Citation Context

...techniques selected in thesproposedsscenarios,sthesimplementationssofsthesGMMsmodelsandsActivesContours59 adjustmentsfromsthesVideosProcessingsandsUnderstandingsLabshavesbeensused.sThesLIBSVM library =-=[31]-=- has been used as the SVM implementation. The implementations ofsthe extraction method and the matching procedure for the Shape Context feature have beenstaken from thesauthor’sswebsites[16].sAddition...

Citation Context

...e typically used to classifysthe video content. In this context, object classification by using shape-based information is one of thesmost common approaches to classify objects in the monitored scene =-=[2]-=-[3]. The success ofsthissprocesssdependssonsthesdescriptionsofsthesobjectsfeaturessandsthesemployedsclassificationsschemesswhichsbringsthesnecessitysofsdeterminingsthesmostsappropriatesshape-based fea...

Citation Context

...ed.sThesusesofsaslog-polarscoordinate system for the histogram gives a better representation. This representation isscalled shape context. From Figure 12, the operations and shape context can be seen =-=[16]-=-. Figure-12:sThe first image represents detected object; the second image is the set ofsvectors that originate from one point to all others; and the third image is the shape contexts[16] 14 2.3.2 Regi...

Citation Context

...ement of the chain code from the previous one. This chain code representation is generally not the preferred one to be used in shapesmatching operations due to its sensitivity to noise and variations =-=[14]-=-.s12 II. Polygon Decomposition: Insthe extractionsofsthissfeature,sthe boundary issdividedsintoslinessegmentssbyspolygon approximation. For each vertices, a four-element string is expressed as a featu...

Citation Context

... video-surveillance systems understwo constraints: stationary camera and low illumination changes of the environment. Thestechnique used in this project is based on the noise introduced by the camera =-=[28]-=-. It hassbeen provided by the Video Processing and Understanding Lab and it is briefly describedsin Appendix 1. The result of this process is depicted in Figure 39. As it can be observed in thissfigur...

Citation Context

...mage pre-processing and shape extraction 2.2.1 Image Preprocessing Thesoperationsperformedsaftersframesacquisitionsissobjectsdetection.sThesmostscommon approaches in object detection are Optical Flow =-=[6]-=-, Background Subtraction [7],sTemporalsDifferencings[7]sandsStatisticalsMethodss[8].sAccordingstosrecentssurveyedspaperss[4][5][6]sobjectsdetectionsbased onsBackground Subtraction followed by objectst...

Citation Context

...ypically used to classifysthe video content. In this context, object classification by using shape-based information is one of thesmost common approaches to classify objects in the monitored scene [2]=-=[3]-=-. The success ofsthissprocesssdependssonsthesdescriptionsofsthesobjectsfeaturessandsthesemployedsclassificationsschemesswhichsbringsthesnecessitysofsdeterminingsthesmostsappropriatesshape-based featur...

Citation Context

...e extraction 2.2.1 Image Preprocessing Thesoperationsperformedsaftersframesacquisitionsissobjectsdetection.sThesmostscommon approaches in object detection are Optical Flow [6], Background Subtraction =-=[7]-=-,sTemporalsDifferencings[7]sandsStatisticalsMethodss[8].sAccordingstosrecentssurveyedspaperss[4][5][6]sobjectsdetectionsbased onsBackground Subtraction followed by objectstracking (in order to maintai...

Citation Context

...pplied, thensSVM classifier becomes:s∑ ∈ += Si iii wxxKyxf 0)()( λ . The decision about the test shape is done according to this function where input value issthe feature vector of the test shape [19]=-=[27]-=-.s34sFigure-32: Support Vector Machine Based Classifier and optimal hyper-plane [19] Inscasesofsmulti-classsclassification,sthesacyclicsgraphsmethodsissusedsandsthesalgorithmsofsthissmethodscansbesobs...

