【作者】 刘俊；

【导师】 黄席樾；

【作者基本信息】 重庆大学 ， 控制理论与控制工程， 2008， 博士

【摘要】 随着内河航运的增长,导致航运事故发生的风险也同样的增加。在雨、雪、雾、夜间等能见度不良气候条件下航行是造成船舶互撞和船撞桥事故的主要因素。船-船、船-桥避碰成功的关键是获取其他运动船舶和危险目标的准确信息。前视红外成像设备具有抗干扰能力强,气候环境适应性强,昼夜连续探测等优点。通过在各种内河行驶船舶上以及桥梁、闸口、限制区等重点区域安装价格便宜、技术成熟的非致冷红外焦平面阵列前视红外成像设备,实时采集红外视频图像,综合利用图像处理、目标检测、目标跟踪等技术,对采集的红外视频图像进行鲁棒的实时的分析处理、实现在内河复杂背景不良气候条件下,对其他内河运动船舶目标进行检测和跟踪,并利用得到的船舶目标检测和跟踪结果辅助船舶航行,提高监管人员和操船人员对航行环境的感知能力,辅助操船人员进行避撞决策,减少操船的失误,提高船、桥避碰成功率,保障人员的生命和财产的安全,减少或避免严重污染水域和自然环境事故的发生,确保航行运输安全。在基于视频的监控系统中有三个关键步骤:检测感兴趣的运动目标、跟踪这些目标、通过分析目标轨迹来识别相应目标的行为。红外目标检测和跟踪技术,作为智能化信息处理的关键环节之一,一直是困扰和制约红外成像探测实用性能的瓶颈问题和技术难点而亟待解决;同时,要把红外技术应用于内河水上交通安全,首先必须解决的关键技术就是基于红外图像的内河运动船舶目标检测和跟踪;因此,基于红外图像的内河运动船舶目标检测和跟踪技术研究,不仅具有重要的实用价值,还有重要的学术研究价值。本论文围绕基于红外图像的内河运动船舶目标检测和跟踪技术展开研究。首先,介绍前视红外成像系统的原理、组成及其优缺点、定性分析红外图像中内河船舶目标和背景的红外特征、红外图像的噪声特性、提出对目标检测和跟踪算法的性能要求;第二,在学习和借鉴已有天水线提取算法的基础上,提出了一种能够在复杂内河背景下进行天水线提取及其评价的方法;实验表明,该方法适应性好,定位精度高,实时性和可靠性高;给出了感兴趣区域ROI的提取方法;第三,总结已有的基于分形技术的人造目标检测算法,在此基础上提出了一种新的多尺度分形特征参数(MFFK);实验证明,当该参数应用于内河红外图像时,可以对内河船舶目标和内河自然背景进行很好的区分;进一步,提出了基于MFFK的内河船舶目标的检测算法;实验证明:该算法适用于内河复杂背景,适应性强,检测精度高,能够满足实时性和可靠性的要求;第四,首先对Mean Shift算法进行回顾,然后提出基于Mean-Shift的内河运动船舶目标跟踪算法。该算法最大的特点是在多尺度分形特征参数MFFK图像中描述内河运动船舶目标;实验证明:对红外图像中处于各种内河复杂背景中的单个运动船舶目标,该算法可实时、可靠、鲁棒的跟踪,但是当出现多个运动船舶目标相互遮挡时,该算法的跟踪可靠性降低;第五,首先回顾了粒子滤波相关理论及其在视频目标跟踪中的应用;然后提出基于单一灰度特征的粒子滤波内河船舶目标跟踪算法。在粒子滤波理论框架下,MFFK灰度图像中的内河船舶目标的状态后验概率分布用加权随机样本集表示,通过这些随机样本的Bayesian迭代进化实现对红外图像序列中的内河船舶目标跟踪;实验证明:单一灰度特征不足以描述内河运动船舶目标,该算法可用于简单内河背景,但不适用于内河复杂背景;第六,由于红外图像中的内河运动船舶目标的轮廓、形状和纹理特征一般不明显、没有颜色信息,同时单一灰度特征又不足以描述目标,因此,提出把船舶的灰度特征和运动特征融合来对内河运动船舶目标进行描述。通过在MFFK灰度图像中提取内河运动船舶目标的灰度特征,在两两MFFK灰度图像帧之间利用时间差分方法提取内河运动船舶目标的运动特征,利用模糊逻辑定义灰度特征与运动特征融合后的多特征融合相似系数;最终提出了基于灰度特征与运动特征融合的粒子滤波跟踪算法。该算法集成了分形几何、Mean Shift、差分运动检测、粒子滤波、模糊论等理论。实验证明:该算法不仅能够在内河复杂背景中对内河运动船舶进行稳健的有效的跟踪,而且能够应付场景的各种变化以及多运动船舶目标交错遮挡等情形,算法具有鲁棒性;算法在应用过程中仅需要很少的状态采样数,满足实时性的要求;最后,对全文的研究工作进行总结,指出今后工作中进一步研究的方向。更多还原

