Automated object’s activity analysis has been and still remains a challenging problem and motion trajectories provide rich spatiotemporal information for this purpose. This paper presents a novel descriptor to analyze object activity based on object trajectories. In the proposed descriptor extraction technique, object’s change in direction is extracted in different level of resolution. One of the most important characteristics of the proposed approach is that the descriptor is translation and rotation invariant. We first segment the trajectories based on the absence of changes in direction via spectral clustering. Long Common Sub-Sequence (LCSS) distance is used to compare the extracted proposed descriptor for unequal length subtrajectories. Experiments using the trajectories of objects datasets (LABOMNI, CROSS and laser monitoring) demonstrate the superiority of using the proposed multiresolution descriptor as a similarity factor in comparison with the similar techniques in the literature.
