Use BEVformer to roll autonomous driving-2

Review of the previous issue: Use BEVformer to roll autonomous driving-1 (qq.com)

      

      In the last issue, we talked about how the transition from 3D to 4D (with the addition of the concept of time) makes BEV even more powerful. Specifically, this kind of power can actually solve the most difficult problems in pure visual solutions, which are based on bad weather, or When the picture is not clear, or when there is sudden obstruction, incorporating the concept of time can greatly improve the accuracy of the solution.

       To summarize what we introduced in the last issue

• 2D: A camera’s 2-dimensional planar features

• 3D: BEV high-dimensional space for multi-camera feature fusion

• 4D: BEV higher dimensional space with time series added

       

        Let's continue