基于D-LinkNet与形态学计算的高分遥感图像车流量监测

Traffic Flow Monitoring of High-resolution Remote Sensing Images Based on D-LinkNet and Morphological Computing

针对城市交通路面车流量监测实时性和准确性高的需求,设计了一种利用高分遥感图像进行道路网自动提取和车辆自动监测的处理方法。综合利用D-LinkNet和形态学计算实现道路区域的二值化语义分割及连贯性、边缘扩展以及平滑性优化,同时将道路信息作为掩码并外溢后进行车辆目标检测,有效避免非道路区域车辆目标干扰。提出采用热力图的形式改进车流量监测方法,可以更直观显示道路拥挤程度。综合利用现有数据集对所提方法进行评价,车辆检测的平均精度达91.7%,道路提取平均交并比达85.3%,可以实现道路车流量的有效监测。

According to the requirement of high real-time and accuracy for urban traffic pavement traffic flow monitoring,a processing method for automatic extraction of road network and automatic vehicle monitoring using high-resolution remote sensing images is designed.The D-LinkNet and morphological computing are comprehensively used to realize the binary semantic segmentation and coherence optimization,edge expansion and smoothness optimization of the road area,and the vehicle target detection is carried out after the road information is used as a mask and spilled over,so as to effectively avoid the interference of vehicle targets in non-road areas.It is proposed to improve the traffic flow monitoring method by using heat map,which can show the road congestion degree more intuitively.Through using the existing data set,the average accuracy of vehicle detection is 91.7%,and the average intersection ratio is 85.3%.The results show that the presented method can realize the effective monitoring of road traffic flow.