[20a-P01-1~21] 3.7 Optical measurement, instrumentation, and sensor

[20a-P01-20] Reducing the Impact of Filters on Stereo Matching of Generative Adversarial Networks

〇Tatsuki Uehara¹, Kent Kumagai¹, Dong Wei¹ (1.NUT)

Keywords:

Generative Adversarial Networks,Stereo Matching,Filters

We aim to realize stereo measurement with a 360-degree camera. Fisheye images need to be corrected for distortion. When stereo matching is performed using images generated from fisheye images by GAN, the accuracy of the depth information may be lower than the result of stereo matching using the correct image. Therefore, we apply a filter to the image generated by GAN and perform stereo matching, and compare the evaluation of the generated image before and after the filter.