Choi, Min-Kook
(Future Automotive Technology Research Center, DGIST, Daegu, Republic of Korea)
,
Lee, Jin-Hee
(Future Automotive Technology Research Center, DGIST, Daegu, Republic of Korea)
,
Jung, Heechul
(Future Automotive Technology Research Center, DGIST, Daegu, Republic of Korea)
,
Tayibnapis, Iman R.
(Future Automotive Technology Research Center, DGIST, Daegu, Republic of Korea)
,
Kown, Soon
(Future Automotive Technology Research Center, DGIST, Daegu, Republic of Korea)
The AR-HUD (Augmented Reality-Head Up Display) overlays the ADAS (Autonomous Driver Assistant System) information to real world objects on the windshield unlike the normal HUD. This projected ADAS information on the windshield usually occurs the irregularity between the objects and the ADAS informat...
The AR-HUD (Augmented Reality-Head Up Display) overlays the ADAS (Autonomous Driver Assistant System) information to real world objects on the windshield unlike the normal HUD. This projected ADAS information on the windshield usually occurs the irregularity between the objects and the ADAS information because of the difference between driver and AR-HUD's view plane. To overcome this problem, we present a simulation framework for the improved AR-HUD in ADAS by means of a homographie registration with inference outputs from a region based deep learning model. In order to build this simulation framework, we set up a simulation test bed to mimic on-road driving environment in a darkroom, and use the inference model based on region-based fully convolutional network to obtain on-road ADAS information. And then we apply the homographie registration method to minimize the irregularity between object and ADAS information in terms of driver's perception. We tested the proposed simulation framework with real world driving recordings, and it showed better display results for improved UI/UX with inference outputs from region-based deep learning model.
The AR-HUD (Augmented Reality-Head Up Display) overlays the ADAS (Autonomous Driver Assistant System) information to real world objects on the windshield unlike the normal HUD. This projected ADAS information on the windshield usually occurs the irregularity between the objects and the ADAS information because of the difference between driver and AR-HUD's view plane. To overcome this problem, we present a simulation framework for the improved AR-HUD in ADAS by means of a homographie registration with inference outputs from a region based deep learning model. In order to build this simulation framework, we set up a simulation test bed to mimic on-road driving environment in a darkroom, and use the inference model based on region-based fully convolutional network to obtain on-road ADAS information. And then we apply the homographie registration method to minimize the irregularity between object and ADAS information in terms of driver's perception. We tested the proposed simulation framework with real world driving recordings, and it showed better display results for improved UI/UX with inference outputs from region-based deep learning model.
※ AI-Helper는 부적절한 답변을 할 수 있습니다.