Binocular Fixation by Tracking of Homogenous Colour Regions

نویسندگان

  • JONAS FORSSTRÖM
  • Mårten Björkman
چکیده

This thesis presents a binocular visual system constructed for fixation of objects of homogenous colour. The purpose of this approach to fixation is to evaluate the use of colour cues, which are well-defined in homogenously coloured regions, where earlier used techniques such as optical flow and correlation are ill-posed. Two colour-based tracking methods are developed and implemented, one probabilistic based on Bayesian probabilities and adaptive Gaussian mixtures, and one clustering algorithm based on the kmeans clustering idea and colour cluster flow. The stereo integration is based on a cyclopean probability map and can handle occlusions and separate same-coloured objects at different depths. Neither the tracking nor the stereo integration relies on explicit matching, which makes the system more robust. The resulting fixation system runs at about 10 Hz and can fixate homogenously coloured objects in most situations. The situations where the system tends to loose fixation is when similarly coloured objects reside close to the target and are not substantially separated in depth. These problems are associated with the inability to establish the boundary of the target, which suggests that colour-based cues should be complemented by edgebased cues. Binokulär fixering genom följning av regioner med homogen färg Sammanfattning I detta examensarbete konstrueras ett seende system som kan fixera objekt som är homogent färgade. Syftet med denna ansats är att utvärdera användningen av färgledtrådar, vilka är väldefinierade på enhetligt färgade ytor där tidigare beprövade tekniker som korrelation och optiskt flöde inte är väldefinierade. Två färgbaserade följningsalgoritmer har utvecklats och implementerats, varav en bygger på bayesianska sannolikheter och adaptiva gaussiska mixturer och den andra spårar färgkluster som skapats med en modifierad k-meansalgoritm. Stereointegrationen är baserad på en cyklopisk sannolikhetskarta och kan både hantera objekt som är dolda i en kamera och separera objekt som befinner sig på olika djup. Varken följning eller stereointegration bygger på explicit matchning vilket gör systemet mer robust. Det resulterande systemet exekverar i ca 10 Hz och kan fixera enfärgade objekt i de flesta situationer. De situationer där systemet tenderar att tappa fixeringen är främst när objekt med liknande färg befinner sig nära det fixerade objektet och dessa inte är tillräckligt separerade i djup. Dessa problem beror till stor del på att systemet inte kan avgöra var målobjektets kanter befinner sig, vilket visar på att färgbaserade ledtrådar bör kompletteras av kantbaserade ledtrådar. Foreword This thesis is the report from a master’s project in computer science at the School of Engineering Physics at the Royal Institute of Technology (KTH) in Stockholm, Sweden. The project has been performed at the center for Computer Vision and Active Perception Laboratory (CVAP), a research group within the Department of Numerical Analysis and Computer Science (NADA) at the Royal Institute of Technology. Acknowledgements I would like to thank my supervisor Mårten Björkman for helping me find and carry through this interesting master’s project. I would also like to thank Mårten for the patience shown in times when I could not put as much time into the project as I should have.

برای دانلود متن کامل این مقاله و بیش از 32 میلیون مقاله دیگر ابتدا ثبت نام کنید

ثبت نام

اگر عضو سایت هستید لطفا وارد حساب کاربری خود شوید

منابع مشابه

Short-latency ocular following in humans: sensitivity to binocular disparity

We show that the initial ocular following responses elicited by motion of a large pattern are modestly attenuated when that pattern is shifted out of the plane of fixation by altering its binocular disparity. If the motion is applied to only restricted regions of the pattern, however, then altering the disparity of those regions severely attenuates their ability to generate ocular following. Th...

متن کامل

Video Segmentation for Markerless Motion Capture in Unconstrained Environments

Segmentation is a first and important step in video-based motion capture applications. A lack of constraints can make this process daunting and difficult to achieve. We propose a technique that makes use of an improved JSEG procedure in the context of markerless motion capture for performance evaluation of human beings in unconstrained environments. In the proposed algorithm a non-parametric cl...

متن کامل

Discriminating classes of sequential and qualitative colour schemes

We present an analysis of the influence of colour distance on the user experience with choropleth and chorochromatic maps. We systematically evaluated five sequential and five qualitative colour schemes in a two-stage user experiment. At first, we conducted an online study to obtain performance metrics accuracy and response time from a large, heterogeneous population. Following this, in a contr...

متن کامل

JEP:HPP, in press Binocular fixation disparity in single word displays

It has been claimed that, when a word is presented in isolation, recognition is affected by the precise location at which the word is fixated. However, this putative role for fixation location has yet to be reconciled with the finding from reading research that fixations made by each eye are frequently misaligned and so more than one location in a word is often fixated simultaneously. The accur...

متن کامل

An Empirical User Study for Measuring the Influence of Colour Distance and Font Size in Map Reading Using Eye Tracking

The primary goal of this study is to empirically analyse the influence of colour distance and font size on map readability. We utilized eye-tracking to complement the classical usability metrics; thus, we studied performance metrics such as effectiveness (i.e. success, accuracy), efficiency (i.e. time to answer, task completion time), and selected eye tracking metrics fixation frequency, fixati...

متن کامل

ذخیره در منابع من


  با ذخیره ی این منبع در منابع من، دسترسی به آن را برای استفاده های بعدی آسان تر کنید

برای دانلود متن کامل این مقاله و بیش از 32 میلیون مقاله دیگر ابتدا ثبت نام کنید

ثبت نام

اگر عضو سایت هستید لطفا وارد حساب کاربری خود شوید

عنوان ژورنال:

دوره   شماره 

صفحات  -

تاریخ انتشار 2004