A new rotation invariant similarity measure for trajectories

We present a new rotation invariant measure for trajectories of dynamically changing locations of mobile objects (vehicles), which appear naturally in applications such as video-tracking, motion capture etc. Similar motion patterns can also be expressed at different orientations. We have modeled each trajectory by its sequence of angles. The similarity measure is defined based on longest common subsequence (LCS) method. To evaluate a system, we have simulated the database consisting of common trajectories of moving vehicles in the cities. First, clustering based on agglomerative algorithm with new similarity measure is applied on the training dataset. To classify new samples, similarity to the median of the clusters is considered and based on the rates of the similarity to the median, some natural language sentences is produced, these sentences express the behavioural descriptions of the vehicles. Experimental results show the accuracy and efficiency of the technique.