LCG-3D: Local Key Object 3D Positioning and Trajectory Generation under Cross-Modal Geometric Consistency Constraints

Abstract

Accurate three-dimensional (3D) localization and trajectory generation of key objects in complex environments remain challenging due to limitations of existing single- or multi-modal methods, such as low accuracy, slow processing speed, and sensitivity to occlusions, especially when relying on single-modality sensing. This paper proposes LCG-3D, a novel framework for local key object 3D positioning and trajectory generation under cross-modal geometric consistency constraints. By integrating heterogeneous sensor data, including RGB images, depth maps, and LiDAR point clouds, LCG-3D enforces local geometric consistency to align multi-modal observations in 3D space. The algorithm selectively focuses on key objects, reducing computational overhead while improving robustness in dynamic or occluded environments. A trajectory generation module further predicts object motion by leveraging both current localization and historical geometric patterns. Extensive experiments on publicly available multi-modal datasets demonstrate that LCG-3D achieves superior localization accuracy and trajectory fidelity compared with state-of-the-art methods, highlighting its potential for applications in intelligent transportation, robotics, and augmented reality.