摘 要

运动目标跟踪吸引了计算机视觉的极大兴趣。这项工作涉及在三维轨迹分析的水面完成中设计运动员的有效跟踪视频。我们的研究是基于在比赛中皮划艇比赛中，使用多台摄像机从视频图像分割和跟踪目标（运动员），其目的是提供能够预测任何遮挡物体的位置跟踪系统，以记录轨迹和进行3D可视化。在计算机视觉领域，水面云动目标是一项具有挑战性的任务。这是一个具有挑战性的任务，因为目标是在混乱的环境中，以及溅泼的水。通过视频跟踪，我们首先分析水面的颜色和运动统计特征。为了提取目标，我们提出了一种两步分割算法，并对运动员中心的三维轨迹进行了估计，建立了一个多视图跟踪系统，利用水的运动性质和颜色属性建立了一个多视图跟踪系统。考虑到算法各有优点和缺点，在文献中我们采用了一些技术方法来介绍和分析算法。为了将表面积分离到运动区域，使用了两种统计特征。摄像头被放置在河的不同一侧，因此为了融合所有的线段结果，我们必须使用扩展的卡尔曼滤波器。为了将我们的结果与标准的三维轨迹进行比较，我们提出了利用独木舟上的标记来计算的多视图系统。在研究文献之后，实践表明我们的轨迹和标准轨迹之间的中间平方误差非常低。

关键字：水面云动目标, 统计模型，透视投影，扩展卡尔曼滤波器，多视点融合。

Abstract

The tracking of moving targets has attracted a great deal of interest in computer vision. The work involves designing of an efficient tracking video of athletes in a water surface competition for 3D trajectory analysis. We based our research on canoeing competition, using multiple cameras from the video image segmentation and tracking the targets (athletes) during the race, the aim is to provide a tracking system able to predict the position of any occluded objects in order to record trajectory and carry on 3D visualization. In the field of computer vision, water surface moving targets is such a challenging task. What makes it such a challenging task is that the targets are in a cluttered environments and the splash water that randomly splash them. Using video tracking, we start by analyzing water surface’s color and motion statistical characters. To extract the targets we proposed a two-step segmentation algorithm and to estimate 3D trajectory of athletes’ center, a multi-view tracking system is established by using the motion property and color property of water. Considering the several amount of algorithm each having strength and weakness,in the literature we took in counts some technologies approaches on algorithm that we introduce and analyzed. In order to separate the surface’s area to the area of the movement two types of statistical feature were used. Cameras are placed on different side of the river, so to fuse all the segments result we had to employ the Extended Kalman Filter. To compare our result to the standard 3D trajectories, we proposed the multi-view system which to be calculated by using markers on canoes. After studying the literature, the practice showed us that the root median square error between our trajectories and the standard one is very low.

Keywords: water surface moving targets, Statistics model, perspective projection, Extended Kalman filter, Multi-view fusion.

Contents

Chapter 1 Introduction…………………………………………….1

1.1 Significance……………………………………….……………………………….1

1.2 Objective………………………………………….……………………………….1

1.3 Motivation…………………………………….…………………………………...2

1.4 Research status……………………………….……………………………………2

1.4.1 Motion Target and Segmentation…………………….……………………….2

1.4.2 Water Surface Moving Target Segmentation………………………….………4

1.4.3 Multi Camera Tracking Technology……………………………….………….5

1.4.4 Moving Target Tracking………………………………………………………6

1.5 Contribution of the Work…………………………………………………………..9

1.6 Thesis Organization……………………………………………………………..10

Chapter 2 Target Extraction Based on Statistic Model……………………….....11

2.1 Introduction……………………………………………………………………..11

2.2 Target Extraction in Cluttered Water Surface Background………………………12

2.3 Moving Areas Segmentation……………………………………………………..13

2.3.1 The image Difference Method……………………………………………….13

2.3.2 Application…………………………………………………..………………14

2.4 Water Surface’s Color Model………………………………….…………………15

2.5 Targets Extraction Based on Color Segmentation………………….…………….15

2.6 Chapter Conclusion……………………………………………………………..17

Chapter 3Designment of Tracking System……………………………………...19

3.1 The Extended Kalman Filter…………………………………………………….19

3.2 The structure of the Tracking System……………………………………………21

3.3 Chapter Conclusion……………………………………………………………...24

Chapter 4 Experiment and Result Analysis……………………………………23 4.1 On a Smooth River……………………………………………………………….23

