[4I2-J-2] Machine learning: uncertainty of targets

[4I2-J-2-04] Monocular depth estimation using overlap information

〇michiru takamine¹, Satoshi Endo¹, Jakub Kolodziejczyk², Taiki Nishime² (1. University of the ryukyus, 2. LiLz Inc.)

Keywords:Depth estimation, Machine learning, Object detection

Depth estimation is an important tool for machines to get spatial information. Human is realizing high-accuracy depth estimation by dividing problem area, but it is difficult for machine to calculate depth from a single RGB image. Our goal is to improve the accuracy of monocular depth estimation.From the past research, it has been found that obtaining the information stepwise in the global and local areas is effective for depth estimation.Therefore, In this research, we propose a method to utilize the anteroposterior relationship information of the object. Experimental results showed that the overlap information is useful for depth prediction.