基于时空融合的交通场景中运动行人检测算法

将视频监控技术应用于交通信息获取与处理，解决行人安全问题，已成为智能交通系统中的一项关键技术。为精准地提取交通场景语义信息，提出了融合时域和空域信息的行人检测算法。通过相邻多帧时域变化和Canny边缘检测算法得到初始检测掩模图像，解决差值局部化和噪声问题；提取空域信息时，引入二次重构和内外标记技术对梯度图像进行修正和分水岭变换，得到空域掩模图像，有效地提高空域分割精度，消除过分割现象；最后将两者结果进行融合和形态学修正，提取出交通场景图像中的运动行人。以北京南站为例进行实验分析，结果表明，该方法能快速准确地检测到较完整的目标信息，在动态场景的行人检测取得了较好的效果。

Pedestrian detection based on temporal-spatio fusion information in traffic scenes

The video surveillance technology is a key part in the intelligent transportation for traffic information acquiring and safe-ty problem solving.A pedestrian detection algorithm,combining time-domain and spatio-domain,is proposed to accurately ex-tract traffic information.To solve the difference localization and noise problem,a rough motion mask is obtained based on the temporal change of adj acent frames and the canny edge detection.Spatial mask image is received by introducing technologies of quadratic reconstruction,internal-external marker and watershed transformation.It improves airspace segmentation accuracy and eliminate over-segmentation.Finally,an accurate motion pedestrian mask is obtained by integrating the rough mask and spatial mask image.The experiment based on Beij ing South Station video shows that target information could be completely and exactly detected,and the pedestrian detection results in dynamic traffic scene are satisfactory.