Cooperative Road Geometry Estimation via Sharing Processed Camera Data

Traffic in the near future is expected to be a mix of legacy vehicles with limited number of on-board sensors and sensor-rich vehicles with advanced sensing capabilities and different levels of automation. In this work, we propose a novel framework to leverage the existence of sensor-rich vehicles to assist legacy vehicles in estimating the road geometry which is an essential task for advanced driver assistance systems (ADAS). In the proposed method, the legacy vehicle, which is not necessarily equipped with any cameras or ranging sensors, receives processed camera data related to the road geometry from nearby sensor-rich vehicles. Then, the legacy vehicle fuses this data to build a local map of the road ahead for up to 200 m. Using experimental data, we show that the proposed method reduces the root mean square estimation error by 209% and the mean absolute estimation error by 857% compared to camera-based systems. The results also show that sensor-rich vehicles benefit from sharing the processed camera data and can significantly improve the accuracy of the road geometry estimate at much higher distances.