CAREER: Observer Design for Intelligent Visual Tracking

提案编号：0846750该奖项是根据2009年《美国复苏和再投资法案》(公法111-5)资助的。该提案中描述的项目旨在提高用于闭环系统的计算机视觉算法的稳定性和健壮性。奋进的研究检验了控制理论在增强闭环系统计算机视觉算法方面可能发挥的作用。随着视觉越来越多地被用于自动化过程，理解控制和计算机视觉之间的相互作用的概念正在变得突出。智能优点：作为传感器，成像系统可以通过实际的传感过程或通过成像场景的几何形状(例如，通过杂乱、遮挡、可变假设等)来处理噪声。因此，视觉任务可以解释为在存在噪声和不确定性的情况下的信号处理任务。通过为寻求估计目标位置和目标边界的视觉跟踪系统推导观测器，PI建议系统地推导出高度稳定和稳健的视觉跟踪算法。Endeavour的研究包括定义概率形状状态和相关的测量模型；推导动态对象和环境的全状态观测器；以及使用机器学习对测量模型施加软和硬几何约束。主要的困难在于视野的丰富性和所有视觉传感器固有的噪声，以及当目标本身既灵活又被成像噪声破坏时对测量模型施加约束的脆弱性。广泛影响：所提出的工作的应用将减少许多单调、重复或时间密集型任务所涉及的人工工作量。工作将具体由两个应用领域推动：用于机场地面作业、建筑工地作业、露天采矿作业安全分析的劳动力跟踪，以及微观蜂窝跟踪和分析。这两个应用领域都关系到民众的健康和福祉，都有拯救生命的潜力，并将产生积极的经济影响。

Proposal Number: 0846750This award is funded under the American Recovery and Reinvestment Act of 2009 (Public Law 111-5).The project described in this proposal seeks to improve the stability and robustness of computer vision algorithms used in closed-loop systems. The research Endeavour examines the role control theory may play in enhancing closed-loop computer vision algorithms. The concept of understanding the interplay between control and computer vision is achieving prominence as vision is increasingly sought for automating processes.Intellectual Merit: As a sensor, the imaging system can be wrought with noise, either through the actual sensing process or through the geometry of the imaged scene (e.g., through clutter, occlusions, variable hypotheses, etc.). Thus, the vision task can be interpreted as a signal processing task in the presence of noise and uncertainty. By deriving observers for visual tracking systems that seek to estimate both target location and target boundary, the PI proposes to systematically derive a highly stable and robust visual tracking algorithm. The research Endeavour involves the definition of a probabilistic shape state and associated measurement models; the derivation of a full state observer for dynamic objects and environments; and the use of machine learning to impose soft and hard geometric constraints on the measurement model. The main difficulties lie in the richness of the visual field coupled with the noise inherent to all visual sensors, and the fragility of imposing constraints on the measurement model when the target itself is both flexible and corrupted by imaging noise.Broader Impact:Application of the proposed work will reduce the amount of human labor involved in many monotonous, repetitive, or time-intensive tasks. Efforts will be specifically driven by two application areas: workforce tracking for safety analysis of airport ground operations, construction site operations, surface mining operations, and microscopic cellular tracking and analysis. Both of these application areas are related to the health and well-being of the populace, have the potential to save lives, and will have a positive economic impact.