Motion trajectories


A major part of the previous research in Computer Vision has dealt with the recovery of 3-D information from 2-D images, using cues like stereo, motion, shading, texture, etc. These methods are called the shape from X methods. Marr envisioned that a goal of early vision should be to combine the outputs from these different channels into an intermediate representation called the intrinsic image or 2.5 D sketch. Recently, it has been argued that shape from X methods may not be that crucial for recognition tasks as have been perceived previously. Most approaches to "Shape from Motion" involve a number of assumptions regarding the objects and their motion, and can handle only restricted cases with a certain minimum number of points in some minimum number of frames. These approaches solve a system of non-linear equations using approximate methods which are very sensitive to noise. Further, the motion information has rarely been used beyond recovering structure, as in a practical system for recognizing objects. The goal of our research is to explore the use of motion information in object recognition, without explicitly recovering the structure of the objects. In our approach, we consider extended trajectories-trajectories consisting of large number of frames followed by objects and analyze them at multiple scales. We foresee the object recognition system using trajectories to be composed of three major modules. Trajectory generation, trajectory segmentation and trajectory matching. This dissertation has contributed to all the three stages. An optimal corner detector has been developed which detects corner points in the scene. These feature points are input to the correspondence module, which uses a good heuristic in minimizing the proximal uniformity cost function and generates trajectories even in the presence of occlusion. The Focus Of Expansion(foe) of a group of motion trajecto- ries, is defined to be a point in the image plane at which the trajectories intersect when they are extended. An algorithm has been designed to segment trajectories in a scene using foe. Procedures for qualitative and quantitative interpretations of trajectories using foe are reported. Simple and efficient algorithms have been developed to match single trajectories and groups of trajectories, which can form part of a model based object recognition system using motion and shape. The various modules developed under this dissertation have a lot of other applications beyond fitting into an object recognition system .


This item is only available in print in the UCF Libraries. If this is your thesis or dissertation, you can help us make it available online for use by researchers around the world by downloading and filling out the Internet Distribution Consent Agreement. You may also contact the project coordinator Kerri Bottorff for more information.

Graduation Date





Shah, Murbarak


Doctor of Philosophy (Ph.D.)


College of Arts and Sciences


Computer Science




193 p.



Length of Campus-only Access


Access Status

Doctoral Dissertation (Open Access)


Orlando (Main) Campus




Arts and Sciences -- Dissertations, Academic; Dissertations, Academic -- Arts and Sciences

Accessibility Status

Searchable text

This document is currently not available here.