Dynamic bayesian networks: representation, inference and learning
Dynamic bayesian networks: representation, inference and learning
Learning and inferring transportation routines
AAAI'04 Proceedings of the 19th national conference on Artifical intelligence
Partial pathfinding using map abstraction and refinement
AAAI'05 Proceedings of the 20th national conference on Artificial intelligence - Volume 3
Fast and Accurate Prediction of the Destination of Moving Objects
SUM '09 Proceedings of the 3rd International Conference on Scalable Uncertainty Management
An AGM-style belief revision mechanism for probabilistic spatio-temporal logics
Artificial Intelligence
| Hi-index | 0.00 |
Tracking the movements of a target based on limited observations plays a role in many interesting applications. Existing probabilistic tracking techniques have shown considerable success but the majority assume simplistic motion models suitable for short-term, local motion prediction. Agent movements are often governed by more sophisticated mechanisms such as a goal-directed path-planning algorithm. In such contexts we must go beyond estimating a target's current location to consider its future path and ultimate goal. We show how to use complex, "black box" motion models to infer distributions over a target's current position, origin, and destination, using only limited observations of the full path. Our approach accommodates motion models defined over a graph, including complex pathing algorithms such as A*. Robust and practical inference is achieved by using hidden semi-Markov models (HSMMs) and graph abstraction. The method has also been extended to effectively track multiple, indistinguishable agents via a greedy heuristic.