Context-Dependent Anomaly Detection for Low Altitude Traffic Surveillance

The detection of contextual anomalies is a challenging task for surveillance\nsince an observation can be considered anomalous or normal in a specific\nenvironmental context. An unmanned aerial vehicle (UAV) can utilize its aerial\nmonitoring capability and employ multiple sensors to gather contextual\ninformation about the environment and perform contextual anomaly detection. In\nthis work, we introduce a deep neural network-based method (CADNet) to find\npoint anomalies (i.e., single instance anomalous data) and contextual anomalies\n(i.e., context-specific abnormality) in an environment using a UAV. The method\nis based on a variational autoencoder (VAE) with a context sub-network. The\ncontext sub-network extracts contextual information regarding the environment\nusing GPS and time data, then feeds it to the VAE to predict anomalies\nconditioned on the context. To the best of our knowledge, our method is the\nfirst contextual anomaly detection method for UAV-assisted aerial surveillance.\nWe evaluate our method on the AU-AIR dataset in a traffic surveillance\nscenario. Quantitative comparisons against several baselines demonstrate the\nsuperiority of our approach in the anomaly detection tasks. The codes and data\nwill be available at https://bozcani.github.io/cadnet.\n