A method for improving the constancy of cone shape and size perception using a single RGB-D image

The 3D shape of a solid is the most crucial object property. Object shape and size information can be derived from depth information as it provides rich 3D geometric information. It can be found from the literature a substantial number of strategies to estimate the shape and size of solid objects from depth information. Nevertheless, the quality of depth information is a function of intrinsic and extrinsic factors. Additionally, noise is inherently embedded in the depth information. As a consequence, using those techniques, the perception of the shape and size of 3D objects from depth solely, results in viewpoint-dependent accuracy and precision. In the current study, we perform a detailed analysis of the degree of constancy of shape and size perception that representative techniques can achieve on depth information of a real cone object, as it is a commonly encountered man-made object. Furthermore, we propose a novel shape and size perception strategy utilizing both 2D and 3D information to enhance the constancy of object shape and size perception. The results clearly show that a remarkable enhancement in the constancy of the shape and size perception of cone objects is observed using the novel method.