【作者】 王焜；

【导师】 李翠华；

【作者基本信息】 厦门大学 ， 计算机应用技术， 2009， 硕士

【摘要】 视频运动目标识别、跟踪是计算机视觉研究领域的重要课题之一,在视频监控、人机交互、机器人视觉导航以及智能交通控制中具有广泛的应用前景。而实现海上运动目标的自动检测、识别和跟踪对于统计海面上来往船只信息、防止船只碰撞以及海上安防都具有重要的意义。本文主要研究海面运动船只的识别与跟踪技术。首先概述了海上运动目标检测和跟踪的研究现状;对目前主要的显著区域提取、运动目标识别和跟踪方法进行了简要概述;提出了基于视觉注意和HOG特征相融合的海上船只目标检测方法;利用多特征融合的粒子滤波算法对运动目标进行了跟踪。本文的主要研究工作和创新点包括以下几点:1、对视觉注意力模型进行了深入研究,提出了基于改进的视觉注意力模型的船只显著区域检测方法。在获得的显著图的基础上,通过阈值分割和区域提取快速获得船只显著区域,为进一步的船只目标识别打下了良好的基础。2、在视觉注意力模型和HOG特征的基础上,提出了视觉注意和HOG特征双向融合的船只目标检测机制。模拟人类视觉中有意识“主动寻找”与无意识“被动受吸引”相交互的视觉过程。通过视觉注意快速获得显著性船只区域,减小目标搜索区域;在显著性船只区域内,通过HOG特征和学习机制实现船只目标的精确识别。同时本文还提出在显著性船只区域内通过边缘检测和区域跟踪相结合来快速获得船只候选区域的方法。实验表明,这种双向融合机制对海上船只目标检测具有准确性高、处理速度快的特点,基本能够满足复杂场景下实时处理的要求。3、对粒子滤波方法进行了深入研究并应用于海面船只跟踪系统中,通过HSV颜色直方图和形状特征相融合的方法,提高了跟踪系统的准确性和鲁棒性。在对多目标、目标交叉遮挡等情况下的跟踪也表现出了良好的性能。更多还原

【Abstract】 Video moving targets detection、recognition and tracking are one of the most important issues in the field of computer vision. It has wide application prospects in many areas, such as video surveillance、human-computer interaction、robot visual navigation and intelligent traffic control. Implement of maritime objects detection、recognition and tracking is significant for statistics of vessel information、avoiding sea collision between vessels and coastal defense.This paper mainly researches maritime ship identification and tracking technology. First, it describes current development status of maritime moving target detection and tracking research, and then outlines some main methods of salient region extraction, moving target identification and tracking, and we present a method of ship detection based on the visual attention and HOG feature. Finally we use the particle filter of multi-feature combining method to track the moving targets. The main research innovations and contributions are summarized as follows:1. We take a deep research into the model of visual attention, and propose a method of salient ship region detection based on improved visual attention. Threshold division and region extraction are used to get salient ship region quickly based on saliency map to lay the foundation for ship identification.2. Based on the attention model and HOG feature, we present a ship target detection mechanism of visual attention and HOG feature two-way integration, and simulate the process of interactive visual about the human visual awareness in the "active search" and unconscious "passive attracted by". We quickly get salient ship region through the visual attention, reduce the search region, and then achieve the precise recognition in the saliency ship region through the HOG feature and Learning mechanism. Simultaneously, we propose a method to get candidate ship location quickly by edge detection and region extraction. Experiment results show that this two-way integration mechanism of ship target detection has feature of high accuracy, fast processing speed, and it basically meet the requirements of real-time processing for complex scene.3. We explore the particle filtering method and put it into the ship tracking system. We improve the accuracy and robustness of the tracking system through the combination of HSV color histogram with shape features. The system also shows a good performance in many situations such as multi-target tracking, target under cross-blocked, and so on.更多还原