基于CNN和Transformer点云图像融合的道路检测
Fusion of Point-Cloud and Image for Road Segmentation Using CNN and Transformer
华怡坦 1黄影平 1过文昊1
作者信息
- 1. 上海理工大学光电信息与计算机工程学院,上海 200093
- 折叠
摘要
针对道路检测模型易受光线及阴影影响而导致精度不高及道路边缘分割不准确的问题,提出一种基于Transformer和卷积神经网络模型混合且以RGB图像和三维激光雷达点云共同为输入的道路分割算法,实现了无人车在自动驾驶过程中对所在行驶道路的精确感知.在KITTI道路数据集上的实验结果表明:与现有的道路检测模型相比,本文方法在分割精度方面具有较好的性能.
Abstract
To address the problem of low accuracy and inaccurate road edge segmentation caused by the susceptibility of road detection models to light and shadows,we propose a road segmentation algorithm based on a hybrid of Transformer and convolutional neural network models,utilizing RGB images and 3D LIDAR point clouds as inputs to enhance the precise perception of driving roads for autonomous vehicles.Experimental results on the KITTI road dataset demonstrate the superior segmentation accuracy of the proposed method compared with existing road detection models.
关键词
道路检测/语义分割/数据融合/TransformerKey words
road detection/semantic segmentation/data fusion/Transformer引用本文复制引用
基金项目
国家自然科学基金(62276167)
上海市自然科学基金(20ZR1437900)
出版年
2024
NETL
NSTL
万方数据