentation about 0.7-3.2 ms and fusion about 3-18 ms; processed rates 15 to 27 fps (Table 1)",null,"not_verified",[],{"id":5,"kind":52,"shortName":7,"title":8,"authors":53,"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":70},"method",[54,55,56,57,58,59],"Maik Keller","Damien Lefloch","Martin Lambers","Shahram Izadi","Tim Weyrich","Andreas Kolb","2013 International Conference on 3D Vision (3DV)","conference","IEEE","pp. 1-8","10.1109\u002F3dv.2013.9","http:\u002F\u002Freality.cs.ucl.ac.uk\u002Fprojects\u002Fkinect\u002Fkeller13realtime.pdf","2013-06","metadata_verified","principle reused: flat point\u002Fsurfel-based fusion with projective association for real-time dense reconstruction, cited as a basis of SuMa's frame-to-model registration.",[11],false,"corrected","author copy","Author PDF hosted at UCL (reality.cs.ucl.ac.uk\u002Fprojects\u002Fkinect\u002Fkeller13realtime.pdf, 8 pp.); IEEE 3DV 2013 version of record (IEEE Xplore document 6599048) not compared",[75,81,86,92,98,104],{"category":76,"model":77,"canonical":77,"role":78,"dataset":48,"specs":79,"locator":80},"rgbd","Microsoft Kinect","method input","near mode; 640x480 depth input; up to 307,200 points per frame; 30 fps input","Sec. 3, Sec. 7, Table 1",{"category":82,"model":83,"canonical":83,"role":78,"dataset":48,"specs":84,"locator":85},"other","PMD CamBoard","time-of-flight camera; 200x200 frames; per-pixel amplitude used in sample confidence; 27 fps input","Sec. 3, Sec. 7, Table 1, Fig. 10",{"category":82,"model":87,"canonical":87,"role":88,"dataset":89,"specs":90,"locator":91},"Vicon","reference or ground truth","Flowerpot and Teapot (Nguyen et al.)","motion capture ground truth for Kinect poses","Sec. 7, Fig. 5",{"category":93,"model":94,"canonical":94,"role":95,"dataset":89,"specs":96,"locator":97},"platform","turntable (model not reported)","dataset sensor","objects rotated in front of a stationary Kinect","Sec. 7",{"category":99,"model":100,"canonical":100,"role":101,"dataset":48,"specs":102,"locator":103},"compute","Intel i7 8-core CPU","compute for runtime","8-core","Table 1 caption",{"category":99,"model":105,"canonical":105,"role":101,"dataset":48,"specs":106,"locator":103},"NVidia GTX 680","GPU",[],{"totalRows":109,"groupCount":110,"groups":111,"others":299},40,2,[112,249],{"slug":113,"group":114,"sourceId":5,"sourceLabel":6,"table":115,"selfRows":116,"metrics":117,"seqs":133,"entrants":161,"cells":165,"outcomes":242,"locators":243,"hardware":244,"wordings":246,"notes":247},"keller2013pointfusion-table-1","keller2013pointfusion:Table 1","Table 1",35,[118,123,125,127,130],{"label":119,"unit":120,"statistic":121,"alignment":122},"Avg. timings [ms], ICP","ms","mean","not_reported",{"label":124,"unit":120,"statistic":121,"alignment":122},"Avg. timings [ms], Dyn-Seg.",{"label":126,"unit":120,"statistic":121,"alignment":122},"Avg. timings [ms], Fusion",{"label":128,"unit":129,"statistic":122,"alignment":122},"fps proc.","fps",{"label":131,"unit":132,"statistic":122,"alignment":122},"#model-points","points",[134,138,142,145,149,153,157],{"dataset":135,"sequence":136,"environment":137},"Sim (synthetic)","950\u002F950 frames","synthetic scene",{"dataset":139,"sequence":140,"environment":141},"Flowerpot (Nguyen et al.)","600\u002F480 frames","turntable object",{"dataset":143,"sequence":144,"environment":141},"Teapot (Nguyen et al.)","1000\u002F923 frames",{"dataset":146,"sequence":147,"environment":148},"Large Office","11892\u002F6704 frames","indoor office, two rooms and corridors (about 10 x 6 x 2.5 m)",{"dataset":150,"sequence":151,"environment":152},"Moving Person","912\u002F623 frames","indoor room with a moving person",{"dataset":154,"sequence":155,"environment":156},"Ballgame","1886\u002F1273 frames","indoor room, two people moving a ball across a table",{"dataset":158,"sequence":159,"environment":160},"PMD","4101\u002F4101 frames","indoor scene, ToF