1. VISPS Overview Capture visual data from two cameras Find laser Triangulate to find X, Y, Z coordinates of laser Output coordinates to RS-232 serial output RS-232 Out

2. Talk Overview VISPS goals and requirements – Cephas Design (Theory + Architecture) – Steve Design (Details + Communication) – Cephas Design analysis – Kevin Issues and testing – Kevin Schedule – Cephas Conclusion – Cephas

3. GOAL Produce X, Y, Z coordinates of laser marker relative to known reference frame Requirements, use: 630-680nm laser pointer 2 RC-2 black and white cameras XSV300 Development Board PIC Microcontroller VISPS Purpose

4. High Level Design Goals Produce X, Y, Z from visual data Simple interface Easy to use General, Accurate Design headroom

5. Requirements Table 3 – VISPS operating parameter ParameterValue Coordinate Accuracy@ 63 inches the coordinates will be within 0.1 inches of its true location. Data OutputRS-232 Protocol (19.2 Kbps / 8 data bits / 1 stop bit / no parity) VoltageRequires 9 and 12 V supply Maximum Operating Distance 60 feet Laser Flare Wavelength630-680 nm Laser Flare Max Output< 1 mW

6. Constraints From Requirements Code size on PIC  C < 20kB Enough pins on XSV300 for two cameras input and communication with PIC  C

7. How Are Design Goals Achieved? Mathematics and theory Logical implementation and architecture Implementation details and subsystem communication

8. Triangulation Problem Summary Side View Top View

9. Visual Data Left CameraRight Camera Pixel coordinates are available.

10. Camera Calibration

11. Pixel Distances h is a constant 340 pixels at any distance

12. Angle Derivation From Pixels

13. Triangulation By Pixels

14. Math And Theory Summary Y, and Z found Similar to X Found X, Y, and Z using in terms of pixel coordinates No trigonometry or complex math No multiplication of error

15. Design Implementation Goals (Output points at 30fps, Preserve design headroom) Black Box design outline Component design specifics and I/O details

16. PIC µC Compute angles from pixel data Compute X, Y, Z from angles Output coordinates using RS-232 protocol XSV300 Locate Marker in pixel data RoboCam RC-2 Deliver pixel data Generic Laser Pointer Project marker RS-232 Coordinate output Generic Red Filters Remove non-laser light VISPS

17. System Pin-outs

18. First saturated point Last saturated point Average of coordinates Finding Center Of Marker

19. XSV300-PIC Protocol NEW READY BUSY C1hiC1loC2hiC2loRhiRlo DATA

20. Design Analysis Goals met Requirements satisfied Error analysis Lab results

21. Goals Met The design will correctly produce X, Y, and Z coordinates. We managed to avoid solutions that would have made using VISPS overly cumbersome or complicated. VISPS can output points at a full 30 points per second, the same rate as the camera Significant upgrades to VISPS accuracy from higher resolution cameras would not require a major design change

22. Requirements Satisfied Reworking our math enabled us to fit all the needed code on the PIC  C With current cameras, VISPS is accurate to within 0.1 inches at 60 inches Moving to XSV300 from XS40 gave us the extra I/O pins we needed

23. Error Sources and Analysis Miscalibration of cameras Camera alignment Camera leveling Camera resolution Output resolution

24. Miscalibration Of Cameras All our mathematics are based on initial measurements at Z = 160cm Introduction of systematic error in Z if h does not really equal 340 pixels Spend extra time taking careful measurements Integer value for h is encouraging Do sanity checks against manual measurements

25. Camera Alignment All our mathematics are based on the assumption that the two cameras are perfectly parallel Introduction of systematic error if cameras are not parallel Measure D at a distance check

26. Camera Leveling To know the reference frame for the X, Y, Z values produced, camera assembly must be perfectly leveled Use bubble levels mounted on camera assembly Do sanity checks against manual measurements

27. Pixel size increases with distance Error is equal to ½ pixel width as a function of Z This error is inherent in VISPS and cannot be reduced except by increasing camera resolution Not bad. Even at 340ft, error only = 5.5 inches! Camera Resolution

28. Output Resolution Output is in tenths of inches Not a factor after 63 inches This error is inherent in VISPS Error can be reduced by reducing the max output distance

29. Lab Results Cameras filtered with red filter produce good, noise free data Able to find laser marker quickly, accurately, and consistently PIC  C output shows that math is correct Unable to test system accuracy until parts are integrated

30. Problem Areas and Testing Resolved issues Remaining issues Reviewers comments Testing

31. Resolved Issues Noise filtering Interfering light sources Complex math and PIC  C memory constraints

32. Remaining Issues Porting FPGA design to XSV300 System accuracy

33. Reviewers Comments ProblemTheir SuggestionWhat We Did Single Pixel Noise In Camera Data 9 or 3 pixel median filter.2 pixel neighbor intensity check. Noise is not saturated. Interfering Light Sources Strobe laser using spinning wheel and take difference. Red Filter to block non- laser light. Jittery Or Unstable Coordinates Average laser location over 4 frames for anti-jitter. Nothing. We want the jitter. Limited Memory of XS40 Move FPGA design to XSV300. Moved FPGA design to XSV300 Floating-Point Math Bloat Use LUT.Reworked math to avoid trigonometry.

34. Testing Do lots of sanity checks against manual measurements Test validity of math theories Visually inspect VGA output of laser finding module and camera perspective Logic analyzer to check FPGA output Extra PIC  C test environment for PIC  C View serial output on Microsoft Hyperterminal Output using printf() to test X, Y, Z production independently of coordinate output protocol

35. Current Status And Roadmap Completed: Find laser in visual data Module to talk to PIC  C PIC  C math PIC  C, FPGA and RS-232 I/O Remaining Port FPGA design to XSV300 Integrate components and test Win32 App for demo Port FPGA Design to XSV300. Integrate Components, test, and debug. More debugging, test, demo. Week 8 Week 9 Week 10

36. Conclusion And Summary Our design will work Design goals are met