A method for road boundary detection and tracking using laser ladar with respect to a vehicle' s local coordinates is proposed. It can be applied to different types of road conditions, such as roads with or without c...A method for road boundary detection and tracking using laser ladar with respect to a vehicle' s local coordinates is proposed. It can be applied to different types of road conditions, such as roads with or without curbs, having relatively rough road surface and with obstacles on road surface. In the method, some line segments are extracted after a series of preprocessing on range data. The extracted line segments are combined and further selected. They are then united to match the road models and generate the road boundary points which are tracked by Kalman filter. Then the obtained road boundary points are transformed to build a precise vector map by least squares fitting algorithm. These fitted line segments represent road boundary vectors. The vector map is precise enough to provide ample road information such as the orientation of road, the road width and the passable road region. Finally, extensive experiments conducted in urban and semi-urban environment demonstrate the robustness, effectiveness and viability of the proposed method.展开更多
基于视频的目标检测在恶劣天气情况下识别效果较差,故需弥补视频缺陷、提高检测框架的鲁棒性。针对此问题,文中设计了一个基于雷达和视频融合的目标检测框架,利用YOLOv5(You Only Look Once version 5)网络获得图片特征图与图片检测框,...基于视频的目标检测在恶劣天气情况下识别效果较差,故需弥补视频缺陷、提高检测框架的鲁棒性。针对此问题,文中设计了一个基于雷达和视频融合的目标检测框架,利用YOLOv5(You Only Look Once version 5)网络获得图片特征图与图片检测框,利用基于密度的聚类获得雷达检测框,并将雷达数据进行编码,得到基于雷达信息的目标检测结果。最后将两者的检测框叠加得到新ROI(Region of Interest),并得到融合雷达信息后的分类向量,提高了在极端天气下检测的准确率。实验结果表明,该框架的mAP(mean Average Precision)达到了60.07%,且参数量仅为7.64×10^(6),表明该框架具有轻量级、计算速度快、鲁棒性高等特点,可以被广泛应用于嵌入式与移动端平台。展开更多
基金Supported by the National Natural Science Foundation of China (61174178)
文摘A method for road boundary detection and tracking using laser ladar with respect to a vehicle' s local coordinates is proposed. It can be applied to different types of road conditions, such as roads with or without curbs, having relatively rough road surface and with obstacles on road surface. In the method, some line segments are extracted after a series of preprocessing on range data. The extracted line segments are combined and further selected. They are then united to match the road models and generate the road boundary points which are tracked by Kalman filter. Then the obtained road boundary points are transformed to build a precise vector map by least squares fitting algorithm. These fitted line segments represent road boundary vectors. The vector map is precise enough to provide ample road information such as the orientation of road, the road width and the passable road region. Finally, extensive experiments conducted in urban and semi-urban environment demonstrate the robustness, effectiveness and viability of the proposed method.
文摘基于视频的目标检测在恶劣天气情况下识别效果较差,故需弥补视频缺陷、提高检测框架的鲁棒性。针对此问题,文中设计了一个基于雷达和视频融合的目标检测框架,利用YOLOv5(You Only Look Once version 5)网络获得图片特征图与图片检测框,利用基于密度的聚类获得雷达检测框,并将雷达数据进行编码,得到基于雷达信息的目标检测结果。最后将两者的检测框叠加得到新ROI(Region of Interest),并得到融合雷达信息后的分类向量,提高了在极端天气下检测的准确率。实验结果表明,该框架的mAP(mean Average Precision)达到了60.07%,且参数量仅为7.64×10^(6),表明该框架具有轻量级、计算速度快、鲁棒性高等特点,可以被广泛应用于嵌入式与移动端平台。