A Generalized Bayesian Approach for Localizing Static Natural Obstacles on Unpaved Roads
This paper presents an approach that implements sensor fusion and recursive Bayesian estimation (RBE) to improve a vehicle's ability to perform obstacle detection and localization in unpaved road environments. The proposed approach utilizes RADAR, LiDAR and stereovision fully for sensor fusion to detect and localize static natural obstacles. Each sensor is characterized by a probabilistic sensor model which quantifies level of confidence (LOC) and probability of detection (POD) associatively. Deploying these sensor models enables the fusion of heterogeneous sensors without extensive formulations and with the incorporation of each sensor's strengths. An Extended Kalman filter (EKF) is formulated and implemented for robust and computationally efficient RBE of obstacles' locations while a sensor-equipped vehicle moves and observes them. Results with a test vehicle show the successful detection and localization of a static natural object on an unpaved road has demonstrated the effectiveness of the proposed approach.
Robot sensing systems, Roads, Radar, Laser radar, Cameras, Bayes methods, Mathematical model
Y. Kinoshita, J. Steckenrider, I. Papakis and T. Furukawa, "A Generalized Bayesian Approach for Localizing Static Natural Obstacles on Unpaved Roads," 2020 IEEE International Symposium on Safety, Security, and Rescue Robotics (SSRR), Abu Dhabi, United Arab Emirates, 2020, pp. 283-289, doi: 10.1109/SSRR50563.2020.9292600.