camera",[162],{"name":163,"methodId":5,"linkable":164,"proposed":164,"self":164},"point-based fusion",true,[166,170,173,175,178,181,183,185,187,189,191,193,195,197,199,201,203,205,207,209,211,213,215,217,219,221,224,226,228,230,232,235,237,239,240],[167,167,167,168,169,167,167,169,167],0,18.9,-1,[167,171,167,172,169,167,167,169,167],1,2.03,[167,110,167,174,169,167,167,169,167],11.5,[167,176,167,177,169,167,167,169,167],3,15,[167,179,167,180,169,167,169,169,167],4,467200,[167,167,171,182,169,167,167,169,167],15.87,[167,171,171,184,169,167,167,169,167],1.9,[167,110,171,186,169,167,167,169,167],6.89,[167,176,171,188,169,167,167,169,167],24,[167,179,171,190,169,167,169,169,167],496260,[167,167,110,192,169,167,167,169,167],15.2,[167,171,110,194,169,167,167,169,167],1.6,[167,110,110,196,169,167,167,169,167],5.56,[167,176,110,198,169,167,167,169,167],27,[167,179,110,200,169,167,169,169,167],191459,[167,167,176,202,169,167,167,169,167],21.75,[167,171,176,204,169,167,167,169,167],2.39,[167,110,176,206,169,167,167,169,167],13.9,[167,176,176,208,169,167,167,169,167],17,[167,179,176,210,169,167,169,169,167],4610800,[167,167,179,212,169,167,167,169,167],15.92,[167,171,179,214,169,167,167,169,167],3.23,[167,110,179,216,169,167,167,169,167],16.61,[167,176,179,218,169,167,167,169,167],20,[167,179,179,220,169,167,169,169,167],210500,[167,167,222,223,169,167,167,169,167],5,16.74,[167,171,222,225,169,167,167,169,167],3.15,[167,110,222,227,169,167,167,169,167],17.66,[167,176,222,229,169,167,167,169,167],21,[167,179,222,231,169,167,169,169,167],350940,[167,167,233,234,169,167,167,169,167],6,10.7,[167,171,233,236,169,167,167,169,167],0.73,[167,110,233,238,169,167,167,169,167],3.06,[167,176,233,198,169,167,167,169,167],[167,179,233,241,169,167,169,169,167],280050,[],[115],[245],"Intel i7 8-core CPU, NVIDIA GTX 680",[],[248],"Average per-frame timings of ICP, dynamic segmentation and fusion; frames input\u002Fprocessed and fps input\u002Fprocessed; input 640x480 except PMD 200x200. Input fps: Sim 15, PMD 27, others 30",{"slug":250,"group":251,"sourceId":5,"sourceLabel":6,"table":252,"selfRows":222,"metrics":253,"seqs":268,"entrants":273,"cells":279,"outcomes":290,"locators":291,"hardware":293,"wordings":295,"notes":296},"keller2013pointfusion-text-sec-7","keller2013pointfusion:Text Sec. 7","Text Sec. 7",[254,257,260,262,265],{"label":255,"unit":256,"statistic":121,"alignment":122},"mean position error of global model points (ground-truth poses)","mm",{"label":258,"unit":259,"statistic":121,"alignment":122},"mean position error of global model points (ICP poses)","cm",{"label":261,"unit":259,"statistic":121,"alignment":122},"mean position error of camera transformations from ICP",{"label":263,"unit":264,"statistic":121,"alignment":122},"mean viewing direction error of camera transformations from ICP","deg",{"label":266,"unit":267,"statistic":122,"alignment":122},"GPU memory for global model points","MB",[269,271],{"dataset":135,"sequence":270,"environment":137},"virtual camera rotating around the scene",{"dataset":146,"sequence":272,"environment":148},"full sequence",[274,276,278],{"name":275,"methodId":5,"linkable":164,"proposed":164,"self":164},"point-based fusion with ground-truth camera poses",{"name":277,"methodId":5,"linkable":164,"proposed":164,"self":164},"point-based fusion with ICP pose estimation",{"name":163,"methodId":5,"linkable":164,"proposed":164,"self":164},[280,282,284,286,288],[167,167,167,281,169,167,169,169,167],0.019,[171,171,167,283,169,167,169,169,167],0.2,[171,110,167,285,169,171,169,169,167],0.87,[171,176,167,287,169,171,169,169,167],0.1,[110,179,171,289,169,171,167,169,171],110,[],[292,97],"Sec. 7, Fig. 3",[294],"NVIDIA GTX 680",[],[297,298],"Synthetic Sim scene with ground-truth camera transformations and geometry; errors are means over model points or frames","GPU memory for the 4.6 million global model points of the Large Office scene (3 floats position, 2 normal, 1 radius, 1 byte confidence)",[],1790510663718]