Issues in the Design of an AutoPTZ Surveillance System

1
Issues in the Design of an Auto-PTZ Surveillance
System

• Jison Hsu
• Intelligent Video Surveillance
• Penpower Technology

2
Contents

• Review of Manual Control of PTZ Cams
• Current Design of Auto PTZ Cams
• Scene Analysis and Zoom In/Out Switch
• Major Makers
• On-going Project at Penpower
• Collaboration with NCTU/NTU

3
Close-up Snapshot by PTZ Cam
4
Manual Control

• Zoom out to scan over the scene
• Zoom in to obtain a close-up view of the
  suspected
• Superb-smart man-machine feedback loop

5
Our Advantage

• 3D Scene Perception and 3D Object Perception
• Object detection and recognition
• Trajectory tracking and prediction,
• Beyond v.s. under control
• Lock on and recover
• Quick Adaptation to Hardware Limits
• Fast close-up snapshot on the spot for those
  moving fast, or zoom out for trajectory and
  speed
• Zoom in for close-up video of those moving slow
• Time-delay is taken into account.

6
Two Types

• PTZ cam with a separated scene analytic cam
• Scene analytic cam detects and classifies objects
  moving in the scene
• PTZ cam zooms in and tracks down each object with
  a preset schedule
• Close-up view together with scene information.
• Standalone PTZ cam
• Zoom out to overwatch the scene, and zoom in to
  track down an intruding object
• Easy installation and an ideal operation mode.

7
Doable for Scene Analytics
8
Challenging Cases
9
Major Player ObjectVideo

• ObjectVideo VEWA intelligence-embedded
  mini-computer for real-time surveillance
• ObjectVideo OnBoard A DSP chipset for 3rd
  parties to embed OVs intelligence into their
  products
• ObjectVideo Forensic A off-line analysis
  package that can adjust the laws in VEW so that
  VEW can be tailor made for application-oriented
  uses
• Buyers DHS, border port controls, state and
  local governments, oil chemical facilities.

10
ObjectVideo System Configuration
11
Major Player ioImage

• Also named Safeguards Video in the US
• TRK Setup and Monitor similar to ObjectVideos
  VEW, for real-time capture of moving objects
• TRK Units similar to OVs On-Board,
  intelligence-embedded solution for 3rd parties
• Focus on integration with speed dome camera to
  capture close-up views of the moving object.
• Buyers Israel parliament, US state department,
  overlapped with OVs territories, and seems a bit
  behind in the US market competition.

12
System Configuration

• Specialized in moving object detection and
  tracking using PTZ cams.

13
Major Player 3VR

• Mostly installed at (indoor) exits, entrances,
  passages
• In addition to motion-sensitive analytics, also
  powered by face detection module for quick
  capture of human faces in the scene
• Standalone and client-server business solutions

14
Major Player Cernium

• Alert monitor keeps displaying important events,
  and call-up monitor offers instant replay and
  close-up check using a manual speed dome.
• Video servers and system server work together to
  push captured behavior events to the alert
  console for event display.

15
A Sample System -- TrackIN
16
TrackIN An Overview
17
A Screenshot of Playback
18
Collaboration with NCTU/NTU

• Scene and Scenario Database,
• Benchmark the performance of frontal object
  detection, tracking, and classification.
• Auto PTZ Control and Performance Evaluation
  Protocol,
• Optimal (or sub-optimal?) way to work with
  reference cams?
• Ways to benchmark the performance of an auto PTZ
  cam or system?

19
Scene and Scenario Database

• Cameras high sensitivity (betn 0.05 and 1 lux),
  infrared, WDR, HR.
• Illumination change Cont. change v.s. sudden
  change, indoor and outdoor, partial and overall
• Distance and range (view tilted from horizon
  090o, object sized from tens of pixels to 90 of
  window. )
• Scenarios
• Vehicles and pedestrians sizes, directions,
  speeds, interactions w/ vehs and peds.
• Noisy v.s. steady backgrounds, such as vehicles
  passing at a distance.

20
Auto PTZ Control
u min EobpP(P,Sz,DT1,DT2,)-P
PTZ Cam
Viewer
Scene Analytics

• Scene analytics offers info of frontal objects,
  incl. positions, speeds, sizes, and classes.
• Considering time delays caused by FPS(contents),
  scheduling, switching, focusing, stabilizing,
• Goal min P(P, Sz, DT1, DT2,) - P

21
Performance Evaluation Protocol

• Experimental setup to benchmark auto PTZ cams
• Controllable parameters
• Movements and timing
• Occlusion and recovery,
• Speeds and directions,
• Changes of poses,
• Crossing,
• Gathering and parting
• Illumination changes
• Continuous v.s. abrupt
• Partial v.s. overall.

22
Questions Welcomed!!