CVPR2024

3D Feature Tracking via Event Camera

Siqi Li, Zhikuan Zhou, Zhou Xue, Yipeng Li, Shaoyi Du, Yue Gao

Abstract

Camera 1 Camera 2 Camera 1 Camera 2 Figure 1 . We present the first high-speed 3D feature tracking method via stereo event cameras and the corresponding high-speed 3D feature tracking dataset. Our proposed method takes high temporal resolution event streams captured from stereo event cameras as input, and could predict the long-term feature motion trajectories of multiple high-speed moving objects within the scene at a rate of 250 FPS.