Frame-based Feature Tracking and EKF Framework for Event Cameras

Abstract

Event cameras are silicon retina sensors that are more advantageous than traditional cameras in low-latency tracking and high dynamic range scenes. In this paper, we present a visual odometry algorithm based on the Dynamic and Active-pixel Vision Sensor (DAVIS), and the 6 Degree-of-Freedom (6-DoF) object motion can be tracked by the proposed algorithm. We detect features and track motion on the image plane, then feature-based pose estimation and extended Kalman filter (EKF) framework are tightly intertwined in event-based visual odometry. In experiments, the accuracy of our approach is evaluated in several object tracking scenarios. The trajectory of a low-latency and high-rate tracking is obtained, and the utilization rate of CPU resources is improved by using an event-driven strategy.