Real-Time Outdoor Illumination Estimation for Camera Tracking in Indoor Environments

Abstract

Dynamic illumination is a challenging problem for visual robot localization and tracking. In indoor environments, the main source of light during the day is outdoor illumination. We propose a method that estimates the appearance of an indoor scene in real-time based on a reflectance map and the current outdoor lighting. Our outdoor illumination model consists of three components, namely sun, sky and ground, where the sun position is obtained from the scene geolocation and the current time of day. The scene illumination is pre-computed using radiosity transport for each of those components. To deal with dynamic illumination resulting from changing weather conditions, we estimate the outdoor light brightness in each input frame and scale the pre-computed illumination accordingly. We evaluate our approach on real-world data covering diverse outdoor illumination settings and show that our adaptable model is beneficial for direct camera tracking.