一种面向复杂路况的自动驾驶场景动态标注系统及方法
By constructing a dynamic target recognition scene model through multi-sensor fusion and deep learning algorithms, a secondary recognition and annotation of autonomous driving image annotation is performed, which solves the problem of inaccurate manual selection and improves annotation efficiency and accuracy.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- CHONGQING FEILIXIN TECH CO LTD
- Filing Date
- 2025-09-04
- Publication Date
- 2026-07-17
AI Technical Summary
In the process of image annotation for autonomous driving, if the bounding box selected manually is too large or too small, it can lead to information confusion or missing target features, affecting the accuracy of model recognition and the efficiency of operators.
Multi-sensor fusion technology is used to acquire multi-source image datasets. A dynamic target recognition scene model is constructed through deep learning algorithms. The manually selected regions are then re-identified and labeled. Convolutional neural networks and Kalman filtering algorithms are used for target classification and trajectory tracking to generate dynamic labeling sequences.
It reduced the workload of operators, improved annotation efficiency, alleviated work fatigue, and ensured the accuracy and continuity of annotation.
Smart Images
Figure CN121121260B_ABST