BEGIN:VCALENDAR
VERSION:2.0
PRODID:-//pretalx//talks.osgeo.org//foss4g-asia-2023//speaker//JZ8MEV
BEGIN:VTIMEZONE
TZID:KST
BEGIN:STANDARD
DTSTART:20000101T000000
RRULE:FREQ=YEARLY;BYMONTH=1
TZNAME:KST
TZOFFSETFROM:+0900
TZOFFSETTO:+0900
END:STANDARD
END:VTIMEZONE
BEGIN:VEVENT
UID:pretalx-foss4g-asia-2023-GGG8B3@talks.osgeo.org
DTSTART;TZID=KST:20231130T135000
DTEND;TZID=KST:20231130T141000
DESCRIPTION:The proliferation of high-resolution satellite imagery has dram
 atically transformed urban planning\, disaster response\, and environmenta
 l monitoring domains. A persistent challenge\, however\, is the precise ex
 traction and delineation of building footprints\, as traditional methods o
 ften yield results with irregular shapes and misalignments. To address the
 se limitations\, we present CornerRegNet\, a pioneering deep neural networ
 k designed to predict oriented building corners and subsequently regulariz
 e building footprints derived from satellite imagery.\nCentral to CornerRe
 gNet's design is its unique ability to harness oriented building corners a
 s primary anchoring points. This approach significantly streamlines and re
 fines the shape of the extracted building footprints. Using advanced multi
 -modal autoencoders\, CornerRegNet adeptly embeds a blend of corner featur
 es\, their inherent orientations\, and overarching semantic context. This 
 intricate fusion of features allows the network to capture micro-level det
 ails and macro-level building structural context.\nEnhancing the accuracy 
 of corner prediction\, the model incorporates active rotating filters. The
 se filters\, combined with uniquely devised equivariant convolutions\, ens
 ure that the predicted corners are not only spatially correct but also mai
 ntain their intended orientation concerning the building's overall layout.
 \nA pivotal component of CornerRegNet is the integrated Graph Convolution 
 Network (GCN). This GCN is instrumental in iteratively refining the positi
 ons of corners from the preliminary footprints. Further\, the in-built sha
 pe regularizer within the GCN aids in achieving architecturally consistent
  and coherent footprints\, effectively minimizing typical extraction anoma
 lies.\nIn comparative evaluations\, CornerRegNet stands out\, consistently
  outperforming contemporary models. The network's resilience in managing a
 nd regularizing building shapes is a distinguishing attribute\, even under
  challenging scenarios marked by occlusions.\nIn conclusion\, CornerRegNet
  represents a unique advancement in building footprint extraction from sat
 ellite imagery. Its unique combination of corner orientation prediction an
 d shape regularization techniques sets new standards for accuracy\, effici
 ency\, and architectural interpretability.
DTSTAMP:20260609T141814Z
LOCATION:Seoul Archive
SUMMARY:CornerRegNet: A Deep Learning Approach for Oriented Corner-Predicti
 on and Shape Regularization of Building Footprints from Satellite Imagery 
 - Gunho Sohn
URL:https://talks.osgeo.org/foss4g-asia-2023/talk/GGG8B3/
END:VEVENT
BEGIN:VEVENT
UID:pretalx-foss4g-asia-2023-U8VLVS@talks.osgeo.org
DTSTART;TZID=KST:20231201T135000
DTEND;TZID=KST:20231201T140500
DESCRIPTION:This research delves into the evolving interactions between sid
 ewalk delivery robots and pedestrians within urban environments\, focusing
  on the dual objectives of pedestrian safety and effective robot navigatio
 n in shared spaces. A comprehensive 3D digital twin model\, replicating re
 al-world urban settings\, was designed to simulate the multifaceted engage
 ments between humans\, robots\, and the urban fabric. Incorporated within 
 this simulation was the Pedestrian Aware Model (PAM)\, a multi-agent syste
 m\, employed to mimic both robot and human movements.\n\nEmploying an agen
 t-based modeling approach\, a series of scenarios involving pedestrians\, 
 wheelchair users\, and robots coexisting in sidewalk spaces were dissected
 . The salient revelation from this study is the non-contributory role of r
 obots to sidewalk congestion. Through programmed safety buffer zones\, rob
 ots not only ensure pedestrian safety but also facilitate streamlined navi
 gation\, hinting at a feasible harmonious integration.\n\nDespite this\, t
 he study unearthed certain challenges. Robots were predominantly implicate
 d in collisions\, whereas pedestrians often infringed upon set distance th
 resholds\, emphasizing the imperative for enhanced strategic measures to a
 lleviate these risks. However\, the overarching inference remains optimist
 ic: with judicious design and continuous research\, robots have the potent
 ial to integrate seamlessly with pedestrians\, enriching the urban milieu.
  The simulation model posited in this study stands as a pivotal resource f
 or urban planners and policymakers\, guiding them in formulating strategie
 s and policies for smooth robot-human cohabitation.\n\nQuantitative analys
 es further cement the significance of this research\, underscoring nuances
  like the importance of ample safety buffers around robots to minimize col
 lisions and enhance sidewalk traffic fluidity. In essence\, this research 
 pioneers in illuminating the pathways for optimized robot-human coexistenc
 e in bustling urban settings.
DTSTAMP:20260609T141814Z
LOCATION:Vium Hall
SUMMARY:Robot-Human Dynamics on Urban Sidewalks: Simulating Safe Coexistenc
 e through GIS-Based Digital Twinning - Gunho Sohn
URL:https://talks.osgeo.org/foss4g-asia-2023/talk/U8VLVS/
END:VEVENT
END:VCALENDAR