Evaluation Technique of 3D Point Clouds for Autonomous Vehicles Using the Convergence of Matching Between the Points

In this paper, we propose a new map evaluation technique for autonomous vehicles using a 3D point cloud. Localization in autonomous driving is an important technology. Attention is focused on accurate 3D mapping and point cloud data, because this map data is needed to estimate vehicle position. However, the constructed 3D point group may have errors due to the measurement. Localization has also been known to fail in places where the terrain has few distinct features. Our technique focuses on localization process to evaluate the map. The goal of our proposal is to calculate the probability of success or failure of localization. This evaluation method uses convergence by matching. Evaluation tests showed that the places where the localization is possible, and the place where the error remains on the map can be clearly separated. In future, the range of the input 3D point cloud is made into the range applicable to Localization, and we evaluate the validity of the proposed method by increasing the set.