摘要
文章提出虚拟传感器的概念,针对城市快速路视频监控系统,改革单一功能摄像机,使其成为具人工智能的新型视频传感器,完成道路车辆跟踪及异常行为检测。异常检测算法运用带有时间和空间信息的车辆轨迹对自组织神经网络进行训练,获得神经网络参数后利用概率模型对实时车辆轨迹进行异常提取。该文所提算法能在嵌入式DM642视频处理平台上有效运行,能够提取诸如超低(高)速行驶、违章停车、违规掉头等异常行为,具有低运算量及较好的鲁棒性。
A novel conception of virtual sensor is propesed for video based urban expressway monitoring system. Compared with traditional cameras,the new intelligent video sensor can carry out vehicle tracking and abnormal behavior detection. Abnormal detection algorithm finishes abnormal trajectories detection by probability model after training of self-organizing neural network using spatiotemporal trajectories data. The proposed algorithms are effectively running on the embedded DM642 video processing platform and feasible for robust detection of extra low or high speeding vehicles,illegal parking and illegal turning with low computational cost.
出处
《现代交通技术》
2012年第1期60-63,共4页
Modern Transportation Technology
基金
"十一五"国家科技支撑计划项目(项目编号:2009BAG13A04)
江苏省自然科学基金(项目编号:BK2010239)
江苏省交通科学研究资助项目(项目编号:08X09)
关键词
视频监控
智能摄像机
行为理解
自组织神经网络
video surveillance
intelligent camera
behavior understanding
self-organizing neural network
