计算机科学
人工智能
职位(财务)
计算机视觉
像面
投影(关系代数)
像素
人工神经网络
失真(音乐)
算法
图像(数学)
财务
计算机网络
经济
放大器
带宽(计算)
作者
Ke Li,Ling Bai,Yinguo Li,Zhongkui Zhou
标识
DOI:10.1109/ismar-adjunct51615.2020.00055
摘要
We propose a distortion correction framework of the AR-HUD virtual image based on a multilayer feedforward neural network model (MFNN) and spatial continuous mapping(SCM). First, we put forward the concept and calculation method of the equivalent plane of the virtual image in the AR-HUD system. Then construct a network structure named MFNN-SCM, and train a network model that can predict the vertex coordinates and pre-distortion map of the equivalent plane of AR-HUD virtual image, and then obtain the eye position of the driver based on training. The network model calculates the virtual image projection mapping relationship under the current eye position of the driver. Finally, the virtual image projection mapping relationship is used to pre-distort the AR-HUD projected image, and the pre-distorted image is projected to improve the AR-HUD imaging effect observed by the driver. In addition, we have embedded the framework into the AR-HUD system of intelligent vehicles and tested it in the real vehicle. The results show that he projected virtual image in this paper has a small relative pixel drift at any eye position. On the premise of ensuring the real-time performance of the algorithm, our method has stronger flexibility, higher accuracy and lower cost than other existing methods.
科研通智能强力驱动
Strongly Powered by AbleSci AI