出于工业和医疗等领域大量现实应用的需要,如今三维物体识别已成为一个很活跃的研究领域.一般来说,三维物体识别系统可以通过两个阶段的处理来完成三维物体的识别和定位,首先用传感器获取的场景输入数据来得到场景的表达;然后将它与数据库中存储的物体表达相匹配.为了推动该领域研究进一步发展,因而对近10a年中该识别过程中必须解决的传感器类型、三维物体表达方法和匹配策略等3个方面问题的研究成果进行了综述,对主要方法进行了分类和总结;并提出了一些三维视觉系统中还需要深入研究的问题,包括对所研究物体形状的限制、复杂背景的影响和表达以及识别中的“整体和局部”的矛盾等.

For the requirement of many practical applications in the area of industry and medicine, the research in three-dimensional objects recognition is very active. In general, the dominant paradigm in 3D objects recognition system proposes to achieve recognition and localization of 3D objects from images by a two-stage process: first derive an internal representation of a scene from the sensed input data and then match it against stored representations of objects in the database. This paper presents a comprehensive survey of the achievement of 3D objects recognition system in the recent decade. The three problems are discussed which are the type of sensors used, 3D objects representation and match strategy. Furthermore, the paper classifies and summarizes the primary schemes. At last, we describe some problems requiring a thorough research on three-dimensional vision system, such as the limitation by the class of shapes that can be described in most of the representation schemes, the clustered background influence on the recognition, and the contradiction between global and local in the 3D objects representation and recognition.