项目名称: 基于高速视觉的实验动物多行为实时检出方法研究
项目编号: No.61201400
项目类型: 青年科学基金项目
立项/批准年度: 2013
项目学科: 电子学与信息系统
项目作者: 聂余满
作者单位: 中国科学院合肥物质科学研究院
项目金额: 24万元
中文摘要: 实验动物的行为分析广泛应用于新药开发和生命科学等领域,传统的人工分析和自动检出方法及设备,通常只能离线检出特定实验中的少量慢速行为,面对多实验环境下多行为实时分析的适应性还有待提高。本研究针对实验动物行动快速、形状多变的运动特点,通过高速视觉获取数百帧每秒的高时空分辨率实时图像,结合硬件处理算法要求,提出基于仿射变换的平移旋转不变特征快速提取算法,以有效避免形状匹配等方法中的误识别问题,为多行为的量化定义和实时检出提供依据;同时,基于高速图像序列分析行为间的细微差别,建立动态行为的多参数描述模型,为发现新型行为和新的时间-行为结构,尤其是交互式社会行为实验设计和分析提供研究支持。本研究所开展的行为特征提取和检出方法将是分析柔软变形体运动特征的重要手段之一,将对定量研究实验动物行为,筛选和验证新的药理病理模型具有切实的促进作用。
中文关键词: 行为分析;高速视觉;行为检出;柔软变形体;人体运动
英文摘要: Behaviors of laboratory animals have been used widely in the fields of new drugs development and life science. However, the traditional manual analysis or automatic detection methods and devices could only detect a few slow behaviors offline, and could not adapt to the multi-environment and multi-behavior online analysis. This research focuses on the features of quick motion and easy-deformable shape, and real-timely captured the hundreds frames per second and high resolution images by high-speed vision system. We propose the quick algorithm of abstracting the shift and rotate invariant features based on affine transform. It will avoid the recognition mistakes in the methods such as shape matching, and provide the basis for quantitative definition and real-time detection. Also, multi-parameter model based the fine difference abstracted from the high-frame-rate images will be build to explore the new behaviors and the new time-behavioral structures, especially for support of the design and analysis of the interactive social behavior experiments. The feature abstraction and behavior detection methods which proposed in this research will be an important approach to solve the motion analysis for the soft and deformable objects. It will also practically promote the quantitative researches of laboratory animals, and t
英文关键词: behaviors analysis;high-speed vision;behavior detection;flexible deformation;human motion