摘要
受光强变化大、亮度低以及车灯光晕等影响,夜间环境下的视频车辆提取相对白天来说较为困难。提出一种联合时空信息的夜间运动车辆提取方法。该方法首先通过帧间差分法提取出运动区域,然后利用均值平滑滤波消除噪声影响并通过大津法(Otsu)自动确定分割阈值,最后通过空间域分析(引入了"D聚类"概念)提取出完整的运动车辆轮廓。实验结果表明,该方法实时性强,鲁棒性高,提取的运动车辆效果好。
Different from daytime condition,vehicle visual detection in evening environment is more difficult owing to the impact of big difference in light intensity,low luminance and halo of car lights.An algorithm of moving vehicle extraction at night is proposed based on spatio-temporal information.The algorithm extracts the moving regions by means of frames difference at first;then it removes noise using mean smooth filtering and determines threshold segmentation automatically with Otsu method;at last,by analyzing spatial information（imported the concept of ＂ distance-clustering＂ ） the holistic profiles of moving cars are extracted.Experimental results showed that the method is real-time and robust,and the effects of the extracted moving vehicles are excellent.
出处
《计算机应用与软件》
CSCD
2010年第10期126-127,198,共3页
Computer Applications and Software
基金
广东省科技厅工业攻关计划项目(2007A010100012)
关键词
运动目标提取
帧差法
均值滤波
区域分割
Moving object extraction
Frame difference
Mean filter
Region segmentation
