Pose Estimation-Based V2I Communication for Autonomous Vehicle Parking

Recently, autonomous driving and parking systems have mainly used in-vehicle sensors to recognize the surrounding environment. However, there is a limitation in that the vehicle manufacturing cost increases and the burden of calibration and maintenance of various sensors. This paper proposes a method that induces precise parking by estimating the position of the vehicle only with the fixed external cameras without using any in-vehicle sensors. The proposed method was implemented based on ROS2, and Fast DDS was utilized for data transfer. By detecting the 'top' and 'bottom' of a client vehicle from an external camera, a vector representing the vehicle's orientation is formed. Based on the angle difference between this vector and a predefined waypoint, real-time steering data is calculated. This lowers the individual cost of the vehicle and allows multiple vehicles to share one infrastructure to build a highly scalable system that can perform parking. As a result of the experiment, it was confirmed that parking was completed with a centimeter-level position error and a direction error within several degrees with only fixed cameras. This result indicates that it is possible to achieve sufficient precision even with a low-cost, low-complexity system.