摘要
结合自动驾驶鸟瞰图(Bird's Eye View,BEV)项目案例,针对基于自动泊车系统的BEV算法开展研究。BEV是端到端的,由神经网络将图像信息从图像空间转换至BEV空间的技术。与传统图像空间感知相比,BEV感知可将多个传感器采集的数据输入到同一空间进行处理,有效避免误差叠加。系统采用环视鱼眼摄像机,通常用于自动驾驶中的近距离感知,车辆四面的4个鱼眼摄像头足以覆盖车辆周围360°的范围,捕捉整个近距离区域,将摄像头感知画面直接传至人工智能(Artificial Intelligence,AI)算法中,生成鸟瞰视角的3D空间,最终研发完成具有自动泊车、自动驾驶功能的BEV系统。
Based on the Bird's Eye View(BEV)project case of autonomous driving,research is conducted on the BEV algorithm based on automatic parking systems.BEV is an end-to-end technique in which neural networks convert image information from image space to BEV space.Compared with traditional image spatial perception,BEV perception can input data collected by multiple sensors into the same space for processing,effectively avoiding error stacking.The system adopts a panoramic fisheye camera,which is usually used for close range perception in autonomous driving.The four fisheye cameras on all sides of the vehicle are sufficient to cover a 360°range around the vehicle,capturing the entire close range area.The camera perception image is directly transmitted to artificial intelligence(AI)algorithms to generate a bird's-eye perspective in 3D space,and ultimately develop a BEV system with automatic parking and autonomous driving functions.
作者
黄烁
Huang Shuo(Shanghai Aiqirui Intelligent Technology Co.,Ltd.,Shanghai 200000)
出处
《现代工程科技》
2024年第12期97-100,共4页
Modern Engineering Technology
作者简介
黄烁(1986-),女,硕士,上海艾琦睿智能科技有限公司工程师,研究方向为计算机软件研发。
