Self driving cars are going to drive the next industrial revolution by freeing up valuable human resources and by democratizing mobility & accessibility. The technology is quite promising and has tremendous amount of scope. However, there are key bottlenecks that need to be addressed, in order for the systems to be economically viable. One of the key bottlenecks is the High Definition Map requirement to enable the car to have scene context. In this paper we describe the state of the art and present a next generation orthogonal approach to creating scene context for self driving cars.
Hyperspec.ai is developing a next generation vision pipeline that will utilize equirectangular imagery instead of standard pin-hole images to gain an understanding of the scene. We present novel innovations that will contribute to the perception stacks which will dramatically revolutionize how self driving cars create scene context while navigating in complex environments