Cooperation-Oriented Object Recognition in Cloeed Space Influenoed by Time-Variant Lighting

时变光照影响的封闭空间中面向合作的目标识别

• 批准号: 11680407
• 负责人: MORI Katsumi
• 金额: $ 2.18万
• 依托单位: Fukuyama University
• 依托单位国家: 日本
• 项目类别: Grant-in-Aid for Scientific Research (C)
• 财政年份: 1999
• 资助国家: 日本
• 起止时间: 1999 至 2001
• 项目状态: 已结题
• 来源: https://kaken.nii.ac.jp/grant/KAKENHI-PROJECT-11680407/
• 关键词: object recognition; lighting recognition; cooperative processing; elliptic contour; multi-layered neural nerwork; predicted scene CG image; color compensation; region segmentation; 光環境認識; シーン予測; 画像の圧縮; CG画像; 回転軸対称物体; 楕円弧

For the purpose of development of the assistant robot for handicapped and aged persons, we have developed some recognition methods for objects in artificial and natural lightings.[1] Recognition of environmental lighting :(1) Color compensation : We have developed a method for the detection of the changing property of the environmental lighting by using neural networks, and shown that this method could be applied to the compensation of real scene images changed by artificial and natural lightings.(2) Segmentation and compression for real scene images : By using multiple neural networks with the cooperative function, we have developed segmentation and compression methods for real scene images changed by artificial and natural lightings. Also we have shown that their methods could detect the changing property of the environmental lighting.[2] Object recognition based on the shape analysis :(1) Recognition of circular objects : We have made some algebraic formulas for representing the geometrical properties of ellipses that they have diameters and conjugate diameters, and shown that circular objects could be recognized accurately in short processing time by using the formulas.(2) Contour segmentation : Because most of contours obtained from real images become links of edges of different objects, we have proposed a method to segment the contours into individual object sections by estimating the curvatures of neighboring contours.[3] Cooperative recognition of environments and objects : We have proposed a scene predicting CG-system for supporting the cooperative recognition of environmental lighting and objects, and shown experimentally that the system is very effective to recognize objects and environmental lightings in real scene. The system drastically cuts the processing cost of recognizing objects in closed space influenced by time-variant lighting.

为了开发残疾人和老年人的辅助机器人，我们开发了一些在人工和自然光下物体识别的方法。[1]环境光照识别：（1）颜色补偿：提出了一种利用神经网络检测环境光照变化特性的方法，并证明了该方法可以应用于人工和自然光照改变的真实的场景图像的颜色补偿。(2)真实的场景图像的分割与压缩：利用具有协同功能的多个神经网络，研究了人工和自然光照变化下的真实的场景图像的分割与压缩方法。此外，我们已经表明，他们的方法可以检测环境照明的变化特性。[2]基于形状分析的目标识别：（1）圆形目标的识别：我们给出了椭圆具有直径和共轭直径的几何性质的代数公式，并证明了利用这些公式可以在较短的处理时间内准确地识别圆形目标。(2)轮廓分割：由于从真实的图像中获取的轮廓大多数是不同目标的边缘的连接点，因此我们提出了一种通过估计相邻轮廓的曲率来分割轮廓的方法。[3]环境和物体的协同识别：我们提出了一个场景预测CG系统，支持环境照明和物体的协同识别，实验表明，该系统是非常有效的识别对象和环境照明在真实的场景。该系统大大降低了在受时变照明影响的封闭空间中识别物体的处理成本。

项目成果

Eiji, Watanabe: "A cooperative learning algonithm for a group of multi-layered neural networks (in Japanese)"Proc. Of 9-th SICE Chugoku Branch Annual Conference. 208-209 (2000)

Eiji, Watanabe：“一组多层神经网络的协作学习算法（日语）”Proc。

渡邊 栄治: "階層型ニューラルネットワークにおける荷重係数値の量子化に基づいた学習データの選択と再学習法"電子情報通信学会技術研究報告. NC2000-91. 45-52 (2001)

Eiji Watanabe：“分层神经网络中基于加权系数值量化的训练数据选择和重新学习方法”IEICE NC2000-91（2001）。

Eiji Watanabe, Katsumi Mori: "A Compression Method for Color Images by Using Multi-Layered Neural Networks"Systems and Computers in Japan. (発表予定).

Eiji Watanabe、Katsumi Mori：“使用多层神经网络的彩色图像压缩方法”日本系统和计算机。

片桐 重和: "物体の重なりを考慮した輪郭線分割法の検討"平成12年度電気・情報関連学会中国支部連合大会(072219). 188-189 (2000)

Shigekazu Katagiri：“考虑重叠对象的轮廓划分方法的研究”2000年中国电气与信息相关学会分会会议（072219）188-189（2000）。

Eiji, Watanabe: "A Distibuted Learning Algorithm for Multi-Layered Neural Networks and Its Applications to Pattern Classification Problems.(in Japanese)"Technical Report of IEICE, Nc2001-63. 1-8 (2001)

Eiji, Watanabe：“多层神经网络的分布式学习算法及其在模式分类问题中的应用。（日语）”IEICE 技术报告，Nc2001-63。