Tracking and Motion

Tracking and Motion

정보산업공학협동과정정우정

Contents

The Basics of Tracking Corner Finding Subpixel Corners Optical Flow Mean-Shift and Camshift Tracking Motion Templates Estimators The Condensation Algorithm

example

…/opencv/samples/c/ lkdemo.c (optical flow) camshiftdemo.c (mean-shift tracking of colore

d regions) motempl.c (motion template) kalman.c (Kalman filter)

The Basics of Tracking

Understand the motion of object:identification and modeling

IdentificationTo finding the object of interest from the video stream.

ModelingProviding us with noisy measurement of the object’s actual position.

Corner Finding (1/2)

Harris corner Shi and Tomasi

Corner Finding (2/2)

image: input image (single-channel)

eigImage, tempImage: scratch by the algorithm

corners: result points after the algorithm

corner_count: maximum number of points

quality_level: minimal eigenvalue (0<x<1)

mask: usual imageblock_size: pixeluse_harris: Harris or Shi-Tomasik: weighting coefficient

Subpixel Corners (1/2)

Subpixel Corners (2/2)

image: inputcorners: initial guesses for

the corner locationcount: compute pointwin: size of windowzero_zone: window that will

not considercriteria: user-specified ter

mination criterion

Optical Flow (1/9)

Optical Flow (Lucas-Kanade Method) (2/9)

Optical Flow (Lucas-Kanade Method) (3/9)

Optical Flow (Lucas-Kanade Method) (4/9)

Optical Flow (Lucas-Kanade Method) (5/9)

Optical Flow (Lucas-Kanade Method) (6/9)

Optical Flow (Lucas-Kanade Method) (7/9)

Optical Flow (Lucas-Kanade Method) (8/9)

Optical Flow (Lucas-Kanade Method) (8’/9)

Optical Flow (Lucas-Kanade Method) (9/9)

imgA: initial imageimgB: final imagepyrA, pyrB: buffers allocated to store the

pyramid imagesfeaturesA: point for motionfeaturesB: new location point from featur

esAcount: number of points int the featureAwinSize: window sizelevel: depth of the stack of imagesstatus: 0/1 correspondingtrack_error: error valuecriteria: user-specified termination criteri

onflags: allow for some fine control

Mean-Shift and Camshift Tracking (1/3)

Camshift: Continuously Adaptive Mean Shift Algorithm

Mean-Shift and Camshift Tracking (2/3)

Mean-Shift and Camshift Tracking (3/3)

prob_image: density of probable locations

window: kernel windowcriteria: user-specified ter

mination criterioncomp: converged search wi

ndow locationbox: contain the newly resi

zed box

Motion Templates (1/5)

Motion Templates (2/5)

Motion Templates (3/5)

Motion Templates (4/5)

Motion Templates (5/5)

Estimators

Estimators (The Kalman Filter)

Estimators (The Kalman Filter)

Estimators (The Kalman Filter)

Estimators (The Kalman Filter)

Estimators (The Kalman Filter)

The Condensation Algorithm

감사합니다 (Q&A)