Batch Size Reconstruction-Distribution Trade-off in Kernel Based Generative Autoencoders

The large amounts of movement data collected from various sources such as wide area motion imagery (WAMI) and mobile phone apps call for innovative technologies to infer valuable information from these data. in this paper, we present two such tools to extract the pattern of life (PoL) information from vehicle trajectories. The first tool, intersection traffic analysis (ITA) detects abnormal traffic patterns in major street intersections; while the second one, frequent trajectory patterns analysis (FTPA) discerns the most frequent trajectory patterns in a given time-interval in the region of concern. Both tools support comprehensive trajectory-based pattern of life situation awareness. Using both simulated trajectories and the measurements extracted from WAMI imagery demonstrate ITA and FTPA utility