Vision-Guided Robot Position Control

1
Vision-Guided Robot Position Control

• SKYNET

Tony Baumgartner Brock Shepard Jeff Clements
Norm Pond Nicholas Vidovich
Advisors Dr. Juliet Hurtig Dr. J.D. Yoder
November 12, 2003
2
Problem Identification

• Goal Develop a vision guided robot positioning
  control system
• Current systems are limited to Teach-Repeat style
• On completion the robot will automatically
  perform tasks specified by the software

3
Budget/Purchasing

• A budget of approximately 20,000 has been
  provided by Ohio Northern University
• Robot
• Gripper
• Desktop computer
• Two CCD (Charge-Coupled Device ) cameras

4
Background Information
SCARA Selectively Compliant Assembly Robot Arm
Articulated
5
Vision Systems

• Eye-On-Hand Configuration
• External Stereo Camera Configuration

6
Prototype

• Final prototype will include
• CCD camera(s)
• Desktop computer
• Robot Control Unit
• Robot
• Gripper

7
Specifications

• Robots are measured according to two
  specifications
• Repeatability
• Absolute Accuracy

8
Object Recognition Algorithms

• Normalized Cross-Correlation
• Slower
• Less Sensitive To Light
• Shape-Based Matching
• Faster
• Similarity Measure
• The Sum of Absolute Differences

9
Difference Algorithm Visuals
10
Difference Algorithm Visuals
11
Design Deliverables

• Purchasing Target Date
• Robot 09/12/03 - 11/21/03
• Gripper 09/12/03 - 11/21/03
• Image Processing
• Camera pan/tilt/zoom 09/30/03 - 02/20/04
• CAD 11/03/03 - 01/09/04
• Difference Algorithm 11/04/03 - 12/19/03
• Object Recognition 01/09/04 - 02/02/04
• User Interface 02/02/04 - 02/20/04
• General
• Controller Integration 12/01/03 - 02/02/04
• Gripper Implementation 12/01/03 - 02/02/04
• Testing 03/08/04 - 04/01/04

12
Overall Block Diagram
13
Software Block Diagram
14
Robot Decision Matrix
0 Unacceptable 1 Acceptable 2 Average 3
Good 4 Excellent 5 Best Case
15
QUESTIONS
?
16
References

• lt1gt ABB Product Specification Sheet (2003)
• lt2gt Lin, C.T., Tsai D.M. (2002) Fast normalized
  cross correlation for defect detection. Machine
  Vision. Yuan-Ze University,1-5.
• lt3gt Phil Baratti Robot Precision (personal
  communication, November 4
• 2003)
• lt4gt Stegar, C.., 2001. Similarity measures for
  occlusion, clutter, and
• illumination invariant object recognition. In
  B. Radig and S. Florczyk(eds), Mustererkennung
  2001, Springer Munchen, pp. 145-154.
• lt5gt Stegar, C., Ulrich, M. 2002 Performance
  Evaluation of 2D Object
• Recognition Techniques. Technical Report
  Technische Universitat Munchen, 1-15.
• lt6gt Robots and Manufacturing Automation, pg.
  220-222.
• lt7gt http//www.prip.tuwien.ac.at/Research/RobotVis
  ion/vs.html

17
Software Flow Description

• Take Initial Picture
• Initialize cameras
• Capture image
• Store image to hard disk or in RAM (will depend
  on speed)
• Take Current Picture
• Start current image capture loop
• Capture image and store in memory

18
Software Flow Description

• Compare Current Picture With Initial
• Use difference algorithm to compare initial image
  and current image this yields a bitmap of the
  pixels that have changed based on a specified
  error factor
• Store difference image in memory

19
Software Flow Description

• Ready to Manipulate Object?
• Check difference image to see that an object has
  entered the area using Boolean to keep track
• If object is already present check to see if it
  has stopped moving
• If done moving, act on object, else return to
  Take Current Picture Block

20
Software Flow Description

• Identify and Find Object Coordinates
• Take pictures of scene using both cameras focused
  on object
• Compare the images received from the cameras
  using the Stereo Vision Algorithm to determine
  object type and relevant points relative to the
  position of the robot arm
• Send Coordinates Instructions to Robot
  Controller
• Communicate with robot controller sending
  incremental instructions based on object type and
  coordinates

21
Software Flow Description

• Use Stereo Vision Algorithm to make sure desired
  actions
• Take pictures of scene using both cameras focused
  on object
• Compare the images received from the cameras
  using the Stereo Vision Algorithm to determine
  whether the robot arm has moved correctly. After
  adjusting to mistakes by the robot, send more
  coordinates and instructions to robot controller
  until it has performed the correct task