【論文要約:自動運転関連】Pre-Trained Vision Models as Perception Backbones for Safety Filters in Autonomous Driving