Deep Neural Networks for Road Sign Detection and Embedded Modeling Using Oblique Aerial Images

Oblique photogrammetry-based three-dimensional (3D) urban models are widely used for smart cities. In 3D models, road signs small but provide valuable information navigation. However, due to the problems of sliced shape features, blurred texture and high incline angles, cannot be fully reconstructed in oblique photogrammetry, even with state-of-the-art algorithms. The poor reconstruction commonly leads less informative guidance unsatisfactory visual appearance. this paper, we present a pipeline embedding sign based on deep convolutional neural networks (CNNs). First, an end-to-end balanced-learning framework object detection that takes advantage region-based CNN data synthesis strategy. Second, under geometric constraints placed by bounding boxes, use scale-invariant feature transform (SIFT) extract corresponding points signs. Third, obtain coarse location single triangulating refine via outlier removal. Least-squares fitting is then applied refined point cloud fit plane orientation prediction. Finally, replace computer-aided design scene predicted orientation. experimental results show proposed method achieves mAP produces visually plausible embedded results, which demonstrates its effectiveness modeling reconstruction.