车辆行为理解中的目标特征抽取理论研究

项目名称： 车辆行为理解中的目标特征抽取理论研究

项目编号： No.61203246

项目类型： 青年科学基金项目

立项/批准年度： 2013

项目学科： 自动化学科

项目作者： 孙涵

作者单位： 南京航空航天大学

项目金额： 24万元

中文摘要： 智能交通系统中的高清图像视频智能分析技术是国家物联网"十二五"发展规划中的重要内容，具有十分重要的研究价值和应用前景。车辆行为理解中的目标特征抽取又是该领域研究的基础和关键。城市道路环境的复杂性、全天候光照和天气影响、摄像机的特殊视角、高清图像信息量巨大和嵌入式设备运算能力有限的矛盾等都显著影响车辆目标特征抽取的实时性和可靠性。现有方法未能同时考虑上述诸因素，故难以满足实际需求。本课题旨在已有工作基础上，发展出能够综合利用多尺度、兴趣区域、关系稳定性的角点、直线段等特征抽取方法和在运动区域检测基础上的目标特征组合和度量方法，形成车辆目标的显著特征描述子，以克服现有方法不足。整个工作围绕建模、算法设计和实现、理论分析和实验对比等方面展开。此外，本项目的研究不仅能解决现实应用需求，而且能为特征抽取理论中的特征尺度、鲁棒性、特征组合和度量等理论研究探索新思路。

中文关键词： 智能交通系统；特征描述子；空间分布约束；运动目标检测与跟踪；图像质量评价

英文摘要： Intelligent high-definition (HD) image and video analysis in the Intelligent Transportation Systems (ITS) is one of the important contents in the National '12th Five-Year Development' Plan of the Internet of Things (IOT). And it also has very important research value and broad application prospect. Object feature extraction in vehicle behavior understanding is the foundation and the key research in this field. There are many factors significantly affecting the ability of vehicle's feature extraction, including the complexity of the urban road environment, the affection of the varied illumination and weather, the special pose of the camera, the contradiction between the huge data in HD image and the limited computing power of embedded device, and so on. Because the existing methods do not consider all these factors at the same time, so they are difficult to meet the demand. Based on our previous work, we plan to develop new corner points and line detection method using such concepts as multi-scale, region of interest (ROI), and the relation stable-state. We also plan to develop the method of feature combination and measure based on the results of moving object region detection. These will result in the robust feature descriptor for vehicle and will overcome the shortcomings of existing methods. The whole work wil

英文关键词： intelligent transportation system；feature descriptor；spatial distribution；moving object detection and tracking；image quality assessment