项目名称: 基于视频的天气现象自动识别方法研究
项目编号: No.41305138
项目类型: 青年科学基金项目
立项/批准年度: 2014
项目学科: 天文学、地球科学
项目作者: 李骞
作者单位: 中国人民解放军理工大学
项目金额: 24万元
中文摘要: 天气现象是地面气象观测的重要内容,对其自动检测与识别方法研究,已成为大气科学基础研究的热点之一。针对当前天气现象器测能力弱,识别准确性较低等问题,本项目尝试从固定摄像机所捕获的室外可见光监控视频中提取有效视觉信息,实现对晴、阴、雨、雪、雾、霾和尘等常见天气进行分类识别。主要研究内容包括:(1)研究可用于当前天气识别的视觉特征。重点研究各类天气对图像与视频成像的影响,从中提取采样帧图像中对比度、饱和度、色调等静态特征和视频中具有时序属性的自相关函数和运动粒子的形态速度等动态特征。(2)研究基于监督学习的天气现象分类方法。重点研究针对提取的视觉特征,设计和训练合适的序列AdaBoost分类器,实现常见天气现象的分类及识别。(3)研究同一时段内多段视频识别结果融合方法。重点研究基于带权投票机制的识别结果融合,降低因单段视频受环境、时间等因素影响而产生的识别误差。
中文关键词: 天气现象;视频;视觉特征;自动识别;监督学习
英文摘要: The weather phenomena is atmospheric physical process occurring in the ground or above ground which is an important content of the surface meteorological observations. To recognition the present weather automatically is becoming one of research hotspots of surface meteorological observation automation, because it mainly observed by human being at present. To overcome the high cost and low accuracy for the current methods, we try to do some research on the classification of common weather situations by extracting useful visual feature from outdoor surveillance videos captured by the fixed camera. The research topics include: (1) Extracting visual features from the video which can be used to identify the present, especially the static features such as contrast, saturation and hue extracted from the sampling frame images and the temporal features such as the autocorrelations and the appearance of moving particles extracted from the videos. (2) The classification methods based on supervised learning will be considered, and the research will focus on designing and training of a set of sequential AdaBoost classifiers suitable for weather classification using the features extracted above. (3) To reduce the generated recognition error due to the factors such as changes in ambient lighting, time and so on, the problem o
英文关键词: Weather phenomena;Video;Visual Feature;Automatic classification;Supervised Learning