摘要
针对行人运动轨迹再现及表达的问题,结合视频检测技术,提出一种结合时空上下文信息的行人运动特征提取和轨迹跟踪算法,并在视频标定计算中引入中心偏移量,改进了共线模型中不考虑镜头畸变的缺点,提高了标定精度.利用该算法对实际视频进行处理,获取多组行人运动参数和轨迹曲线,再将该算法提取的数据与共线标定方法的计算数据及人工调查得到的真实数据进行对比,验证所提算法的准确性.最后定量分析了几种行人过街行为,对过街行为相应的轨迹特征状态进行了表达,为行人交通组织和控制提供支持.
According to the trajectory reappearance and feature expression of pedestrian movement,based on video processing,a pedestrian feature extraction and tracking algorithm combining with temporal and spatial context information is proposed.Then the center offset is introduced into the calibration to overcome the shortcomings of the collinear model which does not consider the lens distortion,thus the calibration precision is improved.This algorithm is used to obtain some groups of pedestrian movement parameters and trajectory curve on real video processing.Then the data extracted by this algorithm is compared with both the data calculated through the collinear calibration method and the real data from artificial investigation,in order to verify the accuracy of this algorithm.Finally,several pedestrian crossing behaviors are quantitatively analyzed and the corresponding feature statuses are expressed,which can support the organization and control of pedestrian traffic.
出处
《东南大学学报（自然科学版）》
EI
CAS
CSCD
北大核心
2012年第6期1233-1237,共5页
Journal of Southeast University：Natural Science Edition
基金
国家自然科学基金资助项目(51108208)
中国博士后科学基金面上资助项目(20110491307)
吉林大学科学前沿与交叉学科创新资助项目(201103146)
关键词
智能交通
视频检测
行人行为
轨迹再现
特征表达
intelligent transportation
video detection
pedestrian behavior
trajectory reappearance
feature expression
