基于LV-DOT算法增强传感器融合与追踪精度研究

Research on Enhancing Sensor Fusion and Tracking Accuracy Based on the LV-DOT Algorithm

摘要: 为了实现对动态目标进行高精度检测与稳定追踪，本文将现有的基于LiDAR和视觉传感器相融合的LV-DOT框架进行改进优化，核心创新涵盖两条：一是提出基于形态跟大小一致性的跨模态检测结果校验举措，在数据融合阶段审视LiDAR点云与视觉检测结果的几何形态和尺度特性一致性，减小错误检测和遗漏检测的发生概率；二是设置基于运动方向的误差补救机制，采用动态校验的方式来确认障碍物运动轨迹连续性与方向一致，抑制传感器噪声、环境遮挡或瞬时观测偏差引发的轨迹漂移，提升追踪的稳健水平。通过实验结果表明，所提方法在检测精度、召回率及实时处理效率方面均有显著改善，此优化框架不仅增强了多传感器融合在动态场景下的稳健性与稳定性，也给移动机器人检测、追踪动态障碍物提供了可行的技术办法。

Abstract: To achieve high-precision detection and stable tracking of dynamic targets, this paper improves the existing LiDAR-Vision ss-modal detection validation method based on morphological and size consistency is proposed, which examines the geometric and scale agreement between LiDAR point clouds and visual detection results during the data fusion stage, thereby reducing false and missed detections; second, a motion-direction-based error remediation mechanism is introduced, which employs dynamic verification to ensure the continuity and directional consistency of obstacle motion trajectories, mitigating trajectory drift caused by sensor noise, environmental occlusion, or transient observation deviations, thus enhancing tracking robustness. Experimental results demonstrate that the proposed approach significantly improves detection accuracy, recall rate, and real‑time processing efficiency. The optimized framework not only strengthens the robustness and stability of multi‑sensor fusion in dynamic scenarios, but also provides a viable technical solution for mobile robots in detecting and tracking dynamic obstacles.