基于协调多尺度特征增强网络的驾驶场景多任务感知方法
By coordinating the multi-scale feature enhancement network CMFANet architecture, the efficiency and accuracy issues of multi-task networks in autonomous driving are solved, achieving comprehensive perception of traffic scenarios, adapting to complex environments, and optimizing network performance.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- SOUTHWEST JIAOTONG UNIV
- Filing Date
- 2025-05-27
- Publication Date
- 2026-07-17
AI Technical Summary
Existing multi-task networks struggle to efficiently and accurately handle traffic object detection, drivable area segmentation, and lane detection in autonomous driving, especially due to the increased inference time and insufficient generalization ability caused by reliance on anchor point methods. Furthermore, existing standardized backbone networks are difficult to adapt to the requirements of different tasks.
The CMFANet architecture, based on a coordinated multi-scale feature enhancement network, is adopted. It includes a shared backbone network layer, independent task neck layers and task heads. Multi-level features are extracted through a hybrid enhancement strategy, and features of different tasks are processed by dynamic deformation modules and dynamic spatial context fusion modules. The output is optimized by combining anchorless decoupling design and lightweight structure.
It enables efficient and accurate simultaneous execution of traffic object detection, drivable area segmentation, and lane line segmentation in resource-constrained environments, adapts to complex geometric deformation scenarios, meets real-time application requirements, and optimizes network performance.
Smart Images
Figure CN120564151B_ABST