[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"doc-seo-512807-105":3,"detail-sidebar-cat-0-en-105":81,"doc-detail-512807-en":130},{"code":4,"msg":5,"data":6},0,"ok",{"site_id":7,"language":8,"slug":9,"title":10,"keywords":11,"description":12,"schema_data":13,"social_meta":74,"head_meta":76,"extra_data":78,"updated_unix":80},105,"en","dnd-dense-depth-estimation-in-crowded-dynamic-indoor-scenes","DnD - Dense Depth Estimation in Crowded Dynamic Indoor Scenes","","A novel approach estimates depth from a monocular camera moving through complex, crowded indoor environments such as department stores or metro stations. The method predicts absolute-scale dense depth maps for a full scene that includes a static background and multiple moving people. Training avoids the need for dense ground-truth depths from sensing devices by using RGB images together with sparse depth maps produced from traditional 3D reconstruction. Two constraints address non-rigid human motion without explicit tracking, and experiments show consistent gains on the NAVERLABS dataset.",{"@graph":14,"@context":73},[15,34,56],{"@type":16,"itemListElement":17},"BreadcrumbList",[18,23,27,31],{"item":19,"name":20,"@type":21,"position":22},"https://docshare.wps.com","Home","ListItem",1,{"item":24,"name":25,"@type":21,"position":26},"https://docshare.wps.com/document/","Document",2,{"item":28,"name":29,"@type":21,"position":30},"https://docshare.wps.com/document/research-report/","Research & Report",3,{"item":32,"name":10,"@type":21,"position":33},"https://docshare.wps.com/document/dnd-dense-depth-estimation-in-crowded-dynamic-indoor-scenes/512807/",4,{"url":32,"name":10,"@type":35,"image":36,"author":41,"headline":10,"publisher":44,"fileFormat":47,"inLanguage":8,"description":12,"dateModified":48,"datePublished":49,"encodingFormat":47,"isAccessibleForFree":50,"interactionStatistic":51},"DigitalDocument",{"url":37,"@type":38,"width":39,"height":40},"https://docshare.wps.com/thumbnails/dnd-dense-depth-estimation-in-crowded-dynamic-indoor-scenes/512807.png","ImageObject",300,407,{"name":42,"@type":43},"Jasmine","Person",{"url":19,"name":45,"@type":46},"DocShare","Organization","application/pdf","2026-11-16","2026-10-06",true,{"@type":52,"interactionType":53,"userInteractionCount":55},"InteractionCounter",{"@type":54},"ViewAction",8,{"@type":57,"mainEntity":58},"FAQPage",[59,65,69],{"name":60,"@type":61,"acceptedAnswer":62},"What problem does DnD address in crowded indoor depth estimation?","Question",{"text":63,"@type":64},"DnD targets the difficulty of obtaining dense, absolute-scale depth from a single moving monocular camera in complex indoor scenes with many non-rigidly moving people.","Answer",{"name":66,"@type":61,"acceptedAnswer":67},"What inputs does DnD use to predict dense depth maps?",{"text":68,"@type":64},"DnD takes an RGB image and a sparse depth map projected from a 3D model generated by traditional 3D reconstruction methods.",{"name":70,"@type":61,"acceptedAnswer":71},"How does DnD handle non-rigid motion without explicit person tracking?",{"text":72,"@type":64},"It introduces two constraints: a flow-guided shape constraint to refine human regions and a normal-guided scale constraint to learn absolute-scale depth guided by human ground contact depth values.","https://schema.org",{"og:url":32,"og:type":75,"og:title":10,"og:site_name":45,"og:description":12},"article",{"robots":77,"canonical":32},"index,follow",{"doc_id":79,"site_id":7},512807,1791389746,{"code":4,"msg":82,"data":83},"success",[84,88,92,96,101,106,111,114,119,122,126],{"id":22,"doc_module":4,"doc_module_name":25,"category_name":85,"show_sort_weight":86,"slug":87},"Story & Novel",90,"story-novel",{"id":26,"doc_module":4,"doc_module_name":25,"category_name":89,"show_sort_weight":90,"slug":91},"Literature",80,"literature",{"id":33,"doc_module":4,"doc_module_name":25,"category_name":93,"show_sort_weight":94,"slug":95},"Exam",70,"exam",{"id":97,"doc_module":4,"doc_module_name":25,"category_name":98,"show_sort_weight":99,"slug":100},5,"Comic",60,"comic",{"id":102,"doc_module":4,"doc_module_name":25,"category_name":103,"show_sort_weight":104,"slug":105},6,"Technology",50,"technology",{"id":107,"doc_module":4,"doc_module_name":25,"category_name":108,"show_sort_weight":109,"slug":110},7,"Healthcare",40,"healthcare",{"id":55,"doc_module":4,"doc_module_name":25,"category_name":29,"show_sort_weight":112,"slug":113},30,"research-report",{"id":115,"doc_module":4,"doc_module_name":25,"category_name":116,"show_sort_weight":117,"slug":118},9,"Religion & Spirituality",20,"religion-spirituality",{"id":117,"doc_module":4,"doc_module_name":25,"category_name":120,"show_sort_weight":117,"slug":121},"World Cup","world-cup",{"id":123,"doc_module":4,"doc_module_name":25,"category_name":124,"show_sort_weight":123,"slug":125},10,"Lifestyle","lifestyle",{"id":127,"doc_module":4,"doc_module_name":25,"category_name":128,"show_sort_weight":97,"slug":129},19,"General","general",{"code":4,"msg":82,"data":131},{"doc_id":79,"user_id":132,"nickname":42,"user_avatar":133,"doc_module":4,"category_id":55,"category_name":29,"doc_title":10,"doc_description":12,"doc_content":134,"file_id":135,"file_url":136,"file_type":137,"file_size":138,"view_count":55,"is_deleted":4,"is_public":22,"is_downloadable":22,"audit_status":22,"page_count":139,"language":140,"language_code":8,"site_id":7,"html_lang":8,"table_of_contents":141,"faqs":142,"seo_title":143,"seo_description":12,"update_tm":144,"read_time":145,"preview_page_count":33,"preview_urls":146},2336478487870,"https://ap-avatar.wpscdn.com/davatar_085a072bc5b1113ac321206ff7593b45","DnD: Dense Depth Estimation in Crowded Dynamic Indoor Scenes  \nDongki Jung* 1 Jaehoon Choi* 1,2 Yonghan Lee 1 Deokhwa Kim 1 Changick Kim3  \nDinesh Manocha2 Donghwan Lee 1  \n1NAVER LABS 2University of Maryland 3 KAIST  \nFigure 1: With traditional 3D reconstruction methods [45, 46], we can obtain a 3D model on complex and crowded indoor environments. Our novel approach, DnD, takes both an RGB image and a sparse depth map projected from this 3D model as input. Our method predicts dense and absolute-scale depth maps of single view dynamic scenes.  \nAbstract  \nWe present a novel approach for estimating depth from a monocular camera as it moves through complex and crowded indoor environments, e.g., a department store or a metro station. Our approach predicts absolute scale depth maps over the entire scene consisting of a static background and multiple moving people, by training on dynamic scenes. Since it is difficult to collect dense depth maps from crowded indoor environments, we design our training framework without requiring depths produced from depth sensing devices. Our network leverages RGB images and sparse depth maps generated from traditional 3D reconstruction methods to estimate dense depth maps. We use two constraints to handle depth for non-rigidly moving people without tracking their motion explicitly. We demonstrate that our approach offers consistent improvements over recent depth estimation methods on the NAVERLABS dataset, which includes complex and crowded scenes.  \n1. Introduction  \nThere is considerable interest in using robotics and aug-  \n* These two authors contributed equally. [Correspondence to dongki.jung@naverlabs.com](Correspondence to dongki.jung@naverlabs.com), [kevchoi@umd.edu](kevchoi@umd.edu)  \nmented reality technologies in crowded real-world spaces corresponding to malls, airports, or public places. In order to perform safe navigation or combine real and virtual worlds, robots [29, 28, 44] or mobile devices [50, 35] need a 3D geometric representation of large-scale indoor environments. While there is considerable progress in terms of capturing depth using LiDARs or stereo cameras, existing devices still have their own limitations. For example, 3DLiDARs [8] tend to produce sparse depth maps for distant objects and may result in noisy point cloud maps due to the high level of occlusions caused by multiple moving people. Moreover, because of the high prices and large volumes of the depth sensors, there is a critical need to consider the case where only a single camera is available.  \nGiven a large number of video frames, traditional 3D reconstruction methods such as structure-from-motion (SfM) and multi-view stereo (MVS) [14, 49, 45, 46] can generatea 3D model. Recently, visual localization techniques [38] have been developed to allow mobile devices to obtain the location and camera pose from which the image is taken. Given the 3D model and visual position, however, mobile devices moving through crowded indoor environments are only able to capture sparse and highly noisy depth maps. This is because traditional reconstruction methods are based on both a static scene assumption (that static areas of the  \nscenes can be observed from two different viewpoints) and correct feature matching in two or more images. However, the moving pedestrians in crowded indoor environments tend to violate the static scenes assumption. Moreover, traditional 3D reconstruction methods may fail to perform the correct matching on non-textured (e.g., walls), specular, and reflective regions (e.g., glass) of the scene. Consequently, these problems lead to a 3D model on complex and crowded indoor environments.  \nMain Results: To address these limitations, we explore new methods that can utilize the 3D model generated using traditional 3D reconstruction methods [45, 46] into learning-based depth estimation algorithms for dynamic scenes. Given the 3D model, our method can be used for general applications because it can compute den","cbCaimV1pqrDonW9","https://ap.wps.com/l/cbCaimV1pqrDonW9","pdf",2187039,11,"English","# Abstract\n# 1. Introduction\n## Motivation and problem with existing depth sensors\n## Limits of traditional 3D reconstruction in crowds\n## Main idea and contributions of DnD","[{\"question\":\"What problem does DnD address in crowded indoor depth estimation?\",\"answer\":\"DnD targets the difficulty of obtaining dense, absolute-scale depth from a single moving monocular camera in complex indoor scenes with many non-rigidly moving people.\"},{\"question\":\"What inputs does DnD use to predict dense depth maps?\",\"answer\":\"DnD takes an RGB image and a sparse depth map projected from a 3D model generated by traditional 3D reconstruction methods.\"},{\"question\":\"How does DnD handle non-rigid motion without explicit person tracking?\",\"answer\":\"It introduces two constraints: a flow-guided shape constraint to refine human regions and a normal-guided scale constraint to learn absolute-scale depth guided by human ground contact depth values.\"}]","DnD - Dense Depth Estimation in Crowded Dynamic Indoor Scenes | PDF",1791276142,28,[147,151,155,159],{"page_no":22,"url":148,"width":149,"height":150},"https://res-academy.cache.wpscdn.com/upload/kdocshare/2026-11-17/0b5127c1d46ece75a7f33f177d7a318a.webp",983,1272,{"page_no":26,"url":152,"width":153,"height":154},"https://res-academy.cache.wpscdn.com/upload/kdocshare/2026-11-17/943be041e399113ab8d914ca5badf1b0.webp",982,1270,{"page_no":30,"url":156,"width":157,"height":158},"https://res-academy.cache.wpscdn.com/upload/kdocshare/2026-11-17/7ff6284bf82f2846a8c4626de62c50fc.webp",981,1269,{"page_no":33,"url":160,"width":161,"height":162},"https://res-academy.cache.wpscdn.com/upload/kdocshare/2026-11-17/e8b9547e5845fa6689bd37bc0011085f.webp",980,1268]