文档下载 文档下载导航
设为首页 | 加入收藏
搜索 请输入内容:  
 导航当前位置: 文档下载 > 所有分类 > Adaptive Dynamic Contour for Real-Time Object Tracking
侵权投诉

Adaptive Dynamic Contour for Real-Time Object Tracking

This paper addresses an efficient tracking algorithm based on adaptive dynamic contours, which can track 2D outlines of objects with arbitrary shapes in canned or real time image sequences. We represent contours using cubic B-splines by a number of control

Adaptive Dynamic Contour for Real-Time Object Tracking

H Wang1 J Leng1 Z M Guo2

1

School of EEE, Nanyang Technological University, Singapore 639798 2

Singapore Technologies Dynamics Pte Ltd, 249 Jalan Boon Lay, Singapore 619523

Fax: (+65) 67920415

E-mail: hw@ntu.edu.sg

Abstract

This paper addresses an efficient tracking algorithm based on adaptive dynamic contours, which can track 2D outlines of objects with arbitrary shapes in canned or real time image sequences. We represent contours using cubic B-splines by a number of control points. Window-based processing is used to reduce

computation. Our tracker is novel in that we use an adaptive contour model: After control points are drawn to the boundary by minimizing image energy, they are adjusted adaptively on each frame to keep to be evenly distributed on the contour, a method using vector product is proposed to prevent control points from crossing over during the tracking process; prediction is made on each control point based on the movement of the centroid of the object by the g-h filter. Edge Polarity and restricted searching region are employed to reject wrong search results. Our method is better comparing with deformable template based algorithms in that it is simpler and doesn’t need prior shape information. It is also applicable to both rigid and non-rigid objects. Experimental results with good accuracy showed that our algorithm is efficient and reliable.

1. Introduction

The tracking of moving contours in an image sequence/video-stream is an important area of research within the computer vision. Given a sequence of images in which an object of interest is in motion, important information such as object’s shape and current position can be obtained by tracking the object’s silhouette across time-varying images. Applications abound in medical diagnosis, surveillance and audiovisual speech recognition.

The active contour (Snake) representation [3] is the most popular method for contour delineation and tracking in still or time-varying images, which is first introduced by Kass et. [1] for lip tracking. By minimizing its energy function: Esnake

=∫Esnake(V(s))ds=∫(Eint(V(s))+Eext(V(s)))ds, (1)

ll

the snake algorithm makes an initial contour get to the boundary of the object. Where V(s)=(x(s), y(s)) parametrically represent the position of the snake. Many object tracking algorithms use this concept by defining the energy differently based on the different requirement [4,5]. Cipolla in [2] proposed to use B-splines to represent object’s contour instead of using the internal energy to reduce computation, which make the contour tracking generally more efficient and reliable.

Deformable templates [3,6] is also a widely used method, which uses a parametric shape-model r(s;X) known as a “deformable template” with relatively few degrees of freedom to model a prior-known object. The template is matched to an image in a similar manner with ‘snake’ by searching for the minimum external energy. The shortcoming of these trackers is that it must have prior knowledge about shape of the object to be tracked, and the outline representing will distort when slight errors occur.

In this paper we present a new method known as “adaptive contour” based on the concept of snake and B-splines. In this method, the outline of the object is represented by cubic B-splines using a number of control points. In section 2 we described window-based preprocessing and constrained edge detecting. Adaptive contour model is introduced in section 3, the adjusting of control points makes tracking process more stable

第1页

你可能喜欢

热门文档

相关文档

站点地图 | 文档上传 | 侵权投诉 | 手机版
本站所有资源均来自互联网,本站只负责收集和整理,均不承担任何法律责任,如有侵权等其它行为请联系我们.
文档下载 Copyright 2013 doc.guandang.net All Rights Reserved.  -  email