Neural Network approximation of nonlinear birds eye view perspective warping with height information

[oswinso]

There's #689 which tries to tackle the problem of the flat plane assumption for image projection using traditional CV techniques. Then, of course, there will always be a deep neural network version that tries to approximate (and maybe produce better results) of traditional CV version.

The hope is that #689 will allow us to create a dataset of transformations from image perspective to birds eye view, which we can hopefully use to train a neural network to approximate.

Or, maybe this approach is dumb. Some related papers I found online:

Related Papers:

• FusionLane: Multi-Sensor Fusion for Lane Marking - Semantic Segmentation Using Deep Neural Networks
• Generative Adversarial Frontal View to Bird View Synthesis
• The Right (Angled) Perspective: Improving the Understanding of RoadScenes Using Boosted Inverse Perspective Mapping