A New Dissimilarity Measure for Trajectories with Applications in Anomaly Detection Dustin L. Espinosa-Isidr´on and Dr. Edel B. Garc´ ıa-Reyes Advanced Technologies Application Center (CENATAV), 7a # 21812 e/ 218 y 222, Rpto. Siboney, Playa, C.P. 12200, La Habana, Cuba. {despinosa,edel}@cenatav.co.cu Abstract. Trajectory clustering has been used to very effectively in the detection of anomalous behavior in video sequences. A key point in trajectory clustering is how to measure the (dis)similarity between two trajectories. This paper deals with a new dissimilarity measure for trajec- tory clustering, giving the same importance to differences and similarities between the trajectories. Experimental results in the task of anomalous detection via hierarchical clustering shows the validity of the proposed approach. Key words: Trajectory Clustering, Dissimilarity, Anomaly Detection 1 Introduction Video surveillance is a research field that has received much interest over the last years. Parking lot surveillance, traffic monitoring, and crime prevention are among the applications of video surveillance systems. A key task that could help improve the effectiveness of these systems is the automatic detection of anomalous behaviors. Trajectory clustering has been established as an effective tool to address the task. A fundamental issue in trajectory clustering is how to measure the (dis)similarity between the trajectories. In this work we propose a new dissimilarity measure for trajectories, namely Dissimilarity for Trajectories(DT). The core of DT is a non-symmetric dissimi- larity which yields the same importance to differences and similarities between the trajectories. All measures are tested in the task of anomaly detection via trajectory clustering. The selected data sets contain different amounts of normal trajectories with different amounts of abnormal trajectories. The remainder of this paper is organized as follows. Sec. 2 describes the repre- sentation and dissimilarity for trajectories, including the proposed dissimilarity measure. Anomaly detection via trajectory clustering is presented alongside ex- perimental results in Sec. 3. Finally, Sec. 4. concludes the paper. 2 (Dis)Similarity Measures for Trajectories Trajectory Representation: Usually a trajectory is represented as a sequence S =(s 1 ,s 2 ,...,s h ), where each s i ,1 ≤ i ≤ h, is a point in a multidimen- sional space containing information about the moving object at time i. Most