Improving of 3-D structure estimation methods by image using ambient light

Ambient light is a concept proposed by visual affordance theory that refers to the distribution of light a single viewpoint receives from its surroundings. By observing the changes in the luminance distribution of light produced by the structure of the environment from a single viewpoint, we can estimate the surrounding structure. When we apply this natural mechanism of recognizing the outside world to image measurement technology, it is possible to obtain the information necessary to recognize the surrounding environment by observing the ambient light without necessarily detecting or recognizing the object. For example, this methods can be used to estimate the 3-D structure in an image using only simple algorithms, without the need for complex image processing or prior data for machine learning. In a previous study, we proposed a structure estimation method to understand the structure of the surrounding environment by capturing ambient light as luminance, thus providing a viewpoint in an indoor environment. The layout of the surface of the surrounding environment structure is estimated by assuming that a horizontal wall is a vertical wall and a vertical wall is a horizontal wall, and that the indoor structure is composed of vertical walls, horizontal walls, and space. However, this method sometimes failed to estimate the structure correctly because changes in the position of the lighting in the room affected the estimation results. In addition, it sometimes misrecognized the boundary of the surface when estimating the structure. In contrast, by using the dynamic thresholding method proposed in the current study, we were able to obtain results that were robust to changes in illumination position. In addition, by focusing on the presence or absence of luminance continuity in an image, we were able to accurately detect the boundary of the surface. In this paper, we describe these results of improving the 3-D structure estimation method.