%0 Conference Proceedings %T Abnormal trajectory detection for security infrastructure %+ Laboratoire Modélisation et Sûreté des Systèmes (LM2S) %A Le, Van-Khoa %A Beauseroy, Pierre %A Grall-Maës, Edith %< avec comité de lecture %B the 2nd International Conference %C Tokyo, Japan %I ACM Press %3 ICDSP 2018: Proceedings of the 2nd International Conference on Digital Signal Processing %P 1-5 %8 2018-02-25 %D 2018 %R 10.1145/3193025.3193026 %Z Engineering Sciences [physics]/Signal and Image processingConference papers %X In this work, an approach for the automatic analysis of people trajectories is presented, using a multi-camera and card reader system. Data is first extracted from surveillance cameras and card readers to create trajectories which are sequences of paths and activities. A distance model is proposed to compare sequences and calculate similarities. The popular unsupervised model One-Class Support Vector Machine (One-Class SVM) is used to train a detector. The proposed method classifies trajectories as normal or abnormal and can be used in two modes: off-line and real-time. Experiments are based on data simulation corresponding to an attack scenario proposed by a security expert. Results show that the proposed method successfully detects the abnormal sequences in the scenario with very low false alarm rate. %G English %L hal-02518529 %U https://utt.hal.science/hal-02518529 %~ CNRS %~ UNIV-TROYES %~ UTT %~ UTT-LIST3N %~ LM2S-UTT