模型基计算机视觉中的假设－验证策略

本文提出了一种以层次化假设－验证为识别策略，基于模型的计算机视觉方法（ＭＢＣＶ），这种方法将ＭＢＣＶ与立体视觉结合在一起，互补相益。进而提出了由初始假设、姿态估计、高层验证和低层验证４部分组成的实施方案，本系统具有两个突出的特点：一是系统不苛求低层处理，能一次提取大量信息以支撑高层识别；二是强的系统容错能力，这两个特点大大减轻了低层处理的压力，并且改善了鲁棒性，实验证明，本系统在带噪和遮挡情况下都能正常工作。

Hypothesis-Verification Strategy in Model Based Computer Vision

A model-based recognition methed based on a hierarchical hypothesis-verification(H-V)strategy is presented. A new approach combining the method with stereo vision is presented, which syn-thesises the robustness of H-V strategy and the searching efficiency of stereo vision. A prototype sys-tem is suggested which consists of four parts:initial hypothesis,pose estimation,high-level verifica-tion and low-level verification. Two most noval aspects of the system are the avoidance of the require-ment to extract a lot of useful information during low level preeessing and the strong power to tolerateerrors. Experiments have been conducted to show the working of the approach in the presence of noise with occlusions.