Multiview Imaging HW Overview

11
“A taster of multiview computer vision” Homework on multiview computer vision Victor Lempitsky (Skoltech) for the “Computational Camera” class

Transcript of Multiview Imaging HW Overview

Lecture 8: Lagrange duality

Homework on multiview computer visionVictor Lempitsky (Skoltech) for the Computational Camera class

A taster of multiview computer visionVideo example

A taster of multiview computer visionPipelineTake photographsDetect/describe featuresMatch featuresEstimate homographyGenerate a sequence of blended framesWrite a video

A taster of multiview computer visionTake photographsTake a sequence that will blend nicely Remember that the algorithm will find the dominant matching planeDownsample to 1-2Mpix

A taster of multiview computer visionFeature Detection/DescriptionFeatureDetector::create (SIFT)FeatureDetector::detectDescriptorDetector::create (SIFT)DescriptorExtractor::computeDetecting features:Describing features:

A taster of multiview computer vision

Reminder

=> Feature vector (128D)

A taster of multiview computer visionMatching featuresDescriptorMatcher::knnMatch(k=2)Take only matches, where the distance to the first neighbor is less then e.g. 0.7*distance to the second neighbor(Do matching the opposite way and keep only reciprocal matches)Use drawMatches to debugUse FlannBasedMatcher if too slow

A taster of multiview computer visionMatching results

A taster of multiview computer visionEstimating homographyYou need findHomography (calib3d module)Use CV_RANSAC optionPlay with reprojThresholdApply the estimated homography to the candidate matches and store the verified matches for later

Use to Mat::inv() to find the reverse mapping

A taster of multiview computer visionReminder how to apply homography

A taster of multiview computer visionImage blendingUse warpPerspective() to map imagesUse image arithmetics (+,*) to blendDefine intermediate frames (e.g. by moving the verified matches along the trajectories and refitting homographies without RANSAC)Use VideoWriter (highgui) to do the video writing

A taster of multiview computer vision