Post on 01-Jan-2016
description
Development of a Whitecap Measurement System
Student: Garry HigginsSupervisor: Dr. Edward JonesCo-Supervisor: Dr. Martin Glavin
Project Outline
• Development of a whitecap measurement system for Mace Head
• Core Components: – Image Capture– Image Processing– Data Handling
• Joint EE and ECI project• Why did I choose this project?
Why I Chose This Project
• Interesting Subject• Outdoors Aspect• Project covered a range of interesting topics
Hardware
• SLR Camera
• CCTV Cameras
CCTV Cameras
• 2 x Marshall 1/3” CCD Cameras• Geovision GV-600 Surveillance• 450 T.V. lines 340x240 resolution at 30fps• BNC Connectors• Connected to P4 2.4GHz PC with 504MB Ram
SLR Type Cameras
• Canon 350D – Sigma AP0DG lens 70-300mm • Nikon D70 – DX lens 18-70mm• Picture resolution of 3456x2304
Nikon D70
• Image exposure varies from picture to picture
Canon 350D
• Consistent exposure for each picture• Sigma zoom lens allowed for testing of
different zoom levels• Increase in zoom causes increase in effect of
tower shaking• Tested capturing area to north-east– Required zoom over 200mm focal length– No significant advantage
Software
• Matlab code
• Perl Scripts
• Mplayer
MATLAB
• Main percentage coverage algorithm– Get % white for each threshold level 0.01 -> 1.0– Get first differential and apply MATLAB “smooth”
function– Get second differential and apply “smooth”
function– Right -> Left : Desired threshold when it goes
above 0.01– Compiled to stand alone executable file
Pre-processing
• Horizon and surf-zone need to be cropped• Mplayer -vf crop=[w:h:x:y] –vo
jpeg:quality=100 –frames a –sstep b• Combination of Perl scripts and batch file to
parse avi files • Compiled MATLAB code called• Event| Year | Month | Day | Hour | Min(5s) |
Secs | Cam |.avi• Output results to txt file based on avi name
Algorithm Accuracy
• Compare results with those obtained by expert in the field on sample data
• Unsmoothed, single smoothed and double smoothed results compared
• Mean and coefficient of correlation between each version and original results calculated
Algorithm Accuracy
Original: Mean: 9.545170e-001Unsmoothed: Mean: 1.124245e+000Single Smooth: Mean: 1.543157e+000Double Smooth: Mean: 1.342003e+000Coefficients of Correlation:Original vs Unsmoothed: 6.855651e-001Original vs Single Smooth: 9.029801e-001Original vs Double Smooth: 9.387241e-001
CCTV Data
• Cameras recorded for month of March from 11:00am to 12:00pm
• Videos retrieved and percentage coverage for each day calculated
• Wind speed for month obtained from MET Eireann
Camera Correlation
• Mean and coefficient of correlation calculated for each camera– Cam01 mean = 1.714857e+000– Cam04 mean = 6.353763e-001– Coefficient of Correlation = 8.348656e-001
• Correlation of +0.83 low for data being compared
Reasons for differences
• Cam01: rain causing droplets of water on lens• Fog• Interference
Whitecap vs Wind Speed
• Daily average of wind speed and whitecap coverage calculated
• 8th -> 14th, 16th, 17th, 19th and 20th • Wind speeds parsed from MET Eireann info• Scatter plot for each camera plotted using
MATLAB “loglog” function• Power law relationship fitted in least square
relationship
Processing Efficiency
• Needs to process data in (pseudo) real-time• Perl includes a “Benchmark” module• Adapted video splitting script to include
benchmarking results• Run on sample of 50 videos• ~2.7x efficient
Avenues for Future Development
• CCTV System:– Remove interference– Cover cam01 from rain– Fog/Environmental
• SLR System:– USB physical limitations– Camera control software– Housing– Algorithm Efficiency
Questions??