Vision algorithms for Emerging Applications

新兴应用的视觉算法

• 批准号: RGPIN-2015-04974
• 负责人: Shehata, Mohamed
• 金额: $ 1.6万
• 依托单位: University of British Columbia
• 依托单位国家: 加拿大
• 项目类别: Discovery Grants Program - Individual
• 财政年份: 2020
• 资助国家: 加拿大
• 起止时间: 2020-01-01 至 2021-12-31
• 项目状态: 已结题
• 来源: https://www.nserc-crsng.gc.ca/ase-oro/Details-Detailles_eng.asp?id=708349
• 关键词: Vision; algorithms; Emerging; Applications

Video surveillance represents one of the fastest growing industries in Canada and the world. It is estimated that the revenue will be $15.9 billion in 2014 compared $14.1 billion in 2013. Terrorism and high crime rates in our societies demand more efficient surveillance systems. With the recent advances in imagery sensor technology and aerial vehicles, aerial surveillance has become a very active area, especially with the introduction of Wide Area Motion Imagery (WAMI) in 2007. At the core of these surveillance systems are object detection, tracking, and activity recognition. This research program investigates the design and development of advanced techniques for object detection and tracking in wide area motion imagery (WAMI); specifically, we will investigate the following three research themes: 3D aerial reconstruction, stabilization and object detection, and object tracking. In the area of 3D aerial reconstruction, we will design and develop new algorithms that formulate the problem differently from the traditional disparity based algorithms and use dense matching and global optimization techniques. In the area of stabilization and object detection, we will design and develop a new robust stabilization technique such that we can treat the outliers in the images that did not follow the transformation function from the stabilization as candidate target objects. We will focus on addressing and minimizing the parallax and ghost effects. In the area of object tracking, a recent study in 2014 compared the performance of different trackers in WAMI and concluded that the best performance accuracy achieved was 70%. This is due to high altitude and small object size in cluttered scenes, especially in high density motion areas with large number of objects. We will design new algorithms that represent the objects in higher dimensional feature vectors based on selected filter responses. The proposed algorithms will be analyzed using both analytical and experimental methods. The algorithms will be tested on Columbus Large Image Format (CLIF) and VIVID datasets, and also by using aerial vehicles to demonstrate real-time performance. This research program will help create new innovative aerial surveillance systems that can be used in the industry of Wide Area Arial Surveillance (WAAS), which is one of the emerging technologies that provide surveillance coverage in city-size coverage.  This research program will also provide training to 20-25 undergraduate and graduate students in total helping them build strong backgrounds in advanced topics in image processing, computer vision, digital signal processing, optimization, information theory, and computational complexity analysis. The students will gain hands-on experience by performing experimental work using state-of-the-art vision equipment and software tools and implementing the algorithms on constrained resources embedded boards.

视频监控是加拿大和世界上增长最快的行业之一。据估计，2014年的收入将为159亿美元，而2013年为141亿美元。我们社会中的恐怖主义和高犯罪率需要更有效的监视系统。随着图像传感器技术和飞行器的最新发展，空中监视已成为一个非常活跃的领域，特别是2007年推出的广域运动图像（WAMI）。这些监视系统的核心是目标检测、跟踪和活动识别。 该研究计划研究广域运动图像（WAMI）中目标检测和跟踪的先进技术的设计和开发;具体而言，我们将研究以下三个研究主题：3D空中重建，稳定和目标检测以及目标跟踪。 在3D空中重建领域，我们将设计和开发新的算法，这些算法与传统的基于视差的算法不同，并使用密集匹配和全局优化技术。 在稳定和目标检测领域，我们将设计和开发一种新的鲁棒稳定技术，这样我们就可以将图像中不遵循稳定转换函数的离群值作为候选目标对象。我们将专注于解决和最小化视差和重影效果。 在目标跟踪领域，2014年的一项研究比较了WAMI中不同跟踪器的性能，得出的结论是，最佳性能精度为70%。这是由于在杂乱场景中，特别是在具有大量对象的高密度运动区域中的高海拔和小对象尺寸。我们将设计新的算法，根据选定的过滤器响应，在高维特征向量中表示对象。 所提出的算法将使用分析和实验方法进行分析。这些算法将在哥伦布大图像格式（CLIF）和VIVID数据集上进行测试，并使用飞行器来演示实时性能。 该研究计划将有助于创建新的创新空中监视系统，可用于广域监视（WAAS）行业，这是在城市范围内提供监视覆盖的新兴技术之一。 该研究计划还将为20-25名本科生和研究生提供培训，帮助他们在图像处理，计算机视觉，数字信号处理，优化，信息论和计算复杂性分析等高级主题方面建立强大的背景。学生将通过使用最先进的视觉设备和软件工具进行实验工作并在资源受限的嵌入式板上实现算法来获得实践经验。