Hierarchical framework for direct gradient-based time-to-contact estimation

The time-to-contact (TTC) estimation is a simple and convenient way to detect approaching objects, potential danger, and to analyze surrounding environment. TTC can be estimated directly from single camera though neither distance nor speed information can be estimated with single cameras. Traditional TTC estimation depends on “interesting feature points” or object boundaries, which is noisy and time consuming. In [13], we propose a direct “gradient-based” method to compute time-to-contact in three special cases that avoid feature points/lines and can take advantages of all related pixels for better computation. In this follow-up paper, we discuss the method to deal with the most general cases and propose a hierarchical fusion framework for direct gradient-based time-to-contact estimation. The new method enhances accuracy, robustness and is computationally efficient, which is important to provide fast response for vehicle applications.