How WormLab Tracks •Supports high mag and low mag (whole plate) tracking •Composed of 2 parts: • Detection (finds new worms as the enter the movie) • Tracking (determining changes in worm position and shape from frame to frame) •Thresholding tools to refine background and improve detection despite moderate background clutter •Uses geometric model, worm motion model, backtracking and Multiple Hypothesis Tracking for accurate detection mbfbioscience.co m
This document is posted to help you gain knowledge. Please leave a comment to let me know what you think about it! Share it to your friends and learn new things together.
Transcript
How WormLab Tracks
• Supports high mag and low mag (whole plate) tracking
• Composed of 2 parts:• Detection (finds new worms as the enter the movie)
• Tracking (determining changes in worm position and shape from frame to frame)
• Thresholding tools to refine background and improve detection despite moderate background clutter
• Uses geometric model, worm motion model, backtracking and Multiple Hypothesis Tracking for accurate detection
mbfbioscience.com
Worm Detection
• The image is inverted and segmented to identify potential worm objects
• The algorithm measures 2 points of high curvature from a closed planar B-spline curve around the boundary of the worm object
mbfbioscience.com
Head and Tail Determination
• Identification based on the worm’s shape and frequency of movement
• We apply the same spatial and temporal cues used by human observers: • The worm’s tail area is lighter than the head • The worm’s tail area is thinner than the
head • The head moves more frequently than the
tail
• Head/tail identification can be swapped for entire track by user
mbfbioscience.com
Detected Head
Geometric Model
• Based on the center line of the worm and boundary
• Modeled on a spline basis to allow easy scaling and resampling at different resolutions
• User can determine the # of points along the center line used in the analysis• 3 pts: head, tail, center
• 17-19 pts: bending analysis
• 59 pts: full resolution (default)
mbfbioscience.com
Worm Motion Model
• ɳ = movement along centerline (peristaltic progression factor)
• Δα = Displacement orthogonal to the trajectory
• Also use elongation and contraction to model motion
mbfbioscience.com
Tracking Across Frames
• A deformable model estimation algorithm fits the geometric model from the previous frame to the current frame
• Backtracking is performed to re-establish worms with their previous tracks if lost
• Backtracking used if video starts with entangled worms
mbfbioscience.com
Multiple Hypothesis Tracking
• Apply a set of hypothesized worm locations across time, thus building a hypothesis tree
• Resolve conflicts by finding the path of Maximum Fitness (best fit across frames)
mbfbioscience.com
Detection of Complex Behaviors
• The geometric model, worm motion model, and MHT help identify worms in ambiguous conformations:• Coiled worms, • Overlapping worms• Omega bends• Reversing worms
mbfbioscience.com
Editing Functions
• Manually draw a worm that is not detected prior to tracking
• Swap head and tail across a track• Join tracks• Split tracks• Delete worms per frame or across all frames
mbfbioscience.com
Metrics and Analyses
• Length: Distance between head and tail along central axis
• Width is calculated from N points along the worm• Direction is the direction of travel• Postion is the center of the median axis• Instaneous speed: Velocity along the central axis from
one frame to the next• Moving Average Speed: Instantaneous speed
averaged over multiple frames• Amplitude: Amplitude of the sine wave that best fits
the worm posture• Wavelength: Period of the sine wave the best fits the
worm’s posture • Bend Angle: Bending angle at the midpoint
mbfbioscience.com
Detection of Omega Bends
mbfbioscience.com
• Begins when the bending angle between head-midpoint and tail-midpoint drops below 1.57 radians ( 90°) and continues until the angle exceeds 1.57 radians
Detection of Reversals
mbfbioscience.com
• Reversal is defined as worm moving backwards for user defined set of frames
Head Bending Analysis
mbfbioscience.com
• Indicates foraging
• Worm sampled with 19pts
• Bending angle is 3pt from head
Imaging Suggestions
• Contrast: dark solid worms on light background• Lawn: replate worms to minimize tracks• Frame Rate: 5-10fps is adequate, faster for swimming
worms• Cameras:
• Industrial machine vision cameras (CCD) work• Webcams (low cost CMOS not so much)• Recommend monochrome cameras
• Image size: • Whole plate: 2500x2500 resolution (5 Megapixels)• Single worms: 800x600, 1200x1024 and faster frame rates
mbfbioscience.com
Video provided by Dr. Kate Harwood
WormLab Overview
• PC & MAC compatible
• Accepts video files in numerous formats
• Includes data and video export (with tracking overlay)
• Workflow based – easy to train and use
• Export metrics to Matlab and Excel
mbfbioscience.com
• Control camera hardware to record videos from stereoscopes, inverted microscopes, or macro photography setup• Automatic Save• Variable Frame Rate • Scaling Tool: Calculate the pixel size• Scaling and frame rate are saved within the video file, and
automatically read by WormLab for analysis• Support DCAM/IIDC compliant cameras (Point Grey, Allied
Vision and Sony)
Camera Control
mbfbioscience.com
• Track swimming, thrashing worms:
• Use a modified worm motion model to map the oscillation of the center point radially
• Quantify pharyngeal pumping
• Synchronization of stimulation and tracking
• New analyses for bending and shape interpretation
• Development of different assays – chemotaxis studies, etc.
WormLab – Future Directions
mbfbioscience.com
Summary
• WormLab for automatic detection and tracking of worms
• Provide metrics including size, speed, direction• Track in complex backgrounds, entanglements,
and shapes• Capture video sequences or open previously