Object class recognition in 3D image data

开发一种解决物体识别问题的解决方案，自动识别和定位场景中的物品，将实现计算机视觉的基本目标之一，使加拿大处于革命性技术的前沿。物体识别是一个长期存在的难题，近年来取得了很大的进展，导致了快速有效的原型。与此同时，识别方法在商业系统中的应用也受到了推动，最初是在相当有限的领域，如光学字符识别、车牌读取器、游戏、面部识别和生物识别，然后越来越多地应用于更普遍的场景。本研究的目的是利用三维距离数据，在一般场景中开发有效的目标类识别方法。有效的对象类别识别方法是实现更自主的智能系统必不可少的组成部分。该领域的未来发展将把这项技术提升到更高的实用水平，使新一代智能机器人系统能够在人类环境中更灵活地互动。社会将受益于制造业应用，如灵活的自动化，但也受益于广泛部署到日常应用，如安全和监控，医疗保健和服务机器人。

Developing a solution to the problem of object recognition which automatically identifies and locates items within a scene would achieve one of the fundamental goals of computer vision, placing Canada at the forefront of revolutionary technology. Object recognition is a longstanding and difficult problem, and there has been much progress in recent years, which has led to fast and effective prototypes. At the same time, there is a pull toward the application of recognition methods in commercial systems, initially in fairly constrained domains such as optical character recognition, licence plate readers, gaming, face recognition and biometrics, and increasingly in more general scenarios. The objective of the proposed research is to develop effective object class recognition methods, in general scenes, using 3D range data. Effective object class recognition methods are the essential missing component to the advent of more autonomous intelligent systems. The coming advances to the field will elevate this technology to a high level of utility, enabling a new generation of intelligent robotic systems that can interact more flexibly in human environments. Society will benefit from manufacturing applications, such as flexible automation, but also from widespread deployment into everyday applications such as security and surveillance, and health-care and service robots.