A Study on Optimal Design of Hough Transform Algorithm for Real Time Vision

实时视觉Hough变换算法的优化设计研究

基本信息

批准号：
12680356
负责人：
MURAKAMI Kazuhito
金额：
$ 2.24万
依托单位：
Aichi Prefectural University
依托单位国家：
日本
项目类别：
Grant-in-Aid for Scientific Research (C)
财政年份：
2000
资助国家：
日本
起止时间：
2000 至 2001
项目状态：
已结题

项目摘要

We promoted our research to realize a real time mechanism to extract figure patterns compounded of line segments. We utilized the Hough transform as the basic feature extraction method. We examined and evaluated from the view points of the processing cost for straight line extraction, the relation between a compound figure pattern and its feature, real time processing, applicability to the image sequences, and we clarified the following results.In order to realize real time processing, we reduced the computation cost by applying the Hough transform not to the whole image but to the partial image and by combining the respective results. We constructed an optimal procedure for the setting of Hough transform parameters from the view point of the relation between the number of space division and the resolution of line segment and the computation cost. From the view point of the extension of the feature extraction mechanism, we improved the feature extraction algorithms for the practical use. As a meta-mechanism of line segment extraction, we introduced the center-fine/peripheral-coarse model which simulates the human vision to search the interesting point or area, and realized the global feature extraction mechanism. Furthermore, we investigated the description of digital line segment, and proposed a new representation method and an algorithm using chain code description. To confirm the effectivity and the applicability of the proposed methods and algorithms to real time vision, we implemented these algorithms to an autonomous robot system for soccer game and experimented, and evaluated under a real and practical condition.

我们促进了我们的研究，以实现一种实时机制，以提取具有线段的复合的图形模式。我们将霍夫变换用作基本特征提取方法。我们从直线提取的处理成本的观点，复合图及其特征，实时处理，对图像序列的适用性之间的关系进行了检查和评估，我们澄清了以下结果。在为了实现实时处理的情况下，我们通过将hough不适用于整个图像，而是通过对部分图像和部分组合来降低计算成本，从而减少了计算成本。我们从空间划分数量与线段分辨率与计算成本之间的关系的角度来构建了一个最佳的过程，用于设置Hough变换参数。从特征提取机制的扩展的角度来看，我们改进了特征提取算法用于实际使用。作为线段提取的元机制，我们介绍了中心及外围/外围 - 核模型，该模型模拟了人类的视野以搜索有趣的点或区域，并实现了全局特征提取机制。此外，我们研究了数字线段的描述，并使用链代码描述提出了一种新的表示方法和算法。为了确认所提出的方法和算法对实时视觉的有效性和适用性，我们将这些算法实施到用于足球游戏的自主机器人系统中，并进行了实验，并在真实和实用的条件下进行了评估。