A Pure Vision-Based 3D Object Detection System and Method Based on Bird's-Eye View
By combining two-dimensional semantic fusion and temporal information enhancement with EfficientNetB0 and ResNet18 networks, the depth prediction accuracy and stability of pure visual 3D object detection are improved. This solves the problems of high model complexity and insufficient detection stability in existing technologies, and enables real-time vehicle deployment of lightweight models.
CN122290076APending Publication Date: 2026-06-26ORIGINAL JIWEI TECHNOLOGY (SHANGHAI) CO LTD
View PDF 0 Cites 0 Cited by
Patent Information
- Application Number
- CN202610757024.4
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2026-05-29
- Publication Date
- 2026-06-26
Smart Images
Figure CN122290076A_ABST
Abstract
This invention relates to the field of environmental perception and discloses a pure visual 3D target detection system and method based on a bird's-eye view. The pure visual 3D target detection system based on a bird's-eye view includes: a 2D image feature extraction module extracting features from the original image to obtain 2D image features; a 2D semantic information fusion module fusing the 2D semantic segmentation mask with the 2D image features to obtain 2D features with fused semantic information; a depth feature extraction module combining the 2D image features to obtain 2D features with depth information; a BEV feature projection module projecting the 2D features with depth information onto the BEV feature space to obtain the current frame BEV features; a temporal information fusion module splicing and fusing the aligned historical frame BEV features with the current frame BEV features in the channel dimension to obtain BEV features with fused temporal information; and a 3D target detection module outputting 3D target detection results.
Need to check novelty before this filing date? Find Prior Art