道路检测方法、设备、装置及存储介质
By combining an improved feature extraction and classification prediction model, the problem of low road detection accuracy in intelligent connected vehicles using existing image semantic segmentation models is solved, achieving pixel-level image classification and improving the accuracy of image semantic segmentation.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- SHANGHAI DATANG MOBILE COMM EQUIP
- Filing Date
- 2022-03-23
- Publication Date
- 2026-07-17
AI Technical Summary
Existing deep learning-based image semantic segmentation models have low accuracy in road detection in intelligent connected vehicles, especially in unrestricted scenarios where their generalization ability is insufficient, leading to a decrease in image segmentation accuracy.
An improved method for fusing a feature extraction network model and a classification prediction model is adopted. The feature extraction network is trained using the SegNet model, and the softmax output layer is replaced with a linear output layer. The gradient boosting decision tree (GBDT) model is combined with the model for pixel classification. The model is trained using the CamVid dataset to achieve pixel-level image classification.
It improves the accuracy of image semantic segmentation, meets the application needs of intelligent connected vehicles in unrestricted scenarios, achieves pixel-level image classification, and enhances the accuracy of image semantic segmentation.
Smart Images
Figure CN116863467B_ABST