An automatic driving semantic occupancy rate prediction method across dimensions branches

Abstract

The application provides an automatic driving semantic occupancy prediction method across a dimension branch, which is optimized for efficient feature extraction from a 2D image and subsequent projection into a high-dimensional space, can more effectively convert image data into a 3D voxel representation, and significantly improves the accuracy of semantic occupancy prediction. It mainly consists of five parts: (1) an ImageEncoder component is used to extract features from the input RGB image; (2) an ImageTrans component is used to extract 2D image features into a 3D environment and maintain channel efficiency; (3) a View Transformation module is used to convert the features of the input image; (4) an XDB-BEVStream module is used to refine the voxel features; and (5) an integrated loss function suite is used to improve the performance of the model. The application uses the SemanticKITTI and nuScenes data sets for verification, and compared with the most advanced SOTA method, the XDB-Occ framework is obviously superior to it in terms of semantic occupancy prediction.

Images