cs-muvi video compressive sensing for spatial multiplexing cameras
DESCRIPTION
CS-MUVI Video compressive sensing for spatial multiplexing cameras. Aswin Sankaranarayanan, Christoph Studer , Richard G. Baraniuk Rice University. Single pixel c amera. Photo-detector. Digital m icro-mirror device. Single pixel c amera. Each configuration of - PowerPoint PPT PresentationTRANSCRIPT
CS-MUVI Video compressive sensing for spatial multiplexing cameras
CS-MUVIVideo compressive sensing for spatial multiplexing camerasAswin Sankaranarayanan, Christoph Studer, Richard G. BaraniukRice University
Single pixel camera
Digital micro-mirror devicePhoto-detectorExplain DMD and linear measurement2Single pixel cameraEach configuration of micro-mirrors yield ONE compressive measurement
Non-visible wavelengthsSensor material costly in IR/UV bands
Light throughputHalf the light in the scene is directed to the photo-detectorMuch higher SNR as compared to traditional sensors
Digital micro-mirror devicePhoto-detectorRemember resolution of the device depends on that of the DMD ,,,,Sensing in non-visible wavelengths where sensing materials are costly3Single pixel cameraEach configuration of micro-mirrors yield ONE compressive measurement
static scene assumption
Key question: Can we ignore motion in the scene ?
Digital micro-mirror devicePhoto-detectorRemember resolution of the device depends on that of the DMD ,,,,Sensing in non-visible wavelengths where sensing materials are costly4SPC on a time-varying sceneNave approach: Collect W measurements together to compute an estimate of an image
what happens ?t=1t=Wmeasurementscompressive recovery
?
time varyingsceneDoes this strategy work ?5SPC on a time-varying sceneTradeoffTemporal resolution vs. spatial resolutiont=1
Small WLess motion blurLower spatial resolutionLarge WHigher spatial resolutionMore motion blurt=W (small)t=W (large)
Does this strategy work ?6
SPC on a time-varying scene
Lower spatial res.Higher temporal res.Higher spatial res.Lower temporal res.sweet spotRecall sweet spot depends on the object motion in the scene7Dealing with MotionMotion information can help in obtaining better tradeoffs [Reddy et al. 2011]State-of-the-art video compression
Dealing with MotionMotion information can help in obtaining better tradeoffs [Reddy et al. 2011]State-of-the-art video compression
nave reconstruction
motion estimatesKey pointsMotion blur and the failure of the sparsity assumptionUse least squares recovery ?
Recover scene at lower spatial resolutionLower dimensional problem requires lesser number of measurementsTradeoff spatial resolution for temporal resolution
Least squares and random matrices Random matrices are ill-conditionedNoise amplification
Hadamard matricesOrthogonal (no noise amplification)Maximum light throughputOptimal for least squares recovery [Harwit and Sloane, 1979]
Hadamard + least sq. recoveryHadamard
Random
Hadamard + least sq. recovery
Designing measurement matricesHadamard matricesHigher temporal resolutionPoor spatial resolution
Random matricesGuarantees successful l1 recoveryFull spatial resolution
Can we simultaneously have both properties in the same measurement matrix ?Dual-scale sensing (DSS) matrices
1. Start with a row of the Hadamardmatrix2. Upsample3. Add high-freq. componentKey Idea: Constructing high-resolution measurement matrices that have good properties when downsampled
Contrast this random matrices14CS-MUVI: Algorithm outlinet=Tt=1t=t0t=t0+Wt=W1. obtain measurementswith DSS matrices2. low-resolution initial estimate
3. motion estimation4. compressive recovery with motion constraints
Simulation result
CS-MUVI on SPC
Single pixel camera setup
Object
InGaAs Photo-detector (Short-wave IR)SPC sampling rate: 10,000 sample/sNumber of compressive measurements: M = 16,384Recovered video: N = 128 x 128 x 61 = 61*MCS-MUVI: IR spectrumJoint work with Xu and KellyRecovered Videoinitial estimateUpsampled
CS-MUVI on SPC
Nave frame-to-frame recoveryCS-MUVI
Joint work with Xu and KellyCS-MUVI summaryKey ingredientsNovel Measurement matrix designExploiting state-of-the-art motion model
One of first practical video recovery algorithm for the SMCdsp.rice.eduPractical - tackles motion blur that is fundamental to the SPC
We are consulting with them to implement this technology 20CS-MUVI summaryLimitationsNeed a priori knowledge of object speedMotion at low-resolutionRobustness to errors in motion estimates
Future workDual-scale to multi-scale matrix constructionsMulti-frame optical flowOnline recovery algorithmsdsp.rice.eduPractical - tackles motion blur that is fundamental to the SPC
We are consulting with them to implement this technology 21