HERO — Vision-Based Parking Assistance
In development · Vision-only parking perception · Low-vision coaching HUD
HERO is a coaching-only parking-assistance project for a Hyundai Ioniq. It combines a camera-only perception stack with a driver interface designed around low-vision legibility. Development and measurement run in CARLA; the system provides guidance and does not intervene in vehicle control.
Vision-Only Perception
- Two-camera ground-plane odometry estimates vehicle motion without vehicle-mounted LiDAR.
- Plane-sweep dense occupancy reconstructs parking-space geometry in C++/CUDA.
- A reproducible CARLA evaluation harness compares perception output against an Autoware LiDAR reference.
- Measurement outputs are separated from the presentation layer so the perception path can be evaluated independently.
Low-Vision Coaching HUD
- Angular legibility budgets connect display geometry to the driver's usable visual angle.
- Guidance includes a parking corridor, stop line, gear state, and explicit degraded-state messaging.
- ROS 2 coaching logic is separated from the Qt presentation layer.
- The current geometry layer includes 26 tests for display and guidance calculations.
Architecture
Camerasdual-view CARLA image streams
→
Perceptionground-plane odometry and dense occupancy
→
Coaching Stateparking geometry, gear, stopping, degraded state
→
HUDlow-vision-oriented Qt presentation
Engineering Status
In development. The current implementation includes a reproducible measurement harness, scenario-driven HUD states, geometry tests, and visual-acuity simulation outputs for interface review.