Optimising Faster R-CNN Training to Enable Video Camera Compression for Assisted and Automated Driving Systems