Learning a distance function with a Siamese network to localize anomalies in videos

This work introduces a new approach to localize anomalies in surveillance\nvideo. The main novelty is the idea of using a Siamese convolutional neural\nnetwork (CNN) to learn a distance function between a pair of video patches\n(spatio-temporal regions of video). The learned distance function, which is not\nspecific to the target video, is used to measure the distance between each\nvideo patch in the testing video and the video patches found in normal training\nvideo. If a testing video patch is not similar to any normal video patch then\nit must be anomalous. We compare our approach to previously published\nalgorithms using 4 evaluation measures and 3 challenging target benchmark\ndatasets. Experiments show that our approach either surpasses or performs\ncomparably to current state-of-the-art methods.\n