다중 물체 추적을 위한 휴버 방식 기반 무향 칼만필터를 활용한 분산형 센서 융합

Abstract

Perception is one of the key factors for the performance and security that needs to be guaranteed for automotives. However, due to the environmental matters, challenging problems still exist and research to improve the performance of perception is actively proceeding. To achieve higher performance, we need various kinds of sensors such as camera, radar, lidar. It is necessary to ensure real-time and high performance so that the results of tracking multiple objects can serve the basis for the latter decisions and planning. Model based state estimator is mainly used which is directly related to the performance of perception when performing multiple object tracking. Kalman filter is widely used among various kinds of state estimators. But Kalman filter has limitations in that they do not consider outlier. When it comes to using outlier data for state estimation, it adversely affects the performance. From this paper, we propose distributed sensor fusion of Unscented Kalman filter based on Huber methodology for the robust performance. Through this, it is possible to reduce computational cost and improve object state estimation performance, which was verified with the Nuscenes dataset.