A lane line recognition method, device, equipment and medium
By receiving images and utilizing a pre-trained recognition model, combined with pixel adjacency relationships and adjustment vectors, the problem of inaccurate lane line recognition was solved, achieving high-accuracy recognition in multi-lane scenarios.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- CHINA TELECOM CLOUD TECH CO LTD
- Filing Date
- 2022-08-08
- Publication Date
- 2026-06-16
AI Technical Summary
The inaccurate lane line recognition in existing technologies includes errors caused by outdated high-precision maps and the influence of the environment on radar recognition.
By receiving an image and inputting it into a pre-trained recognition model, the recognition model outputs a target image carrying whether each pixel is a pixel on the lane line. Combining the adjacency relationship of the pixels and the adjustment vector, the lane line is determined, and the model is optimized through a training sample set to improve accuracy.
It improves the accuracy of lane line recognition, reduces the workload of manually modifying high-precision maps, and ensures accurate lane line recognition in multi-lane scenarios.
Smart Images

Figure 1 
Figure 2