Track papers across labs, topics, and trends.
Search papers, labs, and topics across Lattice.
This paper introduces KLTNet, a novel learning-based sparse feature tracker designed to enhance the robustness and accuracy of monocular visual-inertial odometry (VIO) systems by addressing the limitations of classical KLT trackers. By employing a coarse-to-fine, dense-to-sparse architecture, KLTNet effectively combines low-resolution dense optical flow for global motion initialization with triplet-patch refinement for precise tracking, significantly reducing accumulated drift. Experimental results on both public benchmarks and a self-collected low-texture dataset show that KLTNet outperforms traditional KLT methods in tracking and odometry accuracy while maintaining real-time performance on embedded platforms.
KLTNet achieves superior tracking accuracy and robustness in low-texture environments, outperforming classical methods while ensuring real-time processing capabilities.
Many feature-based visual-inertial odometry (VIO) systems rely on sparse feature tracking, whose accuracy and robustness directly affect state estimation. Classical KLT trackers rely primarily on local image patches and can become unreliable under rapid motion or in low-texture environments. We propose KLTNet, a lightweight learning-based, plug-and-play sparse feature tracker designed to replace classical KLT trackers in KLT-based VIO front ends. KLTNet follows a coarse-to-fine, dense-to-sparse architecture that combines low-resolution dense optical flow for robust global motion initialization with triplet-patch refinement for accurate and temporally consistent tracking. A fixed reference patch provides a stable anchor throughout each feature track and helps reduce accumulated tracking drift. In addition, KLTNet predicts anisotropic confidence weights supervised through differentiable multi-view triangulation, which can be used as observation weights in compatible VIO estimators. Experiments with VINS-Mono and OpenVINS on public benchmarks and a self-collected low-texture dataset demonstrate improved tracking and odometry accuracy over classical KLT, while maintaining real-time performance on an embedded platform.