3D occupancy grid detection method based on voxel entropy value and confidence

The 3D occupancy grid detection method based on voxel entropy and confidence solves the problem of historical feature overlay when objects are occluded, improves the accuracy and stability of detection, and achieves more reliable obstacle detection.

CN122244831APending Publication Date: 2026-06-19YANTAI PORT GRP CO LTD
View PDF 0 Cites 0 Cited by

Patent Information

Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
YANTAI PORT GRP CO LTD
Filing Date
2026-03-20
Publication Date
2026-06-19

AI Technical Summary

Technical Problem

Existing 3D occupancy grid detection methods directly cover historical features when objects are temporarily occluded, making it easy for obstacles to be missed due to occlusion and sparse point clouds. Sensor noise causes the detection box to jump randomly, resulting in occasional false detections in a single frame, which reduces the accuracy, reliability and stability of the detection.

Method used

A 3D occupancy grid detection method based on voxel entropy and confidence is adopted. Point cloud data is acquired from LiDAR mounted on a vehicle, scene features are extracted to generate memory entries, the weight of memory entries is calculated using voxel entropy map and confidence, memory features of historical and current frames are fused, and a pre-trained 3D occupancy grid detection model and static consistency loss function are used to improve detection accuracy.

Benefits of technology

It effectively avoids the problem of historical features being covered when objects are temporarily occluded, avoids the omission of obstacles and random jumping of the detection box, and improves the accuracy, reliability and stability of 3D occupancy grid detection.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN122244831A_ABST
    Figure CN122244831A_ABST
Patent Text Reader

Abstract

This disclosure provides a 3D occupancy raster detection method based on voxel entropy and confidence levels. In this method, data from the 3D occupancy raster detection process is stored as memory entries. The weights of relevant memory entries are determined using their voxel entropy maps and confidence levels. The weights of memory entries in the current frame are determined based on their voxel entropy maps and confidence levels. The memory features of each relevant memory entry are fused with the memory features of the current frame's memory entries based on their respective weights and the weight of the current frame's memory entries to obtain a fused feature. Finally, a 3D occupancy probability raster is obtained using this fused feature. This disclosure effectively improves the accuracy, reliability, and stability of 3D occupancy raster detection.
Need to check novelty before this filing date? Find Prior Art