The invention relates to the technical field of urban road conditions, in particular to an Occ and SLAM-based end-cloud collaborative urban road condition updating method, which comprises the following steps of: obtaining a projection matrix of an initial frame by utilizing an internal reference matrix of a camera; projecting the depth map to a 3D point cloud space to obtain a point cloud; projecting the point cloud to a voxel grid; extracting features of a certain frame of image, performing deformable interaction enhancement on the candidate voxel query set and the features of the certain frame of image by adopting a deformable attention operator, and outputting a three-dimensional semantic occupancy prediction result; the longitude and latitude coordinates of the vehicle at the same moment and the prediction accuracy of each semantic meaning are read through timestamps, different objects are recognized through a DBSCAN clustering function, and the objects and noise are recognized according to the minimum voxel number; and comparing the confidence coefficients of the same object of two continuous frames, and obtaining the repositioned size attribute of the same object by using a size attribute positioning formula. The method solves the problem of positioning error of the same object during high-frequency road condition updating at present.
本发明涉及城市路况技术领域,尤其涉及基于Occ和SLAM的端云协同的城市路况更新方法,包括利用相机内参矩阵得到初始帧的投影矩阵;将深度图投影到3D点云空间,得到点云;将点云投影到体素网格;提取某帧图像的特征,采用可变形注意力算子对候选体素查询集和某帧图像的特征进行可变形交互增强,输出三维语义占用预测结果;通过时间戳读取相同时刻的自车经纬坐标和每种语义的预测准确度,利用DBSCAN聚类函数对不同物体进行识别,并根据最小体素数量进行物体和噪声的识别;比较连续两帧的同一物体的置信度,利用尺寸属性定位公式得到同一物体的重定位后的尺寸属性。本发明解决目前路况高频更新时相同物体存在定位误差问题。
End-cloud collaborative urban road condition updating method based on Occ and SLAM
基于Occ和SLAM的端云协同的城市路况更新方法
2024-09-17
Patent
Electronic Resource
Chinese
Road-SLAM: Road Marking Based SLAM with Lane-Level Accuracy
British Library Conference Proceedings | 2017
|European Patent Office | 2015
|