Identification, segmentation and tracking of objects in colour images and sequences

识别、分割和跟踪彩色图像和序列中的对象

• 批准号: 311626-2006
• 负责人: Androutsos, Dimitrious
• 金额: $ 0.95万
• 依托单位: Ryerson University
• 依托单位国家: 加拿大
• 项目类别: Discovery Grants Program - Individual
• 财政年份: 2008
• 资助国家: 加拿大
• 起止时间: 2008-01-01 至 2009-12-31
• 项目状态: 已结题
• 来源: https://www.nserc-crsng.gc.ca/ase-oro/Details-Detailles_eng.asp?id=396420
• 关键词: Identification; segmentation; tracking; objects; colour

Identifying and segmenting of specific complex objects in general images is still an open problem and an important research area in image processing and analysis, due mostly to the importance of such a task in many applications. Moreover, tracking such objects in a video sequence is of increasingly greater significance, a fact intensified by the fast prevalence and penetration of digital video technology. Research from this proposal will focus on identifying specific image objects in images and video and track them as they move.  By improving the computational methods of such a cognitive task can advance application that are based on visual identification and extraction. Areas such as object-based compression, video surveillance systems, interactive television and rotoscoping for film editing are a few imporant areas that stand to benefit. Research will follow three stages: i) identify specific image objects in general colour images using a modified spatial colour statistical measure;   ii) incorporate this measure and vector colour information into the energy functional, the underlying mathematical mechanism of active contours (snakes), to perform object segmentation; iii) extend concepts in i) and ii) into the video domain and track objects as they move from frame to frame.  This investigation will involve the development and testing of two new experimental concepts, which will: a) consider an image objects path from frame to frame as a 3-D volumetric solid and use this model to improve active contour tracking accuracy; b) apply active contours across frames to localize and lock onto an objects trajectory path. The goal of this proposal is to develop and advance methodologies and tools for identifying specific objects in images and video to then accurately tracking them in a video scene using novel methodologies based on active contours.

一般图像中特定复杂目标的识别和分割仍然是图像处理和分析中的一个开放问题和重要研究领域，这主要是因为该任务在许多应用中都很重要。此外，在视频序列中跟踪这些物体越来越重要，数字视频技术的迅速普及和渗透加剧了这一事实。该提案的研究将侧重于识别图像和视频中的特定图像对象，并在它们移动时跟踪它们。通过改进这种认知任务的计算方法，可以推进基于视觉识别和提取的应用。诸如基于对象的压缩、视频监控系统、交互式电视和用于电影编辑的旋转显微镜等领域是一些重要的受益领域。研究将分为三个阶段：i)使用改进的空间色彩统计度量来识别一般彩色图像中的特定图像对象；Ii)将该度量和矢量颜色信息合并到能量函数中，即活动轮廓（蛇）的潜在数学机制，以执行对象分割；Iii)将i)和ii)中的概念扩展到视频域，并在物体从一帧移动到另一帧时跟踪它们。本研究将涉及两个新的实验概念的开发和测试，它们将：a)将图像对象从一帧到另一帧的路径视为三维体积实体，并使用该模型来提高主动轮廓跟踪精度；B)跨帧应用活动轮廓来定位和锁定对象的轨迹路径。本提案的目标是开发和推进用于识别图像和视频中的特定对象的方法和工具，然后使用基于活动轮廓的新方法在视频场景中准确跟踪它们。