Abstract
In this paper, we propose a novel method for estimating time-to-contact (TTC) in scattering media environments, such as fog and water. For an object moving towards another object, time-to-contact measures the time remaining until collision. Therefore, it is especially useful for safe automated vehicle navigation in the 3D space. Time-to-contact estimation is advantageous because it does not require camera calibration, freeing it from calibration errors. Earlier studies commonly used geometric features of objects such as edges and points to estimate time-to-contact. In scattering media environments, the degradation of image intensity caused by light scattering makes it difficult to obtain clear geometric features. Thus, in this paper we propose a method for estimating time-to-contact in scattering media by using the photometric features in the image instead. We use statistical priors to obtain depth information from the captured image intensity and compute the time-to-contact from estimated transmission of the media.
