Viola and Jones Object DetectorFast!Robust Real-Time Face DetectionDetection basis: FeaturesIntegral ImageComputing featuresClassifier: using AdaBoostPowerPoint PresentationSlide 9The CascadeSlide 11Slide 12Slide 13Slide 14Slide 15Slide 16Slide 17Slide 18Slide 19Slide 20Slide 21Slide 22Slide 23More: Detecting Walking PedestriansExtracting motion informationTraining Set SamplesSlide 27Slide 28Slide 29Slide 30Slide 31Slide 32Slide 33Slide 34Slide 35Slide 36Slide 37Slide 38Slide 39Slide 40Slide 41Slide 42Slide 43Slide 44Slide 45Slide 46Viola and Jones Object DetectorRuxandra Paun EE/CS/CNS 148 - Presentation 04.28.2005Fast! 15 times faster than any previous approach 384 by 288 pixel images detected at 15 frames per second on a conventional700 MHz Intel Pentium IIIRobust Real-Time Face Detection3 key contributors: - a new image representation: the “Integral Image” - a simple and effective classifier, based on the AdaBoost learning algorithm - combining the classifiers in a “cascade”Detection basis: FeaturesIntegral ImageComputing featuresClassifier: using AdaBoost160,000 features for every sub-windowVery small number of these features can be combined to form an effective classifierAdaBoost: constrain each week classifier to depend on a single featureeach stage of boosting = new week classifier selection = feature selectionFirst and Second Features Selected by AdaBoostROC curve for a 200 feature classifierThe Cascadecombining successively more complex classifiers in a cascade structure38 stagesROC curves: cascaded vs. monolithic classifier-> not significantly different accuracy-> but the cascade class. almost 10 times fasterResultsTraining dataset: 4916 imagesROC Curves for Face DetectionComparing Viola-Jones with Other SystemsMore: Detecting Walking PedestriansIntegrating image intensity with motion informationEfficient, detects pedestrians at small scales, and has a very low false positive rateWorks on low resolution images and under difficult weather conditions (rain, snow)Extracting motion informationTraining Set SamplesQuickTime™ and aYUV420 codec decompressorare needed to see this
View Full Document