Vision approach towards fully self-reverse parking system

Some commercial vehicle models have been equipped with semi-autonomous parking systems to a certain extent. But gaps to fully automated solutions still exist and cost considerations further constrain their acceptance among consumers. In this paper, we propose a low-cost vision-based approach to a fully self-reverse parking system. It consists of four key modules: image processing incorporated with Kalman filter for consistent vehicle pose estimation, path planning ensuring a feasible path under any starting pose, a modified sliding mode controller (SMC) for smooth path following and an overall control scheme to ensure final parking accuracy. Experiment results based on 200 on-field tests under different illumination conditions showed that the system can park the vehicle accurately and consistently in all cases with a 4.8cm RMS offset distance from center line and 1.23° RMS orientation deviation. With its easy setup and excellent performance, it can be implemented onto existing vehicles practically and robustly with minimal additional cost.