A Hybrid Temporal-Spatio Fusion Algorithm for Moving Pedestrian Detection in Traffic Scenes

摘要

To accurately extract traffic information, the pedestrian detection algorithm of combining time domain, edge and internal-external marker information is proposed. First, a rough motion mask is extracted based on the temporal change of adjacent multi-frames and Canny edge detection, which effectively solves D-value localization and noise problem. Second, to improve the accuracy of airspace segmentation and eliminate over-segmentation, spatial mask image is received by introducing the technology of quadratic reconstruction, internal-external marker, and watershed transformation. Finally, an accurate pedestrian detection mask is obtained by integrating the rough mask and spatial mask image. The experiment results using Beijing South Station video show that this method can detect the complete target information and obtain the better pedestrian detection results in dynamic traffic scene.