Cooperative pedestrian tracking by multi-vehicles in GPS-denied environments

摘要

This paper presents laser-based pedestrian tracking using multiple vehicles in outdoor environments. Each vehicle locates pedestrians using its own laser scan image by an occupancy-grid-based method, and it tracks the detected pedestrians via Kalman filtering and Global nearest neighbor based data association. In addition, it generates a local map using extended kalman filter-simultaneous localization and mapping (EKF-SLAM), where cylindrical objects in outdoor environments, such as trees, poles, posts, and the corners of the buildings are used as landmarks. When the vehicles are in close proximity, they exchange their local maps and pedestrians tracking information through intercommunication. Tracking data are combined using Covariance Intersection, and a global map is built by merging the local maps, enabling improved tracking performance. In our tracking method, all vehicles share tracking data with each other, so that they can recognize pedestrians that are invisible to any vehicles. The experimental results for tracking two pedestrians with two vehicles validate the proposed method.