Use BEVformer to roll autonomous driving-1

The reason why it is -1 is because there is a high probability that I will not be able to finish writing one article, but I don’t know how many articles I should use to talk about the matter, so I will write it down first.     

As usual, the paper address is: 2203.17270v1.pdf (arxiv.org)

Image

      What is BEV? Bird's -Eye-View means bird's eye view.

      For example, most of the slightly more traditional autonomous driving has been implemented. If you rely on a pure CV solution, then basically no matter what downstream tasks you do, whether it is object detection or image segmentation, you can see the area in front of you, or the field of view in the camera, and then monitor your own cameras respectively. Make decisions based on the results of the tasks in

      The bird's eye view is different, it allows you to have a God's perspective, see Tesla's driving interface

Image

      As shown in the picture above, the middle picture is worthy of being called a bird's-eye view. It can sense all the surrounding environment without blind spots, and then realize the most reasonable driving action (which is different from the high-precision map solution. High-precision maps only present subjective road conditions. Remember, it is the presentation of subjective road conditions, not perception. At the same time, it is impossible to perceive other surrounding objects at the same physical time without cooperation with radar or other solutions.)

     

Guess you like

Origin blog.csdn.net/kingsoftcloud/article/details/135159643