【摘要】 针对自动驾驶汽车,提出一种以道路消失点为导向的结构化道路上远距离障碍物检测方法。分别搭建了现实和虚拟数据采集系统,在现实和虚拟环境中采集了新的数据集。采用深度学习方法,设计了一种新的基于全局特征的道路消失点检测模型,以该模型检测到的道路消失点为导向,确定图像中包含障碍物的区域,在该区域上进行障碍物检测。选择目前主流目标检测网络,在有消失点和无消失点导向的条件下进行对比试验。试验结果表明在增加消失点导向的情况下,对于微小障碍物的检测得到了较好的结果。更多还原

【Abstract】 A method of long-distance small road obstacles detection based on road vanishing point is proposed. The real and virtual data collection systems were built,and new dataset were collected in real and virtual environments. Using the deep learning method,a new road vanishing point detection model based on global features is proposed. The road vanishing point detected by the model is used to determine the area of the image containing obstacles,and obstacle detection is performed on the area. Several excellent object detection networks were selected,and the comparison test was carried out under the contain of vanishing point guidance and vanishing point-free. The experimental results show that the small road obstacle detection results are better under the condition of increasing vanishing point guidance.更多还原