Please use this identifier to cite or link to this item: http://dspace.dtu.ac.in:8080/jspui/handle/repository/14722
Full metadata record
DC FieldValueLanguage
dc.contributor.authorPAWAR, GAURAV-
dc.date.accessioned2016-05-12T12:44:14Z-
dc.date.available2016-05-12T12:44:14Z-
dc.date.issued2016-05-
dc.identifier.urihttp://dspace.dtu.ac.in:8080/jspui/handle/repository/14722-
dc.description.abstractOcclusion handling is one of the most challenging problems in MOT (multiple object tracking). To deal with the problem of occlusion handling we proposed a novel approach based on neural network classifier with particle filter and HOG feature descriptors filter. Initial step in MOT is object detection which is done by background subtraction. After obtaining detected objects further neural network (NN) classifier is applied to verify that weather object detected is of our interest or not. Suppose if we want to track human, then with the help of NN classifier we can discard non-human objects. After object detection we will compute HOG feature descriptors for detected objects. But it is not robust to use all feature descriptors for feature matching as some of them are redundant. So to deal this problem we will assign weights to feature descriptors according to their ability of distinguishing one object from another. Feature descriptors which do not match with other object features or features which are usually similar to other object features should get assigned lower weights. And features which match only half of times to other object features should get assigned maximum weight. To track the objects we will use Particle filter as it can track objects which are moving in non-uniform pattern. Use of Particle filter will reduce region of interest for object matching. To deal with the problem of occlusion handling we will do part based segmentation of object. If for example we use human as an object to be tracked, so in this case we will train a neural network classifier to segment each part of human body like head, hands and torso. So even if object is partially occluded, some part of the object still remains visible. That visible part is detected by Neural Network classifier and matching of that part is done with the corresponding part of other objects. If object is completely occluded or get disappeared then particle filter will be used to predict the location of that object.en_US
dc.language.isoen_USen_US
dc.relation.ispartofseriesTD NO.2016;-
dc.subjectOCCLUSION HANDLINGen_US
dc.subjectMULTIPLE OBJECT TRACKINGen_US
dc.subjectPARTICLE FILTERen_US
dc.subjectNEURAL NETWORKen_US
dc.titleOCCLUSION HANDLING IN MULTIPLE OBJECT TRACKINGen_US
dc.typeThesisen_US
Appears in Collections:M.E./M.Tech. Electronics & Communication Engineering

Files in This Item:
File Description SizeFormat 
2k13spd06.pdf1.61 MBAdobe PDFView/Open


Items in DSpace are protected by copyright, with all rights reserved, unless otherwise indicated.