4.2 On a Disturb River……………………………………………………………….24

4.3 Chapter conclusion……………………………………………………………….25

Chapter 5 Conclusion...…………………………………………………………….26

References………………………………………………………………………….27

Acknowledgement…………………………………………………………………28

Chapter 1 Introduction

1.1 Significance

Sporting events are the most popular form of remote live entertainment in the world attracting millions of viewers on television, personal computers, and a variety of emerging devices. They are produced using a numerous cameras and microphones. This production process continues to evolve as it strives to engage and immerse viewers in the action suspense, and drama of the remote live event Video tracking is the process of locating a moving object or multiple objects over time using a camera. It has a different type of uses, some of them are: human-computer interaction, security and surveillance, video communication and compression, augmented reality, traffic control, medical imaging and video editing. But here we will be focusing on water surface targets tracking. The principal application is to detect and track the contour of objects moving in a cluttered environment that is going to make it so challenging. Video tracking can be a time consuming process due to the amount of data that is contained in video. More again we can to the complexity the possible need to use object recognition methods for tracking, a challenging problem in its own way. Keeping a fix on a target at sea can be demanding. To perform video tracking an algorithm analyzes sequential video frames and outputs the movement of targets between the frames. There are a variety of algorithms, each having strengths and weaknesses[1].It is important to know which algorithm to use basing on the intended use. There are two major components of a visual tracking system: target representation and localization, as well as filtering and data association. Target representation and localization is mostly a bottom-up process. These methods give a lot of tools for identifying the moving target. Locating and tracking the target object successfully is dependent on the algorithm. For example, using blob tracking is useful for identifying human movement because a person's profile changes dynamically. Typically the computational complexity for these algorithms is low.

1.2 Objective

This paper presents a real time tracking system for moving objects on a water surface. The study is operated in a river where athletes perform, that’s a challenging task regarding material we are using cause of the cluttered environment moving targets are and the splashed water that occluded them.For tracking water surface moving targets, some visual tracking methods have been proposed. A region based active contour model is used to extract the moving ship, but doesn’t mention how to initial the snake. In another paper, spatial and temporal statistical information form a Markov Random Field. Then MAP inference is employed to find moving targets. These methods are 2D tracking algorithms and suitable for all situations including water surface targets.

Since there is not special tracking algorithm for water surface target, a multi-view tracking system is constructed in this paper. By employing motion and color statistical characters of water surface, an automatically target extraction algorithm is proposed to extract moving targets without any prior information of background and targets. In the framework of Extended Kalman Filter, the 3D trajectory of water surface moving target is estimated.

We employ this algorithm to track canoes running on the rivers. Then 3D trajectories of the canoes are calculated. Another research group from sport department has put some markers on canoes, and calculated canoes’ standard 3D trajectories according to these markers. Experiments show that the error between our data with the standard trajectory is very little

1.3 Motivation

After a great understanding of the literature, we have come to find that detecting the target from the video sequence and tracking it is a really challenging task. Moving target tracking can be a time consuming operation regarding the amount of data that is contained in the video. From the literature survey it is found that there are manykinds of algorithms based on the video trackingthat work efficiently in both indoor and outdoor tracking system. Video tracking is the most active research topic in computer vision for humans and vehicles. Here the aim is to develop an intelligent visual tracking system by re-placing the age old tradition method of monitoring by human operators. The motivation in doing is to design a video tracking system for water surface moving targets tracking.

1.4 Research status

1.4.1 Motion Target detection and segmentation

Moving object detection and segmentation is a hot research topic in the field of computer vision, when weather or lighting changes in a scene, in the background constantly swaying branches of the trees, shade between tracked target or self-occlusion phenomenon occurs, such events make detection and segmentation difficult[2].

Many world-wide journals (Image Process, IJCV, PAMI, PR, PR Letters, Signal Processing, Computer Vision and Image Process, Digital Signal Process and so on) have taken the target detection and segmentation as an important and special research content. Important international conference as ICCV, ACCV, CVPR, ECCV, ICPR, ICIP, and so on also target the tracking problem as a separate unit for new results released. Target detection, target tracking field as much at the forefront of research on the interdisciplinary research, have produced a large number of high quality, high power, wide applicable scope of the results of theory and practice. The well-known universities in China and abroad have established the laboratory based on computer vision.