Bioinspired framework for real-time collision detection with dynamic obstacles in cluttered outdoor environments using event cameras

Abstract

In the field of robotics and visual-based navigation, event cameras are gaining popularity due to their exceptional dynamic range, low power consumption, and rapid response capabilities. These neuromorphic devices facilitate the efficient detection and avoidance of fast moving obstacles, and address common limitations of traditional hardware. However, the majority of state-of-the-art event-based algorithms still rely on conventional computer vision strategies. The goal is to shift from the standard protocols for dynamic obstacle detection by taking inspiration from the time-computational paradigm of biological vision system. In this paper, the authors present an innovative framework inspired by a biological response mechanism triggered by approaching objects, enabling the perception and identification of potential collision threats. The method, validated through both simulation and real-world experimentation, charts a new path in the application of event cameras for dynamic obstacle detection and avoidance in autonomous unmanned aerial vehicles. When compared to conventional methods, the proposed approach demonstrates a success rate of 97% in detecting obstacles within real-world outdoor settings.