Seeing Stuff: Perceiving Materials and their Properties

Descripción del proyecto

Comprender la percepción y las propiedades de los materiales

Las diversas propiedades físicas, apariencias y comportamientos de los materiales desempeñan un papel fundamental en las tareas cotidianas. A pesar de las numerosas investigaciones sobre el reconocimiento visual, la percepción material se ha descuidado relativamente. Entender cómo percibimos e interactuamos con los materiales, predecir sus comportamientos y adaptar nuestras interacciones en función de sus características físicas es esencial para muchas actividades. El equipo del proyecto STUFF, financiado por el CEI, pretende colmar esta laguna de conocimientos sobre cómo percibimos, entendemos, predecimos e interactuamos con los materiales y sus propiedades. Adopta un método interdisciplinar, que combina técnicas de la psicología experimental, la neurociencia del comportamiento, la infografía, el análisis computacional de imágenes, el aprendizaje automático y el arte. El proyecto consta de cinco paquetes de trabajo interconectados e incluye materiales reales y simulados.

Objetivo

Different materials, such as silk, soil, steel and soap exhibit an astonishing variety of physical properties, appearances and behaviours. The material properties of objects and substances are central to practically every task we perform, from selecting and preparing food, to detecting slippery ground, to using tools effectively. Without touching a surface, we usually enjoy a vivid impression of what it would feel like, through the sense of sight. Yet, how we do so remains mysterious. Decades of research has focussed on the visual recognition of objects, faces and scenes. By comparison, how we see, think about and interact with ‘stuff’ has been relatively neglected. STUFF addresses this major gap in our understanding.

Material perception poses unique and fascinating challenges. The image of a surface is a complex and ambiguous combination of lighting, shape, and material properties. How does the visual system disentangle these intermingled physical factors? Deformable materials like liquids and textiles move and change shape in complex yet lawful ways. How do we infer intrinsic properties like viscosity, compliance and elasticity from such ever-changing stimuli? And how do we reason about and predict their future behaviours as they interact with their surroundings? How do we adapt our own interactions with objects to take into consideration their hardness, density, friction and other physical characteristics, allowing us to pluck a raspberry without crushing it, or pick up wet soap without it slipping through our fingers?

STUFF takes a radically interdisciplinary approach to these questions in five tightly interconnected work-packages, bringing together state-of-the-art methods from experimental psychology and behavioural neuroscience, computer graphics and computational image analysis, machine learning and even art. We draw on real and simulated materials, to uncover how we perceive, reason about, predict and interact with materials and their properties.