BROWSE
Detail
Conference Paper
깊은 나선형 신경망 기반의 시차에 강인한 영상 정합
Cited - time in 
Share 
Share
- Issue Date
- 2020-11
- Citation
- 대한전자공학회 학술 대회 (추계) 2020, pp.301-305
- Publisher
- 대한전자공학회
- Language
- Korean
- Type
- Conference Paper
- Abstract
- Parallax distortion by depth difference occurs frequently when two images are stitched, as objects and backgrounds are usually not located at a uniform distance from the camera. To solve these problems, existing methods tried to obtain multiple homography for each region, then combine them based on certain energy functions. However, these methods are inefficient and complex because they are performed by a series of stages. In this paper, we introduce an end-to-end by a deep learning network for image stitching method, which is robust against distortion by depth difference. In order to train our end-to-end deep convolutional neural network (CNN), we construct a dataset by using CARLA simulator. Our dataset consists of a pair of left and right images with a narrow field of view as inputs, and a center image with a wide field of view. Thus, the proposed network takes left and right images as input and directly generates images with a wide field of view. We show the excellence of the proposed dataset and method through various experiments.
- KSP Keywords
- Convolution neural network(CNN), Deep convolutional neural networks, Deep learning network, End to End(E2E), Field of View(FoV), Stitching method, Wide field, deep learning(DL), depth difference, each region, energy function