激光雷达目标检测模型训练方法、检测方法及装置
By employing a two-stage training method, utilizing LiDAR point cloud data and 3D bounding box annotation information, the LiDAR target detection model achieves accurate prediction of vehicle speed without relying on speed annotations. This solves the problem of insufficient speed inference capability in traditional methods and reduces annotation costs.
CN115761392BActive Publication Date: 2026-07-17JILUO TECH (SHANGHAI) CO LTD
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- JILUO TECH (SHANGHAI) CO LTD
- Filing Date
- 2022-10-14
- Publication Date
- 2026-07-17
AI Technical Summary
Technical Problem
Traditional lidar target detection methods lack velocity inference capabilities, making it impossible to accurately predict vehicle speed in open scenarios, and their reliance on continuous frame annotation increases annotation costs.
Method used
The two-stage training method first uses the original LiDAR point cloud data and 3D bounding box annotation information to pre-train the model, and then performs self-supervised regression training to obtain velocity estimation capability, thus avoiding dependence on velocity annotation information.
Benefits of technology
This invention enables the lidar target detection model to accurately predict vehicle speed without relying on continuous frame annotation, reducing annotation costs and improving the accuracy of speed prediction.
✦ Generated by Eureka AI based on patent content.
Smart Images
Figure CN115761392B_ABST
Abstract
本发明提供一种激光雷达目标检测模型训练方法、检测方法及装置，通过获取原始激光雷达点云数据和3D框标注信息，3D框标注信息包括目标标注位置；根据原始激光雷达点云数据和3D框标注信息对激光雷达目标检测模型进行第一阶段训练，得到预训练激光雷达目标检测模型；根据预训练激光雷达目标检测模型输出的不同时刻目标预测位置与目标标注位置对预训练激光雷达目标检测模型进行第二阶段速度自监督的回归训练，得到具有速度估计能力的激光雷达目标检测模型，通过两阶段的训练，不需要速度标注信息也可以得到精确的速度预测信息，满足用户需求。
Need to check novelty before this filing date? Find Prior Art