Multi-Modality-Enriched Video: Potential, Strategies and Applications

视觉传感器无处不在，并已进入许多不同的应用领域。它们以惊人的速度捕获数据，并与其他传感器集成到单个设备中。例如，一个典型的智能手机有一个摄像头（视觉传感器）、一个GPS传感器、一个加速度计、一个陀螺仪和一个磁力计。这些集成传感器产生多模态丰富的视频信号：视频信号以及一个或多个一维或二维非视频信号。非视频信号可以为视频信号处理提供附加的上下文或直接提供侧信息，方便视频信号的处理和通信。例如，运动传感器可以为运动估计和视频压缩提供有用的运动数据。此外，多模态丰富的视频使我们能够开发新的功能，创造新的用户体验。丰富的非视频信号在改善我们基于视频的沟通和决策方式方面具有巨大的潜力。然而，实现这种潜力取决于管理、分析和从这些多模态数据集中提取有用信息的技术的进步。我们利用丰富的视觉传感器获取的不同信号模态之间的相互依赖性是相当直观的。目前，对于在特定的视频处理任务（例如视频压缩、检测、跟踪）中使用多种信号模式是否有任何好处，以及使用哪一种以及如何使用，还没有确定的基本原理。在这项发现基金研究中，为了系统地利用视频压缩、后处理、检测和分类的不同信号模式之间的相互依赖关系，将进行新的调查。将采用一个基于张量的框架，它将表征传感模式之间的相互作用。本文的研究有望为多模态丰富视频的系统分析和处理提供新的方法。该研究将带来更好的视频通信和更好的基于视频的跟踪、识别和分类。从这项研究中产生的新知识将在教育、医疗保健、制造业和汽车工业等对加拿大非常重要的领域找到新技术和新产品的途径。该领域的加拿大公司将受益于新知识，也将受益于参与研究的高素质人才。

Visual sensors are ubiquitous and have found their ways into many different applications. They capturing data at an incredible pace and are being integrated with other sensors into single devices. For example a typical smartphone has a camera (visual sensor), a GPS sensor, an accelerometer, a gyroscope, and a magnetometer. These consolidated sensors generate multi-modality-enriched video signal: video signal as well as one or multiple 1-D or 2-D non-video signals. The non-video signals can provide additional context for video signal processing or directly provide side information to facilitate video signal processing and communication. For example a motion sensor can provide motion data useful for motion estimation and thus video compression. Moreover, multi-modality enriched video enables us to develop new functionalities and create new user experiences. The enriching non-video signals have a tremendous potential for improving the way we communicate and make decision based on video. However, realization of this potential hinges upon advancement in techniques for managing, analyzing, and extracting useful information from these multi-modality data sets. Our utilization of interdependency between different signal modality acquired by enriched visual sensors is rather intuitive. Currently, there is no established rationale for determining if there are any benefits to use more than one signal modality in a particular video processing task (e.g. video compression, detection, tracking) and which one to use and how to use it. In this Discovery Grant research, a new line of inquiry will be pursued in order to systematically utilize the inter dependencies between different signal modalities for video compression, post processing, detection and classification. A tensor based framework will be adopted that will characterize the interaction between sensing modalities. The proposed research is expected to yield new approaches for systematic analysis and processing of multi-modality enriched video. The research will lead to superior video communication and better video-based tracking, recognition and classifications. The new knowledge arising from this research will find its way into new technologies and products in areas of great importance to Canada such as education, healthcare, manufacturing and auto industry. Canadian companies in this area will benefit from the new knowledge and also the highly qualified personnel that will participate in the research.