Citation Context

...ssand on the rightscorresponding log-polarshistograms can be observed.sEachsrowscontainsssimilarsshapes,ssosthessimilaritysorsdifferencesbetweenslog-polarshistograms of similar shapes can be observed =-=[13]-=-. 38sFigure-36: Database and corresponding log-polar histograms [13] 39 5 System integrations5.1 Introduction In this section, the framework that integrates the different techniques developed insthe p...

Citation Context

...initions:sglobal and structural.sGlobalsapproaches use the whole region of the object while structural methods divide the regionsinto sub-parts. 2.3.2.1 Global Methods: I. Simple Methods: • Fill Ratio=-=[8]-=-: This feature is the value that is calculated by the ratio of size of thesblob segment area (A) over the total minimum bounding rectangle area, defined bysits Height (H) and Width (W), as: ( * ) AF H...

Citation Context

...sdetection.sThesmostscommon approaches in object detection are Optical Flow [6], Background Subtraction [7],sTemporalsDifferencings[7]sandsStatisticalsMethodss[8].sAccordingstosrecentssurveyedspaperss=-=[4]-=-[5][6]sobjectsdetectionsbased onsBackground Subtraction followed by objectstracking (in order to maintain the ID of the classified object) is the most common solutionsadopted due to its simplicity and...

Citation Context

...s, a string is created such as: S=(s1S, s2S,.., sNS) where each s represents the set of these features. Insthesmatchingsstep,sassophisticatedsmodelsbysmodelsmatchingsissemployedsbetween these strings =-=[15]-=-. 13 V. Syntactic Analysis: Insthissmethod,sthe boundary issrepresentedswithspredefinedsprimitivessthatsarescalledscodewords.sForsexample,sascodeword ofsaschromosomesofsassamplesDNAsissrepresented as ...

Citation Context

...es is used where the classes are Human and Object classes wheres22 the samples can be seen from Figure 23. The human shapes class contains 170 samples andsthe object shapes class contains 117 samples =-=[25]-=-.sFigure-23: Human and Object Class database image samples Then foreground mask data of video-surveillance system is selected to match withsthese two-object class databases that are obtained from IPPR...

Citation Context

...ommunity focused their effort on the developmentsof different techniques to automatically detect and classify the video content by analyzingsdifferent patterns. Features like color, texture and shape =-=[1]-=- are typically used to classifysthe video content. In this context, object classification by using shape-based information is one of thesmost common approaches to classify objects in the monitored sce...

Citation Context

...ure and all other twelve objects are database shapes. According tosthe distance signalssshown under each shape,sthe matching shape issfound by distancesbased classification and represented in the box =-=[5]-=-. 37 Figure-35: Distance measure based classification sample [5] 4.2.3.2 Shape Context The Shape Context feature is classified by a different procedure where the idea is tosmeasuresthesdistancesbetwee...

Citation Context

...f them is usually used as a description ofsthe object. Thus, the extracted feature data (or descriptor) is generally an N-dimensionalsvector which can be regarded as a point in an N-dimensional space =-=[10]-=-.s1 2[ , ,..., ]N iDescriptor f f f where f is a feature= Features can be classified as contour-based or region based according to whetherscontoursorsregionsofsthesobjectsissusedsinseachsspecificstech...

Citation Context

...eached. As U is described above, A is found by this formula. This A matrix is the feature that isscalledsshapesmatrix.sFrom Figure 13,sdeformationsand matchingsoperationsson samplesshapes can be seen =-=[18]-=-. Figure-13: Matching and deformation by shape matrix [18] 16 2.3.2.2 Structural Methods: I. Convex Hull: When a convex region is represented assR, its convex hull is the smallest convexsregion that s...

Citation Context

... must be converted to a belonging probability ofsan object to a class. A graphical description of this classifier can be seen in Figure 17. Figure-17: Support Vector Machine Based classifier behavior =-=[19]-=- 2.4.2 Support Vector Data Description Based Classification: Thissclassificationsmethodsdefinessassphericallysshaped boundaryswithsminimalsvolume around the training data set; with this property it de...