Left: features extracted from (a) one frame of a moving-camera video. (b) Haar-feature (summed intensity in adjacent rectangles \cite{Viola}) response map for horizontal lines. (c) Eight Local Binary Patterns histograms (simple texture features \cite{Ojala_2002}) extracted from eight sub-patches. Middle: (d) consecutive saccades (inspired by mammalian visual attention mechanisms \cite{Andreopoulos_2013}), targeted to peaks of a saliency map (e). (f) K-means classifier predictions for a person (top, fourth bar) or bicyclist (bottom, fifth bar); bar height denotes prediction strength for five foreground classes and a null class (sixth bar). Right: (g) grid classifier detects and classifies five types of foreground objects in each frame of an HD video with a vertical aspect ratio (green box = person, magenta = bicyclist, and yellow = car). Adapted from \cite{Akopyan_2015}.