Vehicle Detection Using Aerial Images in Disaster Situations
Vehicle Detection Using Aerial Images in Disaster Situations
复制标题
DOI：
10.1007/978-3-319-99834-3_25
复制
发表时间：
2018-08
期刊：
影响因子：
--
通讯作者：
Ayane Makiuchi;H. Saji
中科院分区：
文献类型：
--
作者：
Ayane Makiuchi;H. Saji
在灾害情况下，需要迅速确定交通拥堵和废弃车辆的位置，以找到交通路线，以便高效地开展救援活动。然而，识别车辆并估计其位置需要时间。我们研究的目的是在发生灾难的情况下快速检测车辆及其位置。提出了一种利用直升机航拍图像和道路地图进行车辆识别和位置估计的方法。虽然这样的图像可以在当地拍摄，但当建筑物反射到道路上时，就会出现遮挡。对于车辆识别，我们使用阴影校正、机器学习去除沥青和形状分析。此外，我们拆除建筑物以解决遮挡问题。首先，通过阴影校正来调整航拍图像的颜色。然后，我们去除道路上的沥青和建筑物区域，并利用它们的形状特征来提取车辆区域。为了估计车辆的位置，我们通过投影变换将道路地图投影到航空图像上。在车辆识别之前，通过投影从航拍图像中提取道路区域，从而提高了识别的效率。使用我们的方法，我们成功地检测到了大多数车辆，因为它们与沥青颜色不同。此外，我们还在路线图上标出了车辆的位置。因此，我们论证了在受灾情况下从航空图像中快速检测车辆的可能性。
In disaster situations, it is necessary to rapidly determine the locations of traffic jams and abandoned vehicles to find traffic routes so that rescue activities can be carried out efficiently. However, it takes time to recognize vehicles and estimate their positions. The purpose of our study is to rapidly detect vehicles and their positions in the case of disasters. We propose a method of vehicle recognition and position estimation using an aerial image taken from a helicopter and a road map. Although such images can be taken locally, occlusion occurs when buildings are reflected on the road. For vehicle recognition, we use shadow correction, asphalt removal by machine learning, and shape analysis. In addition, we remove buildings to solve the problem of occlusion. First, we adjust the color of the aerial image by shadow correction. Then, we remove areas of asphalt and buildings on the road, and we extract vehicle areas by using their shape features. To estimate the positions of vehicles, we project the road map on the aerial image by a projective transformation. We extract the road area from the aerial image by the projection before vehicle recognition, thus increasing the efficiency of the process. Using our method, we successfully detected most vehicles with owing to their different colors from the asphalt. Furthermore, we marked the positions of the vehicles on the road map. We thus demonstrated the possibility of the rapid detection of vehicles from aerial images in disaster situations.