computer vision techniques for underwater navigation
DESCRIPTION
Computer Vision Techniques for Underwater Navigation. Chris Barngrover CSE 291. May 5, 2010. Research Motivation. Doppler Velocity Logger SONAR Cameras. Specific Motivation. AUVSI & ONR’s 13 th Annual AUV Competition. TRANSDEC. Research Goal. Detect and Classify Objects Buoy Pipe. - PowerPoint PPT PresentationTRANSCRIPT
Computer Vision Techniques for Underwater Navigation
Chris BarngroverCSE 291 May 5, 2010
Doppler Velocity Logger
SONAR
Cameras
AUVSI & ONR’s 13th Annual AUV Competition
Specific MotivationSpecific Motivation
TRANSDEC
Detect and Classify Objects
◦ Buoy
◦ Pipe
Research GoalResearch Goal
The StingrayThe StingrayCameras
FrameGrabber
Processor
Labeling Examples
Computer VisionComputer Vision
HSV Classifier◦ Hue – Saturation – Value◦ RGB is lighting dependant
Computer VisionComputer Vision
Boosting Algorithms◦ JBoost
Computer VisionComputer Vision
Binary Image
Computer VisionComputer Vision
Computer VisionComputer Vision
Detect & Classify Determine Center Location
Buoy DetectionBuoy Detection
Baseline Algorithm◦ HSV Range◦ Misses Reflection◦ Noise
Buoy DetectionBuoy Detection
Boosting Benefits◦ HSV Classifier◦ Robust Scoring per pixel◦ Reduced Noise
Buoy DetectionBuoy Detection
Opening◦ Reduces Noise◦ Erosion then Dilation
Buoy DetectionBuoy Detection
Closing◦ Fills holes◦ Dilation then Erosion
Buoy DetectionBuoy Detection
Convex Hull◦ Closes edges
Buoy DetectionBuoy Detection
Center Estimation◦ Centroids of Blobs◦ Largest Area Wins◦ Quality of Classifier
Buoy DetectionBuoy Detection
Hybrid Boosting◦ TRANSDEC & Pool◦ Separate Decision Trees◦ Additive Scoring
Buoy DetectionBuoy Detection
Reflection Problem◦ Larger Reflection Blob◦ Look at 2nd Largest
Buoy DetectionBuoy Detection
Buoy DetectionBuoy DetectionBaseline Metrics
Final Algorithm Metrics
Detect & Classify Determine Center Location Determine Bearing
Pipe DetectionPipe Detection
Baseline Algorithm◦ HSV Range◦ Finds Pipe Generally◦ Lots of Noise
Pipe DetectionPipe Detection
Boosting◦ HSV Classifier
Post Processing◦ Opening◦ Closing◦ Convex Hull◦ Smooth
Pipe DetectionPipe Detection
Edge Detection◦ Blob Perimeter◦ Canny Algorithm
Pipe DetectionPipe Detection
Hough Transform◦ Standard (SHT)◦ Probabilistic (PHT)◦ Multiple lines per edge
Pipe DetectionPipe Detection
Collinear Lines◦ Merge semi-collinear◦ Error from best-fit
Pipe DetectionPipe Detection
Parallel Lines◦ Remove solo lines
Pipe DetectionPipe Detection
Two Pipes◦ Match lines with center
of pipe
Pipe DetectionPipe Detection
Two Line Pairs◦ Choose pair closest
to the center
Pipe DetectionPipe Detection
Pipe DetectionPipe DetectionBaseline Metrics
Final Algorithm Metrics
Fish Detection
Quagga Mussels
Mine Detection
Future EffortsFuture Efforts
Perceptual Robotics Laboratory @ UMich◦ Visually Augmented Navigation◦ Autonomous Ship Hull Inspection
Koch Lab @ Cal Tech◦ Automated Event Detection in Underwater Video
Singh’s Lab @ Woods Hole◦ Underwater Photo Mosaicing
Related WorkRelated Work
Questions?Questions?