End-to-end automatic driving control method and device based on multi-camera fusion
Through multi-camera fusion and deep neural network model, multi-view object detection and tracking are realized, and feature fusion weights are dynamically adjusted, which solves the shortcomings of multi-view information fusion and occlusion processing in traditional methods, and improves the perception and control reliability of the autonomous driving system.
Patent Information
- Application Number
- CN202510370046.0
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-03-26
- Publication Date
- 2025-08-01
AI Technical Summary
The existing end-to-end autonomous driving control methods rely on a single camera perspective and lack the coordinated processing of multi-view information, resulting in insufficient perception accuracy and control stability. There are bottlenecks in the fusion of multi-camera feature, lack of effective perspective consistency constraints and dynamic weight adjustment mechanisms, making it difficult to deal with abnormal situations such as camera occlusion.
Through multi-camera fusion, spatial transformation and coordinate mapping are used to realize multi-view object detection and tracking, a multi-view feature fusion network is designed, including feature extraction, dynamic weight allocation and feature fusion subnet, a geometric consistency constraint between viewing angles and reconstruction loss functions, a deep neural network model is built, a feature fusion weight is dynamically adjusted, and accurate control instructions are output.
It significantly improves the perception ability and control reliability of the autonomous driving system, can deal with abnormal situations such as camera occlusion, and provides stable environmental perception and driving control.