Object Detection and Recognition in Active Video

对象检测和识别定位视觉对象，并识别图像和视频中相应的类别，身份和活动。 该任务是理解图像和视频内容的基础，因此它是计算机视觉和多媒体的中心课题之一。 一般来说，人类拍摄的数字视频是有目的的--针对感兴趣的对象（OOI）。 典型的例子包括拍摄，专业摄像师报道体育赛事，或业余拍摄的旅游景点（例如，建筑物、雕塑/雕像和人群/人的活动）。 努力定位OOI，通过特写镜头或多个视图仔细检查它，如果它在移动，就跟踪它。 我们称这样产生的视频为活动视频。 它的特点是存在显着的OOI，更丰富的背景下产生的有目的的相机运动，它是比在非活动的视频对象为中心。 该研究计划的短期目标是为动态视频中的视觉对象检测和识别这一重要问题寻求新的答案。 特别是，该计划将研究和开发：（a）算法，以探索活动视频中的显着性和注意力模型;（B）功能和描述符，可以更好地捕捉活动上下文;和（c）提升分类器，更具歧视性，鲁棒性和效率。 将在多个应用领域开展研究项目，例如3D Active Video和Faces in Active Video。 我研究的长期目标是推进基于视觉对象的图像和视频处理和分析的知识。 它很大程度上是由来自智能人类视觉系统的线索驱动的。 这项研究的预期影响是双重的：（a）它将产生新的算法和解决方案，以更好地进行视觉对象检测和识别，这是数字视频和多媒体研究社区和行业迫切需要的。 (b)它将直接有助于在顶级会议和期刊上发表学术出版物的形式发现这一领域的知识，并为加拿大培训高素质人才。

Object detection and recognition localize visual objects and recognize their corresponding classes, identities, and activities in images and videos. This task is the foundation of understanding the image and video content, thus it is one of the central topics in Computer Vision and Multimedia. In general, digital video taken by humans is purposive -- aiming for the Object(s) of Interest (OOI). Typical examples include filming, professional video cameramen covering sporting events, or an amateur shooting at a tourist scene (e.g., buildings, sculptures/statues, and activities of the crowd/people). Efforts are made to locate the OOI, examine it closely by close-ups or multiple views, and follow it if it is moving. We call the video thus produced Active Video. It is characterized by the presence of salient OOI, richer context yielded by the purposive camera movements, and it is more object-centered than in non-active videos. The short-term objectives of the proposed research program are to seek new answers to the important problem of visual object detection and recognition in Active Video. In particular, the program will research and develop: (a) Algorithms to explore saliency and attention models in Active Video; (b) Features and descriptors that can better capture the Active Context; and (c) Boosted classifiers that are more discriminative, robust, and efficient. Research projects in several application areas will be conducted including, for example, 3D Active Video and Faces in Active Video. The long-term objective of my research is to advance the knowledge in visual object-based image and video processing and analysis. It is largely motivated by cues from the intelligent human vision system. The anticipated impact of this research is two-fold: (a) It will generate new algorithms and solutions for better visual object detection and recognition which are urgently needed by the digital video and multimedia research community and industries. (b) It will directly contribute to the discovery of knowledge in this area in the form of academic publications in top conferences and journals, and to the training of Highly Qualified Personnel (HQP) for Canada.