【作者】 吴亮；

【导师】 陈宗海； 盛捷；

【作者基本信息】 中国科学技术大学 ， 系统工程， 2009， 硕士

【摘要】 基于计算机视觉分析的目标检测技术是模式识别、图像处理等领域的重要研究课题。环境的复杂性和应用的实时性对处理技术提出了更高的要求,如何在保证实时处理的前提下,提高算法的鲁棒性和适应能力成为算法研究的重点。目标检测技术分为静态目标检测与运动目标检测。本文从静态模型参数估计和动态目标区域检测两方面开展研究内容,以提高算法鲁棒性和实用性为目标。目标模型参数估计中,针对RANSAC方法在模型尺度、数量等环境信息未知时鲁棒性能下降的问题,以椭圆模型为检测对象,提出了一种新的自下而上的估计方法。该方法采用先分类后聚合的估计过程,不仅能够克服离群数据的影响,还能在不提供先验信息的情况下以较高的精度估计模型参数,鲁棒性能良好。实验表明了该方法的有效性。实时运动目标检测技术在监控领域应用广泛,虽然目前较为流行的背景减方法速度快,易于实现,但由于受到阈值的限制,在颜色复杂的环境中适应能力不强。独立分量分析(ICA)是一种利用统计特征进行信号分离的方法,由于处理过程中受局部扰动的影响不明显,近年来已越来越多的应用于图像处理。本文提出了一种新的基于ICA的运动目标检测方法,该方法在两帧之间自适应分割运动区域,无阈值限制,能适应较为复杂的背景环境,且处理速度满足实时性要求。实验结果验证了方法的有效性与鲁棒性。算法的实用性需要在视觉系统中得以验证。本文介绍了一种作者参与设计开发的视觉跟踪系统,该系统可移植不同的视觉算法,用于实现运动目标实时检测,并在检测的基础上对指定目标进行自动跟踪。其中,检测环节采用前述提出的基于ICA的方法,跟踪环节采用基于颜色特征的跟踪方法。运行效果显示出该系统能较好的完成目标检测及后续跟踪任务。更多还原

【Abstract】 Target detection based on computer vision is one of the most important research subjects in some fields such as pattern recognition, image processing and so on. Higher requirement has been put forward by the complexity of environment and the run time of algorithm. How to enhance the robustness and environmental adaptability within real-time performance has become the prime focus of algorithm research. Target detection technology can be divided into static target detection and moving target detection. The research contents of this paper contain parameters estimation of static models and detection of moving target area. The research purpose is to improve the robustness and practicality of algorithms.In the process of model parameters estimation, a novel robust estimation algorithm based on bottom up idea is proposed to improve the robustness of ellipse parameter estimation when the environmental information such as scale and quantity of models is unknown. The method uses the process of classification and clustering. Experimental results show that the method has good performance in accuracy and robustness with no prior information.Real-time moving target detection is widely applied in monitoring field. The background difference is fast and easy to implement, but due to the restrictions of threshold, the adaptability is not good in complicated environment. Independent component analysis (ICA) is a kind of signal separation methods by using statistical characteristics of signals. Because the impact of local disturbance is not obvious in treating processing, ICA has gained more and more applications in the field of image processing. In this paper a novel motion detection method based on ICA is proposed. The method has the following advantages: it is self-adaptive, it needs no threshold, it can adapt to complicated environment, and the processing speed meets real-time requirement. Experimental results show the validity and robustness of the algorithm.The practicality of algorithm should be verified in system. This paper describes a visual system that the author participated in designing. The system can transplant different algorithms to implement targets detection and tracking. The system uses the ICA-based detection algorithm and the color-based tracking algorithm. Operation effects show that the system can accomplish the detection and tracking tasks.更多还原