R-CNN BASED OBJECT DETECTION FROM MMS IMAGERY FOR GENERATION OF ROAD ORTHOPHOTOS

摘要

In recent years, extensive researches have been conducted to automatically generate high-accuracy and high-precision road orthophotos using images and laser point cloud data acquired from Mobile Mapping System (MMS). However, non-road objects such as vehicles, bicycles and pedestrians etc., are necessary to be masked from MMS images, before texturing process in generation of road orthophoto. Hence, we have developed a vehicle detection model based on faster Regions with Convolutional Neural Network (R-CNN) for detecting the vehicle regions from MMS images automatically. The experimental results show that our vehicle detection model identifies the regions of vehicles from MMS images accurately and robustly, and erroneous vehicle textures are efficiently removed from road orthophotos by using the detection results in texturing process.