基于可视范围的图像检索方法

图像是视觉所及环境的自然再现,具有表达直观、内容丰富的特性,其已成为GIS空间数据采集与应用中一种重要的可视数据源。随着硬件终端的高速发展,集成的传感器愈加丰富,多源传感器信息融合已成为移动GIS数据釆集的发展趋势。针对移动终端获取方位角偏差问题,本文重点阐述了基于移动终端的重力传感器与磁力传感器获取镜头拍摄朝向的算法,并基于EXIF图像格式实现了图像元数据、定位信息、方位信息一体化采集生成可定位图像。针对现有可定位图像以点代面建模方式的不足,本文以镜头视野范围AOV（Angle of View）模型为基础,提出了以图像元数据、位置信息以及方位信息构建图像可视域的方法,并采用多级网格方式对图像可视域面状特征建立空间索引,实现基于可视域的图像检索。在此基础上,针对福建省野外遥感核查采集的8022幅可定位图像,采用点建模及可视域建模2种方式建立检索图像集。从检索速度、检索结果2个方面进行对比分析,表明基于可视域图像检索方法不仅能检索出拍摄位置在检索区域内的图像,还可检索出拍摄位置在检索区域外但实际拍摄场景在检索范围内的图像。同时,通过多级空间网格索引可提升6.22-8.64倍检索速度。

A New Image Retrieval Method Based on Image Viewshed

Image is a natural representation of the real world, which has the characteristics of intuitionistic expression and rich content. It has become an important visual data source in GIS data acquisition and application. With the rapid development of hardware terminal, the sensors integrated in mobile terminal become more abundant, and multi-sensor data fusion has become a trend in mobile GIS data collection. Aiming at the problem of obtaining azimuth deviation of mobile terminal, this paper emphasized on the acquisition algorithm of camera shooting directions, which was based on a gravity sensor and a magnetic sensor embedded in mobile terminals. Based on the EXIF format, locatable images were generated by the integrated collection of image metadata, location and direction. To overcome the disadvantages of existing modeling method of taking a surface as a point for locatable image, this paper proposed a method to build image viewshed by using image metadata, location and direction based on the model of lens’ angle of view. Multi-level grid was used to build the spatial index for the polygon feature of the image viewshed. Based on this method, a case study was conducted in Fujian province. In the experiment, 8022 locatable images were collected by field verification of remote sensing. Point based modeling and polygon based modeling were utilized to build the retrieved image set. Retrieval speed and results were compared and analyzed. It showed that the image retrieval results could be more accurate based on image viewshed modeling. It could retrieve images of which the actual shooting position located inside and outside of the retrieval area. Meanwhile, the image retrieval speed could increase 6.22 to 8.64 times by using the multi-level grid spatial index.