Visual Surveillance for Public Transport

1
Visual Surveillance for Public Transport

• Sergio A. Velastin
• Digital Imaging Research Centre
• Kingston University
• sergio.velastin_at_kingston.ac.uk

2
Drivers

• Reduce anti-social and criminal behaviour
• Antisocial behaviour estimated to cost taxpayers
  3.3 billion a year.
• 66,000 reports of nuisance/louts per day.
• 8,000 incidents graffiti/vandalism in one day
  (600m annual damage).
• Wandsworth spends 625K/year on tagging.
• 44 women feel unsafe at bus stops at night.
• Glasgow Vandalism 1/5 of all crime.
• A bus company replaced 8,000 windows and 18,000
  seats in a year 128 passengers and 34 staff
  injured by vandals.
• Improve mobility and shift usage into public
  transport (improve environment, reduce road
  deaths, ).
• Contribute to social inclusion (by protecting
  more vulnerable sectors of the population).

3
Visual Surveillance

• Observe (know what is going on), record (gather
  evidence), intervene (prevent, capable
  guardian).
• CCTV augments our visual ability (wide area).
• UK leads in CCTV in public places (estimated ¼ of
  world total), notorious prosecution successes.
• Balance right of privacy with right to be safe.

4
Conventional Surveillance

• Human beings are very good at situation
  assessment (from visual and non-visual clues).
• Limited attention span ( 40 mins).
• Events are rare Boredom, Too many cameras
  (201).
• Impossible to cover every corner. Crime may be
  displaced to CCTV-poor areas.
• Tendency to use CCTV as reactive tool.
• Once public are aware of limitations, they have
  less confidence.

5
Technology not sufficient

• Need an integrated approach
• Technical means to detect that something of
  interest is occurring, telecommunications
  infrastructure.
• Organisational human-led procedures to make use
  of the technology.
• An ethical framework (privacy vs security).
• Behavioural studies context-driven, multiple
  information sources.

6
Examples (Ipsotek)
Intrusion
Package
Loitering
Mixed
Stationary Object
Stationary Person
7
Progress so far

• Can pick up basic patterns (abandoned package,
  intrusion, stationary people, loitering).
• Possible to track people within a camera and
  between cameras (when uncluttered).
• Face recognition (when correct view point).
• Some posture/gesture recognition (e.g. domotics
  (home) environment).
• Learn behaviour patterns from examples (e.g. car
  parks).

8
Expectations

• Form multi-disciplinary (users, science,
  citizens) group on Crime in Public Transport
• Proposals
• Think Crime
• CEC

9
Needs doing Possible Collaboration

• Infer (criminal/anti-social) behaviour from
  posture/gesture in realistic conditions.
• Localisation and tracking in cluttered scenes
  (and for multiple cameras).
• Reliable unattended operation (24/7), especially
  for varying environmental conditions (day vs
  night, rain, insects!, )
• Constrained environments (e.g. in buses, in
  trains).
• Exploit non-visual context.
• Large systems (hundreds of cameras).
• Ethical safeguards (possibly automatic?)