Egocentric Navigation for Video Surveillance in 3D Virtual Environments
Gerwin de Haan
TU Delft
Josef Scheuer
TU Delft
TNO Science And Industry
Raymond de Vries
TNO Science And Industry
Frits H. Post
TU Delft
Figure 1: View transition sequence of tracking a person between camera A and adjacent camera B in an office hallway surveillance scenario
(see video). The guided 3D navigation enables simple first-person video observation while ensuring a good visual flow for spatial context. Our
method dynamically embeds and blends video canvases in a 3D VE which consists of characteristic model landmarks and perspective lines.
ABSTRACT
Current surveillance systems can display many individual video
streams within spatial context in a 2D map or 3D Virtual Envi-
ronment (VE). The aim of this is to overcome some problems in
traditional systems, e.g. to avoid intensive mental effort to main-
tain orientation and to ease tracking of motions between differ-
ent screens. However, such integrated environments introduce new
challenges in navigation and comprehensive viewing, caused by im-
perfect video alignment and complex 3D interaction. In this pa-
per, we propose a novel, first-person viewing and navigation inter-
face for integrated surveillance monitoring in a VE. It is currently
designed for egocentric tasks, such a tracking persons or vehicles
along several cameras. For these tasks, it aims to minimize the op-
erator’s 3D navigation effort while maximizing coherence between
video streams and spatial context. The user can easily navigate
between adjacent camera views and is guided along 3D guidance
paths. To achieve visual coherence, we use dynamic video embed-
ding: according to the viewer’s position, translucent 3D video can-
vases are smoothly transformed and blended in the simplified 3D
environment. The animated first-person view provides fluent visual
flow which facilitates easier maintenance of orientation and can aid
in spatial awareness. We discuss design considerations, the imple-
mentation of our proposed interface in our prototype surveillance
system and demonstrate its use and limitations in various surveil-
lance environments.
Keywords: video surveillance, virtual environments, navigation
Index Terms: H.5.1 [Information Systems]: Information Inter-
faces And Presentation—Multimedia Information Systems;
1 INTRODUCTION
Video surveillance control rooms have to deal with an ever-growing
amount and diversity of data from complex environments. Cur-
rently, operators observe the video streams directly on large matrix
display arrangements, combined with an interactive camera layout
plan, e.g. see Figure 2. With densely placed cameras in a com-
plex environment, it is hard to maintain orientation and coherence
Corresponding author, g.dehaan@tudelft.nl
Figure 2: Surveillance control room with a traditional video matrix display
arrangement. Operators select sets of individual videos from 2D maps.
between individual streams. Our main application is a surveillance
system in such a complex 3D environment, a newly-built soccer sta-
dium with over 200 cameras. Cognitive overload can easily occur
and it can become hard to perform tasks such as selecting regions
of interest in a camera image or tracking motions between different
cameras. Our current research concerns the design of user inter-
faces to enhance the support of surveillance tasks.
We observed that a major bottleneck in many surveillance tasks
lies in the difference between view reference frames [15] in differ-
ent camera views and the map view. Where operators can naturally
reason and act upon the first-person or egocentric views from a sin-
gle camera, they have to mentally “translate” their reasoning back
to the other camera views or the map, a third-person or exocentric
representation. This dual representation can hinder direct integra-
tion of video information with map data and requires two modes of
interaction . An example is to track a person which walks outside
of the current camera range to the left. The operator needs to select
a camera to the left-side of the current camera. The location of the
current camera has to be found on the exocentric map view and its
left neighbour selected. In an egocentric view, the operator would
naturally indicate the system to move the viewpoint to the left.
In this paper, we propose an egocentric or first-person 3D in-
terface for video surveillance monitoring in VEs. The goal of this
interface is to provide improved situational awareness and to serve