In the last decades, advanced sensors play increasingly important role in modern vehicles. The first step towards fully autonomous driving were Advanced Driver Assistance Systems (ADAS), where smart sensors played a crucial role. For ADAS applications, RGB cameras, radars and ultrasonic sensors were used. In order to achieve higher levels of automated driving, more advanced sensors are required in order to provide better accuracy at higher distances, necessary for highway applications. Moreover, it is necessary to achieve improved robustness in challenging atmospheric and illumination conditions. In order to achieve these goals, different types of sensor were proposed, like radar, thermal cameras and lidar. However, to achieve the required angular resolution for highway operation at distances up to 250m, FMCW lidar proved to be the most promising solution. In this paper we will present methods for advanced processing of FMCW lidar signals, noise reduction and temporal filtering. The first component of the proposed work is enhanced demodulation processing, which improves the accuracy of lidar, in depth resolution, based on neural networks. This processing step includes improved beat frequency detection and noise reduction. To further improve the robustness of the result, we perform temporal filtering of the point clouds. This results in enhanced temporal stability of the point clouds and reduced uncertainty of the object position. We compare the results of the proposed method with the recent state of the art in terms of objective quality measures and prove its effectiveness.
|