Statistical models of object interaction

We present a method for assessing the likelihood of a trajectory of an object through a scene consisting of a number of other objects. The closest points on the trajectory to the other objects are chosen as landmark points and at each landmark we calculate the probability of the interaction based on the speed and distance. Sequences of such probabilities are then sorted in increasing order. Finally a weighted sum of the first few elements in this weighted list is used to classify trajectories in a supervised learning framework.