Improving the Performance of Self Driving Neural Networks

Autonomous vehicle is a rapidly growing field. It will improve traffic safety and eliminate congestion. One of the ways to achieve autonomous driving is to use two cameras simulating human eyes. The video inputs are then fed into a neural network to make decisions on how to operate the vehicle. A neural network is an AI model that is inspired by the neurons in the brain. For the neural network to perform its best, it is necessary to have the cameras always mounted in the same location and orientation. This way, the neural network always has consistent inputs. However, when humans mount the cameras, it is hard to ensure it is installed exactly the same way every single time. This project takes the camera feed from improperly mounted cameras and transforms the image to a consistent orientation for the neural network.