【Abstract】 The recent growth in inland waterway shipping traffic has resulted in a concomitant increase in the risk of shipping accidents, thus making collision avoidance a critical issue in inland waterway shipping traffic safety. The main reason resulted in ships collision is sailing under the restricted visibility conditions such as fog, mist, night and etc. The key of collision avoidance for the reference ship or the bridge, in which FLIR equipments is equipped, is to obtain accuracy navigation information about target ships which locate in front of the reference ship. The forward-looking infrared (FLIR) images have a lot of advantage, such as the capacity of resisting disturbance and the adaptability of weather is strong, the ability of passive detection is continuous day and night. The uncooled infrared focal plane arrays (FGA) FLIR camera, which have lower price and technical matured, is installed on the importance location such as in front of ship, bridge pier, strobe, the restricted area in the river. When infrared images are captured by real time, various technologies that include image processing, object detection, object tracking and etc are integrated and used to robust process and analyze these images in real time way. So, it can be achieved to detection and tracking other moving ships in inland waterway under poorly visible conditions. The detection and tracking other moving ships information are applied to assist sailing for ship. These information are used to improve the capability of apperception for ship’s driver and inspector, to assist ship’s driver decision for avoidance collision, to reduce driving mistake, to enhance the success ratio to avoid ship-ship or ship-bridge collision. Then, People casualty is safeguarded. The damage of ship, bridge and goods are avoided. Economical, social and environmental loss are reduced or avoided furthermore. Finally, the safety sailing is ensured.There are three critical steps in surveillance system analysis based on video frequency, i.e. interesting of moving object detection, object tracking and object’s behavior recognition based on object’s trial analyzed. Object detection and tracking from infrared image, which restrict and bother the practical detection performance, is one of the key stages in intelligent information processing field. It is a bottleneck problem and a technical difficulty unsolved. Meanwhile, when the infrared technology is used to transportation security in inland waterway, the critical technology should be solved is moving ship detection and tracking from infrared image firstly. To sum up, moving ships detection and tracking from infrared image in inland waterway have not only importance practical worthiness but also importance science research value.Firstly, the principle, components and advantage of FLIR system are introduced. The infrared characteristic of Ship and background in inland waterway are qualitative analyzed. The characteristic of noise in infrared image is analyzed. The performance of object detection and tracking algorithm is presented.Secondly, a method is proposed to extract and assessment the sky-water line under complicated inland waterway background based on understanding the existing methods. The result of experiment shows that the proposed method has wide adaptability and high precision, and it has fulfilled the demand of real-time and reliability.Thirdly, a review of man-made object detection algorithms is presented based on various fractal features, which are derived from the blanket covering method. Based on the review, a new multi-scale fractal feature parameter, i.e. multi-scale fractal feature related with K (MFFK), is presented. The results of experiments show that in MFFK image calculated from original infrared image performs the best discriminating capability between natural background and man-made object in fractal feature images.Furthermore, ship detection algorithm in inland waterway based on MFFK is presented. Experimental results have shown that the approach is feasible and effective under complicated inland waterway background. It has achieved real-time and reliable ship detection.Fourthly, mean shift algorithm is introduced. Then a moving ship tracking algorithm in inland waterway is proposed based on mean shift algorithm. The significant characteristic of the algorithm is that MFFK image is used to describe moving ship in inland waterway. Experimental results have shown that the proposed algorithm is effective and robust for tracking single moving ship in inland waterway from infrared image. Moreover, it is satisfied the request of real time tracking. However, the reliability of the proposed algorithm is going to depress due to ship-to-ship occlusions.Fifthly, the method and theory related to particle filter are surveyed. The applications of particle filter in object tracking field based on video sequence are reviewed. Then, a moving ship tracking algorithm is proposed based on single gray characteristic. Under the theory framework of particle filter, the posterior distribution of the moving ship in MFFK image is approximated by a set of weighted samples, while the moving ship tracking is implemented by the Bayesian propagation of the sample set. Experimental results have shown that the moving ship in inland waterway is not enough approximated by single gray feature. The proposed algorithm can be used in simple inland waterway background, and it isn’t applied in clutter inland waterway background.Sixth, the characteristics of moving ship silhouette, shape and texture in infrared image are generally unobvious in inland waterway. Furthermore, the moving ship in inland waterway is not enough described by single gray feature. So, fusion between gray feature and moving cue is presented for approximating moving ship in inland waterway. The gray feature of moving ship is extracted from MFFK image; the moving cue of moving ship is obtained by differencing between two MFFK image frames. The comparability coefficient of multi-characteristic fusion can be obtained by fusing between gray feature and moving cue based on fuzzy logic. Finally, a moving ship tracking algorithm is proposed based on characteristic fusion between gray feature and moving cue under the theory framework of particle filter. The algorithm is integrated with fractal geometry, mean shift, temporal differences method, particle filter, fuzzy, etc. Experimental results have shown that the proposed algorithm is not only used to tracking moving ship in complicated inland waterway background steadily, but also adapted to changing moving ship and the scene, non-rigid ship structures, ship-to-ship and ship-to-scene occlusion. The proposed algorithm is effective and robust. Moreover, it is satisfied the request of real time tracking due to require a low number of particles from the prior in real applications.Finally, the summary of the thesis is given. Furthermore, the further work and research prospects are introduced.更多还原