Learning Stochastic Path Planning Models from Video Images
Sridevi Parise, Padhraic Smyth · 2004
We describe a probabilistic framework for learning models of pedestrian trajectories in general outdoor scenes. Possible applications include simulation of motion in computer graphics, video surveillance, and architectural design and analysis. The models are based on a combination of Kalman filters and stochastic path-planning via landmarks, where the landmarks are learned from the data. A dynamic Bayesian network (DBN) framework is used to represent the model as a position-dependent switching state space model. We illustrate how such models can be learned and used for prediction using the block Gibbs sampler with forward-backward recursions. The ideas In this paper we describe a probabilistic landmark-based framework for modeling, learning, and predicting pedestrian trajectories in a scene (as obtained from video-based tracking, for example). Figure 1 shows an example of a set of trajectories. Trajectory modeling and prediction has applications in problems such as computer graphics, video surveillance, path usage analysis for planning