Development of autonomous driving sensing system capable of real-time weather condition recognition and object detection/tracking in snowy weather

Abstract

As research on autonomous driving technology is actively conducted around the world, research using autonomous driving sensors such as LiDAR, radar, and vision camera is also being actively conducted. Accordingly, research on object detection and tracking using these autonomous driving sensors is also being actively conducted. However, due to the advantages and disadvantages of each sensor and the vulnerability to environmental changes such as snow, rain, fog, and night situations, research is currently focused on autonomous driving in daytime and good weather conditions. In recent years, research on autonomous driving in snowy environments has been gradually progressing mainly in Northern Europe and North America, which are regions where there is a lot of snow. In particular, as the military, which is essential for combat in extreme conditions day and night, is also conducting unmanned weapon systems, research on autonomous driving in the defense field is actively progressing, and interest in developing autonomous driving technology in extreme environment conditions is growing. Accordingly, in this paper, we propose an autonomous driving sensing system that can robustly detect and track objects even in extreme environmental conditions such as snow, rain, fog, and night. First, the radar sensor is robust to environmental changes, but the resolution is relatively low when acquiring position information or object shape compared to other sensors. Therefore, we improve location accuracy through Kalman filter fusion with LiDAR sensor, and complement object classification and visualization abilities through sensor fusion with mono camera. At this time, to overcome the shortcomings of LiDAR sensor vulnerable to environmental changes such as snow, rain, fog, and dust, a filter was developed that can remove noise in real time, and then an object detection system using LiDAR point cloud applied with this filter was developed and fused with LiDAR sensor through Kalman filter. In addition, a real-time weather condition recognition system using snow particles removed by noise removal filters was developed to provide additional weather information necessary for controlling autonomous vehicle. Through the development of an all-weather detection and tracking system, it is expected that not only autonomous vehicle technology, but also unmanned combat vehicles, unmanned robots, GOP border security systems, attack drones, and UAVs in the defense field will be available.