Not really possible to do with a simple delay, and if it was that easy it'd have already been done. However, the math you do need is a close relative of the pan

Not really possible to do with a simple delay, and if it was that easy it'd have already been done. However, the math you do need is a close relative of the panorama mode you have on your phone, and I feel like it's about the right time for it to have been refined and hardened well enough to show up on a car. Conceptually it's simple: Take the images, stitch them together, center the view on the cameras. Repeat with the next frame, but instead of using only the new images, stitch them into the same panorama you started building last frame. Keep stitching even as older images start to move entirely out of the cameras' field of view. Garbage-collect eventually, but you have a constant target synthetic FoV so you can guarantee bounded consumption. Add a few years of engineering and you have a really limited, single-propose SLAM system cobbled together from OTC algorithms that's just good enough to handle the ground underneath an SUV.