Files in this item



application/pdfECE499-Sp2017-sun.pdf (2MB)
(no description provided)PDF


Title:Object tracking with background elimination
Author(s):Sun, Yu
Contributor(s):Levinson, Stephen E.
Subject(s):computer vision
Gaussian mixture model
Abstract:Humans can follow moving objects and grasp an object of interest starting from early childhood. To grasp an object successfully, one needs to be able to recognize the object of interest against its background. If the object is moving, it is also critical to realize that the same object may look different from different perspectives. Then, an estimation of depth is required to decide if the object is within reach. Finally, when ready to grasp the object, one needs to figure out the part of the object to be held and the gesture of the hand to grasp the object. This process might be intuitive for humans, but it is not as simple for robots. The purpose of this research is to program the robot so that it can behave like a child who would follow moving objects unconsciously. Therefore, the thesis mainly focuses on the first step: object recognition. To begin, a simple object detector using color detection and segmentation was implemented. Next was experimentation with object tracking algorithms which perform feature matching based on feature descriptors like SIFT or ORB. Last, the Haar Feature-based Cascade Classifier was used to look at the learning based method. All these methods helped to understand how people approach the problem of object recognition, but they either have substantial limitations or do not quite satisfy the purpose of the research. Therefore, finally implemented was an object detection method by performing background elimination, which is able to track an arbitrary number of moving objects fast.
Issue Date:2017-12
Date Available in IDEALS:2017-08-28

This item appears in the following Collection(s)

Item Statistics